CUDA Fortran

Programming Guide and Reference

Release 2011

The Portland Group

While every precaution has been taken in the preparation of this document, The Portland Group® (PGI®), a wholly-owned subsidiary of STMicroelectronics, Inc., makes no warranty for the use of its products and assumes no responsibility for any errors that may appear, or for damages resulting from the use of the information contained herein. The Portland Group retains the right to make changes to this information at any time, without notice. The software described in this document is distributed under license from STMicroelectronics and/or The Portland Group and may be used or copied only in accordance with the terms of the license agreement ("EULA"). PGI Workstation, PGI Server, PGI Accelerator, PGF95, PGF90, PGFORTRAN, and PGI Unified Binary are trademarks; and PGI, PGHPF, PGF77, PGCC, PGC++, PGI Visual Fortran, PVF, PGI CDK, Cluster Development Kit, PGPROF, PGDBG, and The Portland Group are registered trademarks of The Portland Group Incorporated. No part of this document may be reproduced or transmitted in any form or by any means, for any purpose other than the purchaser's or the end user's personal use without the express written permission of STMicroelectronics and/or The Portland Group.

Copyright © 2010 The Portland Group® and STMicroelectronics, Inc.All rights reserved. Printed in the United States of America Release 2010, version 10.8, August 2010 Release 2010, version 10.9, September 2010 Release 2011, version 11.0, November 2010

Technical support: trs@pgroup.com Sales: sales@pgroup.com Web: www.pgroup.com

ID: 103151117

Contents
Preface ...................................................................................................................................... ix
Intended Audience ................................................................................................................. ix Organization .......................................................................................................................... ix Conventions ........................................................................................................................... ix Terminology ........................................................................................................................... x Related Publications ................................................................................................................ x

1. Introduction ........................................................................................................................ 1 2. Programming Guide .......................................................................................................... 3
CUDA Fortran Kernels .............................................................................................................. 3 Thread Blocks ........................................................................................................................ 4 Memory Hierarchy .................................................................................................................. 4 Subroutine / Function Qualifiers ................................................................................................ 4 Attributes(host) .............................................................................................................. 5 Attributes(global) ............................................................................................................ 5 Attributes(device) ........................................................................................................... 5 Restrictions .................................................................................................................... 5 Variable Qualifiers .................................................................................................................. 5 Attributes(device) ........................................................................................................... 5 Attributes(constant) ........................................................................................................ 6 Attributes(shared) ........................................................................................................... 6 Attributes(pinned) .......................................................................................................... 6 Datatypes in Device Subprograms ............................................................................................. 6 Predefined Variables in Device Subprograms .............................................................................. 7 Execution Configuration ........................................................................................................... 7 Asynchronous concurrent execution .......................................................................................... 8 Concurrent Host and Device Execution .............................................................................. 8 Concurrent Stream Execution ........................................................................................... 8 Kernel Loop Directive .............................................................................................................. 8 Restrictions on the CUF kernel directive ........................................................................... 10 Building a CUDA Fortran Program ........................................................................................... 11 Emulation Mode ................................................................................................................... 11 iii

3. Reference ........................................................................................................................... 13
New Subroutine and Function Attributes ................................................................................... Host Subroutines and Functions ...................................................................................... Global Subroutines ........................................................................................................ Device Subroutines and Functions ................................................................................... Restrictions on Device Subprograms ................................................................................ Variable Attributes ................................................................................................................. Device data .................................................................................................................. Constant data ................................................................................................................ Shared data .................................................................................................................. Value dummy arguments ................................................................................................ Pinned arrays ............................................................................................................... Allocating Device and Pinned Arrays ........................................................................................ Allocating Device Memory .............................................................................................. Allocating Device Memory Using Runtime Routines ............................................................ Allocating Pinned Memory .............................................................................................. Data transfer between host and device memory ......................................................................... Data Transfer Using Assignment Statements ....................................................................... Implicit Data Transfer in Expressions ............................................................................... Data Transfer Using Runtime Routines ............................................................................. Invoking a kernel subroutine .................................................................................................. Device code ......................................................................................................................... Datatypes allowed ......................................................................................................... Builtin variables ............................................................................................................ Fortran intrinsics .......................................................................................................... New Intrinsic Functions ................................................................................................. Warp-Vote Operations .................................................................................................... Atomic Functions .......................................................................................................... Restrictions .................................................................................................................. Host code ............................................................................................................................ SIZEOF Intrinsic ............................................................................................................ 13 13 13 14 14 14 14 15 16 17 17 17 17 18 18 19 19 19 20 20 21 21 22 22 24 26 27 29 29 29

4. Runtime APIs .................................................................................................................... 31
Initialization ......................................................................................................................... Device Management ............................................................................................................... cudaGetDeviceCount ...................................................................................................... cudaSetDevice ............................................................................................................... cudaSetDeviceFlags ........................................................................................................ cudaGetDevice .............................................................................................................. cudaGetDeviceProperties ................................................................................................ cudaChooseDevice ......................................................................................................... Thread Management .............................................................................................................. cudaThreadSynchronize ................................................................................................. cudaThreadExit ............................................................................................................. Memory Management ............................................................................................................ cudaMalloc .................................................................................................................. iv 31 31 31 31 31 32 32 32 32 32 32 32 33

cudaMallocPitch ............................................................................................................ cudaFree ...................................................................................................................... cudaMallocArray ........................................................................................................... cudaFreeArray .............................................................................................................. cudaMemset ................................................................................................................. cudaMemset2D ............................................................................................................. cudaMemcpy ................................................................................................................ cudaMemcpyAsync ........................................................................................................ cudaMemcpy2D ............................................................................................................ cudaMemcpy2DAsync .................................................................................................... cudaMemcpyToArray ...................................................................................................... cudaMemcpy2DToArray .................................................................................................. cudaMemcpyFromArray .................................................................................................. cudaMemcpy2DFromArray .............................................................................................. cudaMemcpyArrayToArray .............................................................................................. cudaMemcpy2DArrayToArray .......................................................................................... cudaMalloc3D .............................................................................................................. cudaMalloc3DArray ....................................................................................................... cudaMemset3D ............................................................................................................. cudaMemcpy3D ............................................................................................................ cudaMemcpy3DAsync .................................................................................................... cudaMemcpyToSymbol ................................................................................................... cudaMemcpyFromSymbol ............................................................................................... cudaMemcpyToSymbolAsync ........................................................................................... cudaMemcpyFromSymbolAsync ....................................................................................... cudaGetSymbolAddress ................................................................................................... cudaGetSymbolSize ........................................................................................................ cudaHostAlloc ............................................................................................................... cudaHostGetDevicePointer .............................................................................................. cudaMallocHost ............................................................................................................ cudaFreeHost ................................................................................................................ Stream Management .............................................................................................................. cudaStreamCreate .......................................................................................................... cudaStreamQuery .......................................................................................................... cudaStreamSynchronize .................................................................................................. cudaStreamDestroy ........................................................................................................ Event Management ................................................................................................................ cudaEventCreate ............................................................................................................ cudaEventRecord .......................................................................................................... cudaEventQuery ............................................................................................................ cudaEventSynchronize .................................................................................................... cudaEventDestroy .......................................................................................................... cudaEventElapsedTime ................................................................................................... Error Handling ..................................................................................................................... cudaGetLastError .......................................................................................................... cudaGetErrorString ........................................................................................................

33 33 33 33 34 34 34 34 34 35 35 35 35 35 36 36 36 36 36 36 37 37 37 37 37 38 38 38 38 39 39 39 39 39 39 39 40 40 40 40 40 40 40 41 41 41 v

....... Contact Information ........ 41 5............................................... Examples ................................................................................ 41 cudaRuntimeGetVersion ............................................ 43 Source Code Listing ............................................................................................................................................................................................................................................................................................................ 45 6............................. 41 cudaDriverGetVersion .............................Version Management ............ 47 vi ............................................................................................................ 43 Source Code Description .................................. 43 Matrix Multiplication Example ......................................................................................................................................

.....................................................................................................4........................Tables 2......................10........................................ Fortran Numeric and Logical Intrinsics ................................5... 23 3.............. 27 3....................................... 23 3.......1......... Fortran Numeric Inquiry Intrinsics ...................................... Fortran Bit Manipulation Intrinsics .................. Fortran Mathematical Intrinsics ...................................................6........ 28 3......... Arithmetic and Bitwise Atomic Functions ...................... Intrinsic Datatypes ................ 6 3... 24 3............................................................... 21 3................................................................................ Device Code Intrinsic Datatypes ...... Counting Atomic Functions ....9.............8......................... 23 3................ Compare and Swap Atomic Function ............................................................................................2.................... 29 vii ............................................................3.................... Fortran Random Number Intrinsics .......................................... 24 3............. 22 3.............................7...... Fortran Reduction Intrinsics ...................1........................................................................................................

viii .

This guide assumes familiarity with basic operating system usage. Organization The organization of this document is as follows: Chapter 1. C. Intended Audience This guide is intended for application programmers. AMD64. “Introduction” contains a general introduction Chapter 2. Conventions This guide uses the following conventions: italic is used for emphasis. a small set of extensions to Fortran that supports and is built upon the CUDA computing architecture. “Reference” describes the CUDA Fortran language reference Chapter 4. “Runtime APIs” describes the interface between CUDA Fortran and the CUDA Runtime API Chapter 5. ix . “Programming Guide” serves as a programming guide for CUDA Fortran Chapter 3. and/or C++ languages. and Intel 64 hardware platforms. scientists and engineers proficient in programming with the Fortran.Preface This document describes CUDA Fortran. “Examples” provides sample code and an explanation of the simple example. The PGI tools are available on a variety of operating systems for the X86.

nvidia. C/C++ C/C++ language statements are shown in the test of this guide using a reduced fixed point size. Terminology If there are terms in this guide with which you are unfamiliar.nvidia.htm Related Publications The following documents contain additional information related to CUDA Fortran programming. In this case item1 is optional.Terminology Constant Width is used for filenames. and for language statements in the text. Information Technology – Programming Languages – FORTRAN.1. NVIDIA. In this case. multiple filenames are allowed. Available online at http://www. PGI provides a glossary of terms which you can access at www.com/doc/pgiug. options. [ item1 ] in general. square brackets are required to specify a p/t-set. 7/21/2010. Available online at http:// www.1.com/cuda. Geneva. you must select either item2 or item3. square brackets indicate optional items. NVIDIA. { item2 | item 3} braces indicate that a selection is required.. x . • ISO/IEC 1539-1:1997. Available online at http:// www.pgroup. There are a wide variety of releases and distributions of each of these types of operating systems. • NVIDIA CUDA Compute Unified Device Architecture Reference Manual. The Portland Group. filename .com/support/definitions. directories. including assembly language statements. Release 10.pdf. 1997 (Fortran 95).1. • PGI Users Guide. Version 3. September.. Bold is used for commands. Version 3. ellipsis indicate a repetition. • NVIDIA CUDA Programming Guide.9. The PGI compilers and tools are supported on both 32-bit and 64-bit variants of the Linux and Windows operating systems on a variety of x86-compatible processors. arguments.pgroup. In the context of p/tsets. 2009. June 2010. examples. In this example.com/cuda. FORTRAN Fortran language statements are shown in the text of this guide using a reduced fixed point size. Zero or more of the preceding item may occur.

CUDA Fortran includes a Fortran 2003 compiler and tool chain for programming NVIDIA GPUs using Fortran. The CUDA Fortran extensions described in this document allow the following operations in a Fortran program: • Declaring variables that are allocated in the GPU device memory • Allocating dynamic memory in the GPU device memory • Copying data from the host memory to the GPU memory. The programming model supports four key abstractions: cooperating threads organized into thread groups.Chapter 1. GPU designs are optimized for the computations found in graphics rendering. a small set of extensions to Fortran that supports and is built upon the CUDA computing architecture. which can cooperate using shared memory and barrier synchronization. CUDA Fortran is an analog to NVIDIA's CUDA C compiler. here called CUDA C. a general purpose parallel programming architecture. Introduction Welcome to Release 2011 of PGI CUDA Fortran. CUDA Fortran is a lower-level explicit programming model with substantial runtime library components that give expert programmers direct control of all aspects of GPGPU programming. allowing direct programming of the GPU from a high level language. Compared to the PGI Accelerator directives-based model and compilers. Each block is partitioned into fine grain threads. and tremendous potential for many applications. A properly designed CUDA program will run on any CUDA-enabled GPU. and back • Writing subroutines and functions to execute on the GPU • Invoking GPU subroutines from the host • Allocating pinned memory on the host • Using asynchronous transfers between the host and GPU 1 . highly parallel computational units with very high memory bandwidth. with compilers and libraries to support the programming of NVIDIA GPUs. but are general enough to be useful in many data-parallel. regardless of the number of available processor cores. A CUDA programmer must partition the program into coarse grain blocks that can be executed in parallel. shared memory and barrier synchronization within thread groups. NVIDIA introduced CUDA™. PGI 2011 includes support for CUDA Fortran on Linux. compute-intensive programs. Graphic processing units or GPUs have evolved into programmable. Mac OS X and Windows. CUDA comes with an extended C compiler. and coordinated independent thread groups organized into a grid.

2 .

a.y real. dimension(*) :: x. value :: a integer. each with 64 threads. refer to Chapter 3. For a reference for CUDA Fortran.” on page 13. dimension(*) :: x. “Reference. a kernel is called using special chevron syntax to specify the number of thread blocks and threads within each thread block: ! Kernel definition attributes(global) subroutine ksaxpy( n. CUDA Fortran Kernels CUDA Fortran allows the definition of Fortran subroutines that execute in parallel on the GPU when called from the Fortran program which has been invoked and is running on the host. a. value :: n. and a local thread index within its thread block. A kernel is defined using the attributes(global) specifier on the subroutine statement. x. y ) real. Such a subroutine is called a device kernel or kernel. Programming Guide This chapter introduces the CUDA programming model through examples written in CUDA Fortran.Chapter 2. The CUDA threads are organized into thread blocks. and a thread index accessed through threadidx. a. each instance will be executed by a different CUDA thread. each thread performs one iteration of the common SAXPY loop operation. x. x. 3 . the call to the kernel ksaxpy specifies n/64 thread blocks. y ) end subroutine In this case. y real :: a integer :: n ! call the kernel call ksaxpy<<<n/64. In this example. i i = (blockidx%x-1) * blockdim%x + threadidx%x if( i <= n ) y(i) = a * x(i) + y(i) end subroutine ! Host subroutine subroutine solve( n. y ) real. and each thread has a global thread block index. 64>>>( n. device. Each thread is assigned a thread block index accessed through the built-in blockidx variable. A call to a kernel specifies how many parallel instances of the kernel must be executed.

high bandwidth software managed cache memory. and a thread index accessed through threadidx. computed from the thread index.Dy). When invoking a kernel. again implemented using DMA access. data in shared memory has a lifetime of the thread block. The thread index may be a one-. Each thread can also have private local memory. Thread blocks must be able to execute independently. Threads in the same thread block can access and share data in shared memory. 4 . or with the host attribute by default. called from the host. The host can also set the values in the device constant memory. The thread blocks are organized into a one. or three-dimensional index. two-. the thread ID is (x+Dx(y-1)+Dy(z-1)). all threads can read data in constant memory. A unique thread ID is assigned to each thread. For a one-dimensional thread block. the thread index for each dimension starts at one. each with the same thread block size. Threads in the same thread block may cooperate by using shared memory.or two-dimensional grid of blocks. For a three-dimensional thread block of size (Dx.Dz). On the host side. the first argument in the chevron <<<>>> syntax is the grid size. but it is read-only to the threads and limited in size. A kernel may be invoked with many thread blocks. Each thread in the block waits at the call to SYNCTHREADS() until all threads have reached that call.Thread Blocks Thread Blocks Each thread is assigned a thread block index accessed through the built-in blockidx variable. In CUDA Fortran. Memory Hierarchy CUDA Fortran programs have access to several memory spaces. and if the latter. the maximum number of threads in a thread block is 512. such data copies require DMA access to the device. For a two-dimensional thread block of size (Dx. Data in constant memory space is initialized by the host program. data in global device memory can be read or written by all threads. It can also directly copy data to and from the device global memory. the thread ID is equal to (x+Dx(y-1)). On the device side. and by synchronizing at a barrier using the SYNCTHREADS() intrinsic. data in thread local memory may be implemented as processor registers or may be allocated in the global device memory. two thread blocks may be executed in parallel or one after the other. Currently. so each thread has a thread index within the block. designating whether it is executed on the host or on the device. the thread index is equal to the thread ID. is called a host subprogram. • A subprogram declared with attributes(host). so are slow relative to the host memory. best performance will often be obtained when thread local data Subroutine / Function Qualifiers A subroutine or function in CUDA Fortran has an additional attribute. The shared memory acts like a low-latency. by the same core or by different cores. or called from another device subprogram. and a block index within the grid. whether it is a kernel. the host program can directly access data in the host main memory.Dy. Accesses to constant memory are typically faster than accesses to global memory. and the second argument is the thread block size. • A subprogram declared with attributes(global) or attributes(device) is called a device subprogram.

global. specified on the subroutine or function statement. specified on the subroutine or function statement. to be executed on the device. that is. • Calls to a kernel subroutine must specify the execution configuration. and may only be called from the host using a kernel call containing the chevron syntax and runtime mapping parameters. Attributes(host) The host attribute. Attributes(device) A variable with the device attribute is called a device variable. 5 . Variable Qualifiers Variables in CUDA Fortran have a new attribute. Restrictions The following restrictions apply to subprograms: • A device subprogram must not be recursive. the variable may be accessed by any subprogram in that module and by any subprogram that uses the module. as described in “Predefined Variables in Device Subprograms. it declares that the subroutine is a kernel subroutine. which declares in which memory the data is allocated. if none of the host. the host routine making the call continues to execute before the device has completed its execution of the kernel subroutine.” on page 7. Such a subprogram can only be called from another host subprogram. Dummy arguments in a device subprogram must not have the pointer attribute. or device attributes is specified. • A kernel subroutine may not also have the device or host attribute. variables declared in modules or host subprograms are allocated in the host main memory. • Device subprograms may not be contained in a host subroutine or function. and may not contain any subroutines or functions. constant. The default is attributes(host). or pinned attributes may be specified for a variable. and is allocated in the device global memory. • A device subprogram must not have optional arguments. Attributes(device) The device attribute. • A device subprogram must not contain variables with the SAVE attribute. Programming Guide • A subroutine declared with attributes(global) is also called a kernel subroutine. such a routine must be called from a subprogram with the global or device attribute.Chapter 2. Attributes(global) The global attribute may only be specified on a subroutine statement. At most one of the device. By default. Such a call is asynchronous. or with data initialization. declares that the subroutine or function is to be executed on the host. • If declared in a module. declares that the subprogram is to be executed on the device. shared.

Intrinsic Datatypes Type integer logical real double precision Type Kind 1. and has a lifetime of the thread block. from when it is allocated until it is deallocated. If declared in a module.2.8 equivalent to real(kind=8) 6 . A shared variable may not be data initialized. it will be allocated in host pagelocked memory. Device constant data may not be assigned or modified in any device subprogram.8 1. if the allocation in page-locked memory fails. and have a lifetime of the entire application. Other device variables have a lifetime of the entire application. the variable will be allocated in the normal host paged memory.4.Datatypes in Device Subprograms • If declared in a host subprogram. Some operating systems or installations may restrict the use. Datatypes in Device Subprograms The following intrinsic datatypes are allowed in device subprograms and device data: Table 2. Device constant variables may not be allocatable.8 4. or an assumed-shape dummy array. It can be read or written by all threads in the block. A pinned variable must be an allocatable array. or by other device subprograms to which it is passed as an argument. A shared variable may only be declared in a device subprogram. A shared variable is allocated in the device shared memory for a thread block. but may be modified in host subprograms.2. the variable may be accessed by any subprogram in that module and by any subprogram that uses the module. An allocatable device variable has a dynamic lifetime. Attributes(constant) A variable with the constant attributes is called a device constant variable. an allocatable array. though a write in one thread is only guaranteed to be visible to other threads after the next call to the SYNCTHREADS() intrinsic. availability. When a pinned variable is allocated. Device constant variables are allocated in the device constant memory space. and may only be accessed within that subprogram.1.4. The advantage of using pinned variables is that copies from page-locked memory to device memory are faster than copies from normal paged host memory. Attributes(pinned) A variable with the pinned attributes is called a pinned variable. or size of page-locked memory. A device array may be an explicit-shape array. Attributes(shared) A variable with the shared attributed is called a device shared variable or a shared variable. the variable may be accessed by that subprogram or subprograms contained in that subprogram.

the threadidx%y and/or threadidx%z components have the value one. as with threadidx.or two-dimensional thread blocks. the value of each component must be equal to or greater than one.8 1 Derived types may contain members with these intrinsic datatypes or other allowed derived types. and a stream identifier. Programming Guide complex character(len=1) 4. or three-dimensional thread blocks. are not accessible in host subprograms. and block%x and block%y must be equal to or greater than one. Execution Configuration A call to a kernel subroutine must specify an execution configuration. blockdim has the same value for all thread blocks in the same grid. The execution configuration defines the dimensionality and extent of the grid and thread blocks that execute the subroutine.Chapter 2. Threads are executed in groups of 32.block[.streamid]]>>>(arg1.y. blockidx%y has the value one. It may also specify a dynamic shared memory extent. These variables are of type dim3. for one. called warps. for one-dimensional grids. or are type(dim3). two-. for one.…) where • grid and block are either integer expressions (for one-dimensional grids and thread blocks). • The variable threadidx contains the thread index within its thread block. in bytes. • If grid is type(dim3). • The variable griddim contains the dimensions of the grid. A kernel subroutine call looks like this: call kernel<<<grid. and the product of the component values must be less than or equal to 512. The value of blockidx%z is always one.z end type These predefined variables. This memory is used for the 7 . in addition to the statically allocated shared memory. the module cudafor defines the derived type dim3 as follows: type(dim3) integer(kind=4) :: x. warpsize contains the number of threads in a warp.arg2.bytes[. • The variable blockidx contains the block index within the grid. to support concurrent stream execution on the device. the value of grid%z must be one. • The variable is warpsize is declared to be type integer.or two-dimensional grids and one-. except for warpsize. • The value of bytes must be an integer. • The variable blockdim contains the dimensions of the thread block. the value of griddim%z is always one. it specifies the number of bytes of shared memory to be allocated for each thread block. Predefined Variables in Device Subprograms Device subprograms have access to block and grid indices and dimensions through several builtin read-only variables. • If block is type(dim3).

Concurrent Stream Execution Operations involving the device. in cases where the host program reads or modifies device or constant data. and no subsequent operations on any queue begin until the stream zero operation is complete.Asynchronous concurrent execution assumed-size shared variables in the thread block. and will only be initiated when all previous operations on that queue have been completed. The default stream. including calling or queueing more kernels for execution on the device. and may execute concurrently with each other. where the launch actually corresponds to placing the kernel on a queue for execution by the device.2. it specifies the stream to which this call is associated. • The value of streamid must be an integer greater than or equal to zero. the launch is asynchronous. see Section 3. An operation is placed at the end of the stream queue. is stream zero. The call can be treated as a kernel launch operation. Launch configuration and mapping of the loop iterations onto the hardware is controlled and specified as part of the directive body using the familiar CUDA chevron syntax. The program can use cudaThreadSynchronize() or CUDA Events to wait for the completion of the kernel. operations on different stream queues may execute out-of-order with respect to when they were placed on the queues. used when no stream identifier is specified. the host program can synchronize and wait for all previously launched or queued kernels. the host can continue executing. the compiler handles the generation of the final reduction kernel. The work in the loops specified by the directive is executed in parallel. are implemented using stream queues. inserting synchronization into the kernel as appropriate.3. Programmers must be careful when using concurrent host and device execution. If not specified. such as summing the values in a vector or matrix. including kernel execution and data copies to and from device memory. By calling the runtime routine cudaThreadSynchronize. 8 . The one exception to this rule is a scalar reduction operation. For these operations. stream zero is special in that operations on the stream zero queue will begin only after all preceding operations on all queues are complete. In this way. As with any kernel. Each user-created stream manages its own queue. the host program should synchronize with the device to avoid erroneous results. Kernel Loop Directive CUDA Fortran allows automatic kernel generation and invocation from a region of host code containing one or more tightly nested loops. An application can manage more concurrency by using multiple streams. across the thread blocks and grid. it is the programmer's responsibility to ensure that parallel execution is legal and produces the correct answer. Asynchronous concurrent execution There are two components to asynchronous concurrent execution with CUDA Fortran. its value is treated as zero. Concurrent Host and Device Execution When a host subprogram calls a kernel subroutine. the call actually returns to the host program before the kernel subroutine begins execution.

Example 2 !$cuf kernel do <<< *.j) = b(i. n a(i. Alternatively.j) end do end do Without an explicit n on the do. Programming Guide Syntax The general form of the kernel directive is: !$cuf kernel do[(n)] <<< grid. specified as (*.j) end do end do In this example.*). 256 >>> do j = 1. !$cuf kernel do(2) <<< *.j) = b(i. ThreadIdx%y runs from 1 to 4 and is mapped onto the outer j loop.j) +c(i. the schedule applies just to the outermost loop. The grid shape.Chapter 2. In this case.j) = b(i. (32. using asterisks tells the compiler to choose a thread block shape and/or compute the grid shape from the thread block shape and the loop limits. The inner i dimension is run sequentially on each thread. the directive defines a two-dimensional thread block of size 32x4. Alternatively.4) >>> do j = 1.j) end do end do 9 . You might consider if the code in Example 2 would perform better if the two loops were interchanged.*). m do i = 1. Example 1 !$cuf kernel do(2) <<< (*. that is. m do i = 1. Loops which are not mapped onto the grid and block values are run sequentially on each thread. (256. The body of the doubly-nested loop is turned in the kernel body: ThreadIdx%x runs from 1 to 32 and is mapped onto the inner i loop. is computed by the compiler and runtime by dividing the loop trip counts n and m by the thread block size. so all iterations are computed. The grid and block values can be an integer scalar or a parenthesized list. block >>> The compiler maps the launch configuration specified by the grid and block values onto the outermost n loops. you could specify a configuration like the following in which the threads read and write the matrices in coalesced fashion. n a(i.j) +c(i. starting at loop n and working out. n a(i. m do i = 1. only the outer j loop is run in parallel with a thread block size of 256.j) +c(i.1) >>> do j = 1. the default value is 1.

• Arrays used or assigned in the loop must have the device attribute. atomic functions. To "unroll" more work into each thread. • Implicit loops and F90 array syntax are not allowed within the directive loops. 10 . as illustrated here: !$cuf kernel do(2) <<< (1. the j loop is mapped onto the grid x-dimension. gridDim%x is equal to m. • Only CUDA Fortran dataypes are allowed within the loops.Kernel Loop Directive Example 3 In Example 2. the compiler generates a correct sequence of synchronized operations to produce one copy either in device global memory or on the host. • CUDA Fortran intrinsic functions are allowed. when a reduction has a scalar target.j) +c(i. specify non-asterisk values for the grid. upper limit. in concert. Except in the case of reductions. • Subroutine and function calls to attributes(device) subprograms are allowed if they are in the same module as the code containing the directive. • The body of the loops may contain assignment statements.j) end do end do Now the threads in a thread block handle all values in the i direction. n a(i. • Scalars used or assigned in the loop must either have the device attribute. IF statements. • There can be no GOTO or EXIT statements within or between any loops that have been mapped onto the grid and block configuration values.j) = b(i. Specifically. Restrictions on the CUF kernel directive The following restrictions apply to CUF kernel directives: • If the directive specifies n dimensions. Further expansion of the work along the i direction and all work across the j dimension is handled by the mapping onto the grid dimensions. because the compiler skips over the constant 1 in the i loop grid size. the 256 threads in each block each do one element of the matrix addition. one for each thread.1) >>> do j = 1. • The tightly-nested DO loops must have invariant loop limits: the lower limit. One thread block is created for each j. or the compiler will make a device copy of that variable live for the duration of the loops. In CUDA built-in language. are not. incrementing by 256. but the device-specific intrinsics such as syncthreads. and GOTO statements. it must be followed by at least that many tightly-nested DO loops. if they are allowed in device code.*). Summation Example The simplest directive form for performing a dot product on two device arrays takes advantage of the properties for scalar use outlined previously. and increment must be invariant with respect to any other loop in the kernel do. (256. loops. etc. m do i = 1.

the device code is compiled for and runs on the host. the . allowing the programmer to use a host debugger. add the command line option –Mfixed.cuf extension specifies that the file is a free-format CUDA Fortran program. In emulation mode. Building a CUDA Fortran Program CUDA Fortran is supported by the PGI Fortran compilers when the filename uses a CUDA Fortran extension. The . It’s important to note that the emulation is far from exact. the compiler recognizes rsum as a scalar temporary which should be allocated locally on every thread. the compiler recognizes the use of the scalar and generates just one final result. emulation mode may execute a single thread block at a time. This CUF kernel can be followed by another CUF kernel in the same subprogram: !cuf kernel do <<< *. Programming Guide rsum = 0.CUF extension may also be used. To compile a fixed-format program. the host floating point units and intrinsics are used. * >>> do i = 1. in which case the program is processed by the preprocessor before being compiled. which may produce slightly different answers than the device units and intrinsics.0 !cuf kernel do <<< *. use of rsum on the host following this loop is undefined. 11 . n rsum = rsum + x(i) * y(i) end do For reductions. * >>> do i = 1. compile and link with the –Mcuda=emu command line option. n rsum = x(i) * y(i) z(i) = rsum end do In this CUF kernel. such as memory races. However. To build a program using emulation mode. This will not expose certain errors. Emulation Mode PGI Fortran compilers support an emulation mode for program development on workstations or systems without a CUDA-enabled GPU and for debugging. In emulation mode. In particular.Chapter 2. CUDA Fortran extensions can be enabled in any Fortran source file by adding the –Mcuda command line option.

12 .

Global Subroutines The global attribute may be explicitly specified on the Subroutine statement as follows: attributes(global) subroutine sub(…) Functions may not have the global attribute. These attributes are specified using the attributes(attr) prefix on the Subroutine or Function statement. A kernel subroutine is compiled as a kernel for execution on the device.Chapter 3. pure. A Subroutine may have the host. A host subprogram is compiled for execution on the host processor. global. A subroutine with the global attribute is called a kernel subroutine. elemental. or a host subprogram. A subprogram with no attributes prefix has the host attribute by default. and may not contain any other subprogram. or both. Host Subroutines and Functions The host attribute may be explicitly specified on the Subroutine or Function statement as follows: attributes(host) subroutine sub(…) attributes(host) integer function func(…) integer attributes(host) function func(…) The host attributes prefix may be preceded or followed by any other allowable subroutine or function prefix specifiers (recursive. if there is no attributes prefix on the subprogram statement. or may have both host and device attribute. Reference This chapter is the CUDA Fortran Language Reference. A kernel subroutine may not be contained in another subroutine or function. pure. This chapter describes how to specify the new attributes. function return datatype). or elemental. A kernel subroutine may not be recursive. or device attribute. so no other subroutine prefixes are allowed. then default rules are used. to be called from a host routine using an execution configuration. 13 . A subroutine or function with the host attribute is called a host subroutine or function. their meaning and restrictions. as described in the following sections. A Function may have the host or device attribute. New Subroutine and Function Attributes CUDA Fortran adds new attributes to subroutines and functions.

nor have the pure or elemental prefix on the subprogram statement. that is. Variable Attributes CUDA Fortran adds new attributes for variables and arrays. Variables declared in a device subprogram may have one of four attributes: they may be declared to be in device global memory. The device attribute can be specified with the attributes statement. in page-locked host memory space. The advantage of using pinned memory is that transfers between the device and pinned memory are faster and can be asynchronous. Device data A variable or array with the device attribute is defined to reside in the device global memory. Restrictions on Device Subprograms A subroutine or function with the device or global attribute must satisfy the following restrictions: • It may not be recursive. in constant memory space. refer to “Device code.” on page 21. A subroutine or function with the device attribute must appear within a Fortran module.Variable Attributes Device Subroutines and Functions The device attribute may be explicitly specified on the Subroutine or Function statement as follows: attributes(device) subroutine sub(…) attributes(device) datatype function func(…) datatype attributes(device) function func(…) A subroutine or function with the device attribute may not be recursive. Variables in modules may be declared to be in device global memory or constant memory space. to be device arrays of size 100. so no other subroutine or function prefixes are allowed. A subroutine or function with the device or kernel attribute is called a device subprogram. A device subprogram is compiled for execution on the device. the array may be declared to be in pinned memory. This section describes how to specify the new attributes and their meaning and restriction. a and b. nor have the recursive prefix on the subprogram statement. The following example declares two arrays. 14 . • It may not be pure or elemental. pure. CUDA Fortran adds a new attribute for allocatable arrays in host memory. or as an attribute on the type declaration statement. For more information. or elemental. except for the function return datatype. or in thread local memory. and may only be called from device subprograms in the same module. • It may not be contained in another subroutine or function. in the thread block shared memory. • It may not contain another subprogram.

or Allocatable attributes. c and d. • Constant variables and arrays may appear in modules.Chapter 3. Reference real :: a(100) attributes(device) :: a real. in that case. but may not be in a Common block or an Equivalence statement. In host subprograms. The following example declares two arrays. and the matching dummy argument must also have the device attribute. or may have adjustable extent. Target. • Device variables and arrays declared in a host subprogram cannot have the Save attribute. the subprogram interface must be explicit (in the Fortran sense). • Members of a derived type may not have the device attribute unless they are allocatable. • Device variables and arrays may not have the Pointer or Target attributes. real :: c(100) attributes(constant) :: c real. device data may only be used in the following manner: • In declaration statements • In Allocate and Deallocate statements • As an argument to the Allocated intrinsic function • As the source or destination in a data transfer assignment statement • As an actual argument to a kernel subroutine • As an actual argument to another host subprogram or runtime API call • As a dummy argument in a host subprogram A device array may have the allocatable attribute. • Device variables and arrays may appear in modules. constant :: d(100) These rules apply to constant data: • Constant data may not have the Pointer. but may not be in a Common block or an Equivalence statement. 15 . Constant data A variable or array with the constant attribute is defined to reside in the device constant memory space. or as an attribute on the type declaration statement. to be constant arrays of size 100. device :: b(100) These rules apply to device data: • An allocatable device array dynamically allocates device global memory. The constant attribute can be specified with the attributes statement. • Device variables and arrays may be passed as actual arguments to host and device subprograms.

In host subprograms. variables and arrays with the constant attribute may not be assigned or modified. • Within device subprograms. • Shared variables and arrays may be passed as actual arguments to from a device subprogram to another device subprogram. • Shared variables may not be in a Common block or Equivalence statement. s and t. Its size is determined at run time by the call to the kernel. that is. This memory is used for the 16 . the last dimension of a shared array may have an asterisk as its upper bound: real. the value of the bytes argument in the execution configuration is used to specify the number of bytes of shared memory that is dynamically allocated for each thread block. to be shared arrays of size 100. shared :: d(100) These rules apply to shared data: • Shared data may not have the Pointer.Variable Attributes • Members of a derived type may not have the constant attribute. The shared attribute can be specified with the attributes statement. Target. as long as the interface is explicit and the matching dummy argument has the shared attribute. When the kernel is called. • Constant variables and arrays may be passed as actual arguments to host and device subprograms. Shared arrays that are not dummy arguments may be declared as assumed-size arrays. A shared variable or array may only be declared and used inside a device subprogram. • Arrays with the constant attribute must have fixed size. data with the constant attribute may only be used in the following manner: • In declaration statements • As the source or destination in a data transfer assignment statement • As an actual argument to another host subprogram • As a dummy argument in a host subprogram Shared data A variable or array with the shared attribute is defined to reside in the shared memory space of a thread block. as long as the subprogram interface is explicit. and the matching dummy argument also has the constant attribute. real :: c(100) attributes(shared) :: c real. variables and arrays with the constant attribute may be read and written. The following example declares two arrays. or Allocatable attributes. shared :: x(*) Such an array has special significance. • Within host subprograms. • Members of a derived type may not have the shared attribute. or as an attribute on the type declaration statement.

and other supported methods for allocating device memory. specifically for dynamically allocating device arrays and host pinned arrays. attributes(global) subroutine madd( a. it must be fixed size. or whether the dummy argument has the pinned and allocatable attributes. or as an attribute on the type declaration statement. n ) real. Value dummy arguments In device subprograms. as is done in C. These arrays are dynamically allocated in host subprograms using the Allocate statement. allocatable.n) :: a. If a device array declared in a host subprogram does not have the Save attribute. dimension(n. p and q. Allocating Device Memory Device arrays can have the allocatable attribute. and the address of the argument is passed to the subprogram. Where the array is deallocated. The pinned attribute can be specified with the attributes statement. An array with the pinned attribute may be declared in a module or in a host subprogram. value :: n In this case. pinned :: q(:) Pinned arrays may be passed as arguments to host subprograms regardless of whether the interface is explicit. if there is more than one assumed-size shared memory array. Pinned arrays An allocatable array with the pinned attribute will be allocated in special page-locked host memory. and dynamically deallocated using the Deallocate statement. Allocating Device and Pinned Arrays This section describes extensions to the Allocate statement. This means the actual argument must be stored in device global memory. by adding the value attribute to the variable declaration. it will be automatically deallocated when the subprogram returns. b integer. 17 . If a shared array is not a dummy argument and not assumed-size. the declaration for the array must still have the pinned attribute. they are all implicitly equivalenced. or the deallocation may fail.Chapter 3. Reference assumed-size shared memory arrays in that thread block. starting at the same shared memory address. b. Scalar arguments can be passed by value. to be pinned allocatable arrays. Programmers must take this into account when coding. following the rules of Fortran. The variable arrays corresponding to the dummy arguments a and b must be set up before the call to reside on the device. real :: p(:) allocatable :: p attributes(pinned) :: p real. the value of n can be passed from the host without needing to reside in device memory. when such memory is available. dummy arguments are passed by default by reference. The following example declares two arrays.

pinned :: p(:) allocate(p(5000).Allocating Device and Pinned Arrays real. device :: ndev allocate(ndev) ndev = 100 The language also supports the ability to create the equivalent of automatic and local device arrays without using the allocate statement. refer to“Memory Management. If no such memory space is available. The cudaMalloc function can be used to allocate single-dimensional arrays of the supported intrinsic datatypes. allocatable.c. the compiler allocates the array in normal paged host memory. These arrays will also have a lifetime of the subprogram as is usual with the Fortran language: subroutine vfunc(a. or if it is exhausted. These functions return memory which will bypass certain Fortran allocatable properties such as automatic deallocation. Fortran interfaces to the CUDA memory management runtime routines are provided. To use these. declare and initialize the scalar on the host as: integer. and thus the arrays are treated more like C malloc’ed areas. device :: b(:) allocate(b(5024).stat=istat) … if(allocated(b)) deallocate(b) Scalar variables can be allocated on the device using the Fortran 2003 allocatable scalar feature. real. allocatable. device :: atmp(4) … end subroutine vfunc ! adev and atmp are deallocated Allocating Device Memory Using Runtime Routines For programmers comfortable with the CUDA C programming environment.” on page 32.stat=istat) … if(allocated(p)) deallocate(p) 18 . Otherwise. allocatable. allocatable. device :: adev(n) real. pinned allocatable arrays work and act like any other allocatable array on the host.n) real. The compiler will generate code to allocate the array in host page-locked memory. Allocating Pinned Memory Allocatable arrays with the pinned attribute are dynamically allocated using the Allocate statement. if available. Mixing standard Fortran allocate/deallocate with the runtime Malloc/Free for a given array is not supported. 100) … istat = cudaFree(v) For a complete list of the memory management runtime routines. and cudaFree can be used to free it: real. device :: v(:) istat = cudaMalloc(v.

copies data from the host memory to the device global memory. Data transfer between host and device memory This section provides methods to transfer data between the host and device memory. and cdev are conforming device arrays. copies data from the device global memory to the host memory. and the right hand is a host variable or host array or array section. if a. • An assignment statement where the left hand side is a device variable or device array or array section. the rule of thumb is all arithmetic or operations must occur on the host. meaning the data copy waits until all previous kernels and data copies complete. and adev. Note Using assignment statements to read or write device or constant data implicitly uses CUDA stream zero. pinned=plog) if (. Similarly. • An assignment statement with a device variable or device array or array section on both sides of the assignment statement copies data between two device variables or arrays.not. Compiler-generated temporary arrays are generated to accommodate the host copies of device data as needed.Chapter 3. For instance. plog) then . In general. logical plog allocate(p(5000). which normally only allows one device array to appear on the righthand-side of an expression. . • An assignment statement where the left hand side is a host variable or host array or array section. Data Transfer Using Assignment Statements You can copy variables and arrays from the host memory to the device memory by using simple assignment statements in host subprograms. . Reference To determine whether or not the allocation from page-locked memory was successful. b. you can use simple assignment statements to copy or assign variables or arrays with the constant attribute. Implicit Data Transfer in Expressions Some limited data transfer can be enclosed within expressions. and the right hand side is a device variable or device array or array section. bdev. an additional PINNED keyword is added to the allocate statement. stat=istat. This means such data copies are synchronous. and c are conforming host arrays. the following expressions are legal: a = adev adev = a b = a + adev c = x * adev + b 19 . It returns a logical success value.

or of type(dim3). If it is type(dim3). the value of grid%z must be one. as well as the amount of shared memory to use for assumed-size shared memory arrays. it is converted to dim3(grid. If block is an integer. stream >>> • grid is an integer. and the implementation. or from one device array to another. which must be less than the maximum supported by the device. or of type(dim3). bytes. The execution configuration is specified after the subroutine name in the call statement. and the associated stream. refer to “Memory Management. These functions can transfer data either from the host to device.1.1. bl • block is an integer. For a complete list of memory management runtime routines. If it is type(dim3). it has the form: <<< grid. Counts expressed in arguments to the Fortran runtime routines are expressed in terms of data type elements.1). or would be more efficient written in another way.1). Array slices are also supported. cur. Invoking a kernel subroutine A call to a kernel subroutine must give the execution configuration for the call. amount of contiguous data in the slices. device to host. the kind parameter to the Memcpy routines is optional in Fortran since the attributes of the arrays are explicitly declared. 512) For those familiar with the CUDA C routines. 20 . block. The cudaMemcpy function can be used to copy data between the host and the GPU: real.” on page 32. not bytes. Fortran interfaces to the CUDA memory management runtime routines are provided. The product grid%x*grid%y gives the number of thread blocks to launch. it is converted to dim3(block. The execution configuration gives the size and shape of the grid and thread blocks that execute the function. and their performance is dependent on the size of the slice. the number of threads per thread block is block%x*block%y*block%z. or both: c = adev + bdev adev = adev + a b = sqrt(adev) Elemental transfers are supported by the language but will perform poorly. device :: wrk(1024) real cur(512) istat = cudaMemcpy(wrk.Invoking a kernel subroutine The following expressions are not legal as they either promote a false impression of where the actual computation occurs. Data Transfer Using Runtime Routines For programmers comfortable with the CUDA C programming environment. If grid is an integer.

4(default). or declared in device subprograms.4(default). Table 3. It specifies the stream to which this call is enqueued. they may be of derived type. and specifies the number of bytes of shared memory to be allocated for each thread block to use for assumed-size shared memory arrays. or shared attributes. such as the function arguments and execution configuration arguments.1. a kernel subroutine attributes(global) subroutine sub( a ) can be called like: call sub <<< DG.8 1 (default) Additionally. if present. Device Code Intrinsic Datatypes Type integer logical real double precision complex character(len=1) Type Kind 1. where the members of the derived type have one of the allowed intrinsic datatypes.5 on page 41.8 1. and have a value of zero.Chapter 3. • stream is optional.8 equivalent to real(kind=8) 4(default). Device code Datatypes allowed Variables and arrays with the device. constant. For instance. or if bytes is greater than the shared memory available. Reference • bytes is optional. defined as type(dim3) integer(kind=4) :: x. or another allowed derived type. are limited to the types described in this section. If not specified. or a value returned by a call to cudaStreamCreate. DB >>> ( A ) The function call will fail if the grid or block arguments are greater than the maximum sizes allowed. See Section 4. if present.2. For more information. The system module cudafor includes definitions of the derived type dim3. allowing for static shared memory declared in the kernel and for other dedicated uses.2.y. it must be a scalar integer.8 4(default).” on page 16. it must be an integer. refer to “Shared data. the value zero is used. They may have any of the intrinsic datatypes in the following table.z end type 21 .

• The variables threadidx.Device code Builtin variables The system module cudafor declares several predefined variables. for one. Fortran Numeric and Logical Intrinsics Name abs aimag aint anint ceiling cmplx conjg dim floor Argument Datatypes integer. • The variable griddim contains the dimensions of the grid. They are declared as follows: type(dim3) :: threadidx. blockdim has the same value for all threads in the same grid. It has constant value. The value of blockidx%z is always one. blockdim. the threadidx%y and/or threadidx%z components have the value one. for onedimensional grids. complex complex real real real real or (real. blockidx. These variables are read-only. real 22 . blockidx. the value of griddim%z is always one.real) complex integer. • The variable warpsize contains the number of threads in a warp. real. and griddim are available only in device subprograms. • The variable blockidx contains the block index within the grid. complex integer. real integer.or two-dimensional thread blocks.2. real real Name int logical max min mod modulo nint real sign Argument Datatypes integer. complex logical integer. currently defined to be 32. real. real. the value of griddim%z is always one. • The variable blockdim contains the dimensions of the thread block. Fortran intrinsics This section lists the Fortran intrinsic functions allowed in device subprograms.or two-dimensional thread blocks. real integer. griddim integer(4) :: warpsize • The variable threadidx contains the thread index within its thread block. The value of griddim is the same for all threads in the same grid. the value of griddim%y is one for one-dimensional grids. blockdim. real real integer. The value of blockidx is the same for all threads in the same thread block. blockidx%y has the value one. the blockdim%y and/or blockdim %z components have the value one. Fortran Numeric and Logical Intrinsics Table 3. real integer. as with threadidx. for one.

complex real real. real. complex real real. real integer.3. complex integer (integer. complex Name log log10 sin sinh sqrt tan tanh Argument Datatypes real. real real real Name precision radix range selected_int_kind selected_real_kind tiny Argument Datatypes real. Reference Fortran Mathematical Intrinsics Table 3. complex real real.5. complex real real Fortran Numeric Inquiry Intrinsics Table 3. complex integer.integer) real Fortran Bit Manipulation Intrinsics Table 3. Fortran Bit Manipulation Intrinsics Name btest iand ibclr ibits ibset ieor ior Argument Datatypes integer integer integer integer integer integer integer Name ishft ishftc leadz mvbits not popcnt poppar Argument Datatypes integer integer integer integer integer integer integer 23 .4. real real integer.real) real. Fortran Numeric Inquiry Intrinsics Name bit_size digits epsilon huge maxexponent minexponent Argument Datatypes integer integer. Fortran Mathematical Intrinsics Name acos asin atan atan2 cos cosh exp Argument Datatypes real real real (real.Chapter 3.

it has no arguments: void syncthreads() Sometimes threads within a block access the same addresses in shared or global memory. real integer.7. SYNCTHREADS The syncthreads intrinsic subroutine acts as a barrier synchronization for all threads in a single thread block. Synchronization Functions The synchronization functions control the synchronization of various threads during execution of thread blocks. real integer. thus creating potential read-after-write. complex integer. . syncthreads syncthreads_count syncthreads_and syncthread_or threadfence threadfence_block threadfence_system For detailed information on these functions. This 24 . real. complex Fortran Random Number Intrinsics Table 3. real integer. real. refer to “Thread Management.6.” on page 32. or write-after-write hazards for some of these memory accesses. use syncthreads()to specify synchronization points in the kernel. Fortran Reduction Intrinsics Name all any count maxloc maxval Argument Datatypes logical logical logical integer.Device code Fortran Reduction Intrinsics Table 3. To avoid these potential issues. Fortran Random Number Intrinsics Name random_number random_seed Argument Datatypes real integer New Intrinsic Functions This section describes the new intrinsic functions and subroutines supported in device subprograms. real Name minloc minval product sum Argument Datatypes integer. write-after-read.

If any thread in a thread block issues a call to syncthreads. When all blocks are done. Each thread in a thread block pauses at the syncthreads call until all threads have reached that call. If no fence is placed between storing the partial sum and incrementing the counter. For example. each block atomically increments a counter to signal that it is done with computing and storing its partial sum. In addition. suppose you use a kernel to compute the sum of an array of N numbers in one call. syncthreads_and evaluates the integer argument int_value for all threads of the block and returns non-zero if and only if int_value evaluates to non-zero for all of them. all threads must also reach and execute the same call statement. Each block first sums a subset of the array and stores the result in global memory. like syncthreads. like syncthreads. syncthreads_or evaluates the integer argument int_value for all threads of the block and returns non-zero if and only if int_value evaluates to non-zero for any of them. the last block done reads each of these partial sums from global memory and sums them to obtain the final result. Reference intrinsic acts as a barrier at which all threads in the block must wait before any thread is allowed to proceed. the counter might increment before the partial sum is stored. syncthreads_count evaluates the integer argument int_value for all threads of the block and returns the number of threads for which int_value evaluates to non-zero.Chapter 3. SYNCTHREADS_OR integer syncthreads_or(int_value) syncthreads_or. acts as a barrier at which all threads in the block must wait before any thread is allowed to proceed. THREADFENCE void threadfence() threadfence acts as a memory fence. other threads may see the effects of these memory writes in a different order. SYNCTHREADS_AND integer syncthreads_and(int_value) syncthreads_and. Memory Fences In general. when a thread issues a series of writes to memory in a particular order. threadfence_block(). other threads may see the effects of these memory writes in a different order. acts as a barrier at which all threads in the block must wait before any thread is allowed to proceed. To determine which block is finished last. like syncthreads. Typically. Threads within a block cooperate and share data by synchronizing their execution to coordinate memory accesses. In addition. 25 . In addition. creating a wait. SYNCTHREADS_COUNT integer syncthreads_count(int_value) syncthreads_count. or the kernel fails to complete correctly. You can use threadfence(). when a thread issues a series of writes to memory in a particular order. and threadfence_system() to create a memory fence to enforce ordering. acts a as a barrier at which all threads in the block must wait before any thread is allowed to proceed.

true.Device code threadfence() is one method to enforce a specific order. The function allthreads evaluates its argument for all threads in the current warp. THREADFENCE_SYSTEM void threadfence_system() threadfence_system acts as a memory fence. only if the value of the argument is . The clock frequency can be determined by calling cudaGetDeviceProperties. GPU_TIME The gpu_time intrinsic returns the value of the clock cycle counter on the GPU. ALLTHREADS The allthreads function is a warp-vote operation with a single scalar logical argument: if( allthreads(a(i)<0.” on page 32. creating a wait until all global and shared memory accesses made by the calling thread prior to threadfence_block() are visible to all threads in the thread block for all accesses.0 or higher. Warp-Vote Operations Warp-vote operations are only supported by devices with compute capability 1. for all threads in the warp. It has a single argument: integer(8) clock call gpu_time(clock) The argument to gpu_time is set to the value of the clock cycle counter.0) ) allneg = . Each of these functions has a single argument.2 and higher. The value of the function is .true. 26 . as described in “cudaGetDeviceProperties. creating a wait until all global and shared memory accesses made by the calling thread prior to threadfence_system() are visible to: • All threads in the thread block for shared memory accesses • All threads in the device for global memory accesses • Host threads for page-locked host memory accesses threadfence_system() is only supported by devices of compute capability 2.true. All global and shared memory accesses made by the calling thread prior to threadfence() are visible to: • All threads in the thread block for shared memory accesses • All threads in the device for global memory accesses THREADFENCE_BLOCK void threadfence_block() threadfence_block acts as a memory fence.

false. value ) Additional Atomic Update mem = mem + value mem = mem – value 27 .1 and higher.0) ) allneg = .Chapter 3. Arithmetic and Bitwise Atomic Functions Function atomicadd( mem. This function is only supported by devices of compute capability 2.2 or higher is required if the first argument has the shared attribute.8. The function anythread evaluates its argument for all threads in the current warp. Atomic Functions The atomic functions read and write the value of their first operand.0. The atomic functions return correct values even if multiple threads in the same or different thread blocks try to read and update the same location without any synchronization. Note The return value for each of these functions is the first argument. These functions are: Table 3. They also combine that value with the value of the second argument. The value of the function is . value ) atomicsub( mem. Compute capability 1. for all threads in the warp. which must be a variable or array element in shared memory (with the shared attribute) or in device global memory (with the device attribute). depending on the function. Reference ANYTHREAD The anythread function is a warp-vote operation with a single scalar logical argument: if( anythread(a(i)<0.true. and store the combined value back to the first argument location.true. only if the value of the argument is . BALLOT The ballot function is a warp-vote operation with a single integer argument: unsigned integer ballot(int_value) The function ballot evaluates the argument int_value for all threads of the warp and returns an integer whose Nth bit is set if and only if int_value evaluates to non-zero for the Nth thread of the warp. Atomic functions are only supported by devices with compute capability 1.false. Example: if( ballot(int_value) ) allneg = . Arithmetic and Bitwise Atomic Functions These atomic functions read and return the value of the first argument. mem. Both arguments must be of type integer(kind=4).

Counting Atomic Functions Function atomicinc( mem. These functions are: Table 3. value ) atomicexch( mem.value) mem = value Counting Atomic Functions These atomic functions read and return the value of the first argument. It also compares the first argument with the second argument. The function is: 28 . value ) atomicand( mem. value ) atomicmin( mem. mem. Note The return value for each of these functions is the first argument.Device code atomicmax( mem.value) mem = ior(mem.and.value) mem = iand(mem. imax ) Additional Atomic Update if (mem<imax) then mem = mem+1 else mem = 0 endif if (mem<imax . Note The return value for this function is the first argument. imax ) Compare and Swap Atomic Function This atomic function reads and returns the value of the first argument. value ) atomicxor( mem.value) mem = min(mem. value ) mem = max(mem. mem>0) then mem = mem-1 else mem = imax endif atomicdec( mem. These functions are intended to implement circular counters. value ) atomicor( mem. counting up to or down from a maximum value specified in the second argument.value) mem = ieor(mem.9. and atomically stores a new value back to the first argument location if the first and second argument are equal. They also compare the first argument with the second argument. mem. Both arguments must be of type integer(kind=4). depending on the result of the comparison. All three arguments must be of type integer(kind=4). and stores a new value back to the first argument location.

• Optional arguments are not allowed. OPEN. • Automatic arrays must be fixed size. • Fortran intrinsic functions not listed in Section 3.val) Additional Atomic Update if (mem == comp) then mem = val endif Restrictions This section lists restrictions on statements and features that can appear in device subprograms. • Subroutine and function calls are supported only if they can be inlined. • Recursive subroutines and functions are not allowed. ENDFILE. character substrings are not supported. BACKSPACE. where A is a variable or expression.comp.3 are not supported. • Floating point exception handling is not supported. INQUIRE. returns the number of bytes required to hold the value of A. FORMAT. WRITE.6. REWIND. such as the new sizeof intrinsic function. • Cray pointers are not supported. Compare and Swap Atomic Function Function atomiccas(mem. • Objects with character type must have LEN=1. • STOP and PAUSE statements are not allowed. j j = sizeof(i) ! this assigns the value 4 to j 29 . Host code Host subprograms may use intrinsic functions. NAMELIST. • Input/Output statements are not allowed: READ. PRINT. • Alternate return specifications are not allowed. • Objects with the Pointer and Allocatable attribute are not allowed. • ENTRY statements are not allowed. Reference Table 3. SIZEOF Intrinsic A call to sizeof(A). CLOSE. integer(kind=4) :: i.10.Chapter 3.

30 .

Runtime APIs The system module cudafor defines the interfaces to the Runtime API routines. cudaSetDevice integer function cudaSetDevice( devnum ) integer. intent(out) :: numdev cudaGetDeviceCount assigns the number of available devices to its first argument. To interpret the error codes. Tip When doing timing runs. intent(in) :: devnum cudaSetDevice selects the device to associate with this host thread. Initialization No explicit initialization is required. Device Management Use the functions in this section for device management. Most of the runtime API routines are integer functions that return an error code. refer to “Error Handling. and a nonzero value if there was an error. the runtime initializes and connects to the device the first time a runtime routine is called or a device array is allocated. cudaSetDeviceFlags integer function cudaSetDevice( flags ) integer. cudaGetDeviceCount integer function cudaGetDeviceCount( numdev ) integer. they return a value of zero if the call was successful. intent(in) :: flags cudaSetDeviceFlags records how the CUDA runtime interacts with this host thread.Chapter 4. 31 .” on page 41. be aware that initialization can add some overhead.

To avoid these potential issues. thus creating potential read-after-write. Consistent use of TYPE(C_PTR) and 32 . Some can also take C types. use the functions in this section for thread management. cudaThreadExit integer function cudaThreadExit() cudaThreadExit explicitly cleans up all runtime-related CUDA resources associated with the host thread. it is implicitly called when the host thread exits. It may return an error condition if one of the preceding operations fails. devnum ) type(cudadeviceprop). provided through the Fortran 2003 iso_c_binding module.Thread Management cudaGetDevice integer function cudaGetDevice( devnum ) integer. as arguments to simplify interfacing to existing CUDA C code. or write-after-write hazards for some of these memory accesses. prop ) integer. cudaGetDeviceProperties integer function cudaGetDeviceProperties( prop. Thread Management Sometimes threads within a block access the same addresses in shared or global memory. Any subsequent CUDA calls or operations will reinitialize the runtime. cudaThreadSynchronize integer function cudaThreadSynchronize() cudaThreadSynchronize blocks execution of the host subprogram until all preceding kernels and operations are complete. intent(out) :: devnum type(cudadeviceprop). Calling cudaThreadExit is optional. intent(out) :: prop integer. write-after-read. Memory Management Many of the memory management routines can take device arrays as arguments. defined in the cudafor module. intent(in) :: devnum cudaGetDeviceProperties returns the properties of a given device. intent(out) :: devnum cudaGetDevice assigns the device number associated with this host thread to its first argument. cudaChooseDevice integer function cudaChooseDevice ( devnum. intent(in) :: prop cudaChooseDevice assigns the device number that best matches the properties given in prop to its first argument. as TYPE(C_DEVPTR). CUDA Fortran has extended the F2003 derived type TYPE(C_PTR) by providing a C device pointer.

proper Fortran pointers for device data are supported. an allocatable device array as the second argument. Use the functions in this section for memory management. in the future. Runtime APIs TYPE(C_DEVPTR). in which case the count is in bytes. height) cudaMallocPitch allocates data on the device. cudaMallocArray integer function cudaMallocArray(carray. Similarly.1.1. “Device Code Intrinsic Datatypes. cudaFreeArray integer function cudaFreeArray(carray) type(cudaArrayPtr) :: carray 33 .Chapter 4. Or. height cudaMallocArray allocates a data array on the device. devptr may be any allocatable. cudaMalloc integer function cudaMalloc(devptr. Under CUDA Fortran. two-dimensional device array of a supported type specified in Table 3. Or. width. Both of these features are subject to change when. width. and in effect transfer the allocation to the Fortran array. count) cudaMalloc allocates data on the device.1. Currently. devptr may be of TYPE(C_DEVPTR). The count is in terms of elements. it is possible to construct a Fortran device array out of a TYPE(C_DEVPTR) by using an extension of the iso_c_binding subroutine c_f_pointer. “Device Code Intrinsic Datatypes. a shape as the third argument. devptr may be any allocatable device array of a supported type specified in Table 3. cdesc. one-dimensional device array of a supported type specified in Table 3. cudaMallocPitch may pad the data. should be of benefit. “Device Code Intrinsic Datatypes. devptr may be any allocatable.” on page 21.” on page 21. height) type(cudaArrayPtr) :: carray type(cudaChannelFormatDesc) :: cdesc integer :: width. The height is an integer. as well as consistency checks between Fortran device arrays and host arrays. and the padded width is returned in the variable pitch. cudaFree integer function cudaFree(devptr) cudaFree deallocates data on the device. pitch. in which case the integer values are expressed in bytes. c_f_pointer will take a TYPE(C_DEVPTR) as the first argument. The width is in terms of number of elements. devptr may also be of TYPE(C_DEVPTR).” on page 21. cudaMallocPitch integer function cudaMallocPitch(devptr. devptr may be of TYPE(C_DEVPTR). there is also a function C_DEVLOC() defined which will create a TYPE(C_DEVPTR) that holds the C address of the Fortran device array argument.

Memory Management cudaFreeArray frees an array that was allocated on the device. The value must match in type and kind. src. “Device Code Intrinsic Datatypes. it must be one of the defined enums cudaMemcpyHostToDevice. width. height) cudaMemset2D sets an array to the specified value. and the lowest byte of value is used.” on page 21. dst and src may be of TYPE(C_DEVPTR) or TYPE(C_PTR). for more information.1. and height are in terms of elements. cudaMemcpyAsync integer function cudaMemcpyAsync(dst. count. count) cudaMemset sets a location or array to the specified value. cudaMemset integer function cudaMemset(devptr. kdir may be optional. cudaMemcpy2D integer function cudaMemcpy2D(dst. The pitch. for more information. spitch. Or. The value must match in type and kind. src. “Device Code Intrinsic Datatypes. value. scalar or array. in which case the pitch. refer to “Data Transfer Using Runtime Routines. refer to “Data Transfer Using Runtime Routines. in which case the count is in term of bytes.1. dst and src may be any device or host. count. height. it must be one of the defined enums cudaMemcpyHostToDevice. width. or cudaMemcpyDeviceToDevice. stream) cudaMemcpy copies data from one location to another.” on page 20. Alternatively. If kdir is specified. kdir) cudaMemcpy copies data from one location to another. otherwise the stream argument is optional and defaults to zero.1. “Device Code Intrinsic Datatypes. and height are in terms of bytes. “Device Code Intrinsic Datatypes. cudaMemcpyDeviceToHost.” on page 21. The count is in terms of elements.1. “Device Code Intrinsic Datatypes. The count is in terms of elements. The count is in terms of elements. value. cudaMemset2D integer function cudaMemset2D(devptr. of a supported type specified in Table 3. of a supported type specified in Table 3. The copy can be associated with a stream by passing a non-zero stream argument. kdir) cudaMemcpy2D copies data from one location to another. cudaMemcpy integer function cudaMemcpy(dst. The width 34 . devptr may be any device array of a supported type specified in Table 3.” on page 21. devptr may be of TYPE(C_DEVPTR).” on page 21. in which case the count is in term of bytes. dpitch. If kdir is specified. dst and src may be any device or host array. src. pitch. This function operates on page-locked host memory only. Or. of a supported type specified in Table 3. dst and src may be of TYPE(C_DEVPTR) or TYPE(C_PTR). in which case the count is in term of bytes. width. kdir may be optional. cudaMemcpyDeviceToHost. and the lowest byte of value is used. or cudaMemcpyDeviceToDevice. dst and src may be any device or host. scalar or array. Alternatively.1. kdir. devptr may be any device scalar or array of a supported type specified in Table 3.” on page 20. width.” on page 21. devptr may be of TYPE(C_DEVPTR).

“Device Code Intrinsic Datatypes. If kdir is specified. src. kdir cudaMemcpyToArray copies array data to and from the device. width. If kdir is specified. Alternatively. dpitch. kdir) type(cudaArrayPtr) :: dsta integer :: dstx. or cudaMemcpyDeviceToDevice. refer to “Data Transfer Using Runtime Routines. dsty. count. Alternatively. This function operates on page-locked host memory only. refer to “Data Transfer Using Runtime Routines. cudaMemcpy2DAsync integer function cudaMemcpy2DAsync(dst. otherwise the stream argument is optional and defaults to zero.” on page 20. spitch. srca. cudaMemcpyToArray integer function cudaMemcpyToArray(dsta. srcy. count. spitch. height. dstx. cudaMemcpyDeviceToHost. height.” on page 20. srca. kdir) type(cudaArrayPtr) :: srca integer :: dpitch.1. kdir. height. dsty. for more information. cudaMemcpy2DFromArray integer function cudaMemcpy2DFromArray(dst. cudaMemcpyDeviceToHost. 35 . in which case the width and height are in term of bytes.” on page 21. dsty. dsty. dstx. or cudaMemcpyDeviceToDevice. in which case the width and height are in term of bytes. srcy. cudaMemcpy2DToArray integer function cudaMemcpy2DToArray(dsta. The copy can be associated with a stream by passing a non-zero stream argument. for more information. width. kdir may be optional. dst and src may be of TYPE(C_DEVPTR) or TYPE(C_PTR). kdir cudaMemcpy2DFromArray copies array data to and from the device. srcx. kdir cudaMemcpy2DToArray copies array data to and from the device. dst and src may be of TYPE(C_DEVPTR) or TYPE(C_PTR). stream) cudaMemcpy2D copies data from one location to another. The width and height are in terms of elements. width. src. srcx. cudaMemcpyFromArray integer function cudaMemcpyFromArray(dst.Chapter 4. count. srcy. width. src. count. kdir may be optional. it must be one of the defined enums cudaMemcpyHostToDevice. dst and src may be any device or host array. height. Runtime APIs and height are in terms of elements. of a supported type specified in Table 3. dsty. it must be one of the defined enums cudaMemcpyHostToDevice. kdir cudaMemcpyFromArray copies array data to and from the device. dpitch. kdir) type(cudaArrayPtr) :: dsta integer :: dstx. height. kdir) type(cudaArrayPtr) :: srca integer :: dstx. srcx. width. spitch.

srca. srca integer :: dstx. dsty. cext) type(cudaPitchedPtr) :: pitchptr integer :: value type(cudaExtent) :: cext cudaMemset3D sets elements of an array. width. dsty. intent(in) :: cext cudaMalloc3D allocates data on the device. srcx. count.” on page 21. Alternatively. kdir) type(cudaArrayPtr) :: dsta. dsty. cdesc. kdir cudaMemcpy2DArrayToArray copies array data to and from the device. cudaMalloc3D integer function cudaMalloc3D(pitchptr. srcx. dstx. cext is also a derived type which holds the extents of the allocated array. height. cudaMemset3D integer function cudaMemset3D(pitchptr. cext) type(cudaArrayPtr) :: carray type(cudaChannelFormatDesc) :: cdesc type(cudaExtent) :: cext cudaMalloc3DArray allocates array data on the device. srcy. kdir cudaMemcpyArrayToArray copies array data to and from the device. srcy. width. the extents in each dimension specified by cext. three-dimensional device array of a supported type specified in “Datatypes allowed. which was allocated with cudaMalloc3D to a specified value. 36 . intent(out) :: pitchptr type(cudaExtent). srcy. srca integer :: dstx. dsty. value. cext) type(cudaPitchedPtr).Memory Management cudaMemcpyArrayToArray integer function cudaMemcpyArrayToArray(dsta. count. cudaMemcpy3D integer function cudaMemcpy3D(p) type(cudaMemcpy3DParms) :: p cudaMemcpy3D copies elements from one 3D array to another as specified by the data held in the derived type p. kdir) type(cudaArrayPtr) :: dsta. pitchptr is a derived type defined in the cudafor module. cudaMalloc3DArray integer function cudaMalloc3DArray(carray. height. pitchptr may be any allocatable. srcx. dstx. srca. srcx. srcy. cudaMemcpy2DArrayToArray integer function cudaMemcpy2DArrayToArray(dsta.

optional :: stream cudaMemcpyToSymbol copies data from the source to a device area in global or constant memory space referenced by a symbol.” on page 21. kdir integer. src. kdir integer. kdir. offset. count. stream) type(cudaSymbol) :: symbol integer :: count. src may be any host scalar or array of a supported type specified in “Datatypes allowed. cudaMemcpyFromSymbol integer function cudaMemcpyFromSymbol(dst. offset. count. src. This function operates on page-locked host memory only. offset.” on page 21. The copy can be associated with a stream by passing a non-zero stream argument. optional :: stream 37 . kdir) type(cudaSymbol) :: symbol integer :: count. The copy can be associated with a stream by passing a non-zero stream argument. symbol. kdir. count. cudaMemcpyFromSymbolAsync integer function cudaMemcpyFromSymbolAsync(dst. kdir) type(cudaSymbol) :: symbol integer :: count. stream) type(cudaSymbol) :: symbol integer :: count. dst may be any host scalar or array of a supported type specified in “Datatypes allowed. offset. symbol.” on page 21. cudaMemcpyToSymbol integer function cudaMemcpyToSymbol(symbol. offset. count. offset.Chapter 4. stream) type(cudaMemcpy3DParms) :: p integer :: stream cudaMemcpy3DAsync copies elements from one 3D array to another as specified by the data held in the derived type p. offset. offset. Runtime APIs cudaMemcpy3DAsync integer function cudaMemcpy3D(p. kdir cudaMemcpyFromSymbol copies data from a device area in global or constant memory space referenced by a symbol to a destination on the host. kdir cudaMemcpyToSymbol copies data from the source to a device area in global or constant memory space referenced by a symbol. This function operates on page-locked host memory only. The count is in terms of elements. The count is in terms of elements. The count is in terms of elements. cudaMemcpyToSymbolAsync integer function cudaMemcpyToSymbolAsync(symbol. src may be any host scalar or array of a supported type as specified in “Datatypes allowed.

” on page 21. This function operates on page-locked host memory only. flags) type(C_DEVPTR) :: devptr type(C_PTR) :: hostptr integer :: flags cudaHostGetDevicePointer returns a pointer to a device memory address corresponding to the pinned memory on the host. hostptr. flags cudaHostAlloc allocates pinned memory on the host. 100) Vx = 0. device :: vx(:) csvx = “vx” Istat = cudaGetSymbolAddress(cdvx. cudaGetSymbolAddress integer function cudaGetSymbolAddress(devptr. dst may be any host scalar or array of a supported type specified in “Datatypes allowed. and read and written by. cudaHostGetDevicePointer integer function cudaHostGetDevicePointer(devptr. hostptr is a pinned memory buffer that was allocated via cudaHostAlloc(). The copy can be associated with a stream by passing a non-zero stream argument. It returns in hostptr the address of the pagelocked allocation. allocatable.0 cudaGetSymbolSize integer function cudaGetSymbolSize(size.Memory Management cudaMemcpyFromSymbol copies data from a device area in global or constant memory space referenced by a symbol to a destination on the host. csvx) Call c_f_pointer(cdvx. Size is in bytes. The count is in terms of elements. a kernel which runs on 38 . symbol) type(C_DEVPTR) :: devptr type(cudaSymbol) :: symbol cudaGetSymbolAddress returns in the devptr argument the address of symbol on the device. cudaHostAlloc integer function cudaHostAlloc(hostptr. symbol) integer :: size type(cudaSymbol) :: symbol cudaGetSymbolSize sets the variable size to the size of a device area in global or constant memory space referenced by the symbol. flags) type(C_PTR) :: hostptr integer :: size. size. The normal iso_c_binding subroutine c_f_pointer can be used to move the type(c_ptr) to a Fortran pointer. The following code sequence initializes a global device array “vx” from a CUDA C kernel: type(cudaSymbol) :: csvx type(c_devptr) :: cdvx real. The flags argument enables different options to be specified that affect the allocation. vx. or returns an error if the memory is unavailable. A symbol can be set to an external device name via a character string. It returns in devptr an address that can be passed to.

asynchronous operations. and the value cudaErrorNotReady if not. cudaStreamCreate integer function cudaStreamCreate( stream ) integer. It returns in hostptr the address of the page-locked allocation. The normal iso_c_binding subroutine c_f_pointer can be used to move the type(c_ptr) to a Fortran pointer. size) type(C_PTR) :: hostptr integer :: size cudaMallocHost allocates pinned memory on the host. or returns an error if the memory is unavailable. intent(in) :: stream cudaStreamDestroy releases any resources associated with the given stream. The normal iso_c_binding subroutine c_f_pointer can be used to move the type(c_devptr)to a device array. it returns zero (success) if all operations are complete. cudaStreamSynchronize integer function cudaStreamSynchronize( stream ) integer.Chapter 4. It may return error codes from previous. cudaMallocHost integer function cudaMallocHost(hostptr. Stream Management Use the functions in this section for stream management. Runtime APIs the device. 39 . intent(out) :: stream cudaStreamCreate creates an asynchronous stream and assigns its identifier to its first argument. intent(in) :: stream cudaStreamQuery tests whether all operations enqueued to the selected stream are complete. cudaStreamDestroy integer function cudaStreamDestroy( stream ) integer. cudaStreamQuery integer function cudaStreamQuery( stream ) integer. size is in bytes. intent(in) :: stream cudaStreamSynchronize blocks execution of the host subprogram until all preceding kernels and operations associated with the given stream are complete. cudaFreeHost integer function cudaFreeHost(hostptr) type(C_PTR) :: hostptr cudaFreeHost deallocates pinned memory on the host allocated with cudaMalloHost. It may also return another error condition if some asynchronous operations failed. The flags argument is provided for future releases.

intent(in) :: event integer. cudaEventCreate integer function cudaEventCreate( event ) type(cudaEvent). intent(in) :: stream cudaEventRecord issues an operation to the given stream to record an event. It returns cudaErrorInvalidValue if cudaEventRecord has not been called for this event. It returns success (zero) if the event has been recorded. intent(in) :: event cudaEventSynchronize blocks until the event has been recorded.Event Management Event Management Use the functions in this section to manage events. end) float :: time type(cudaEvent). The event is recorded after all preceding operations in the stream are complete. It returns a value of cudaErrorInvalidValue if cudaEventRecord has not been called for this event. cudaEventSynchronize integer function cudaEventSynchronize( event ) type(cudaEvent). It returns cudaErrorInvalidValue if either event has not yet been recorded. the event is recorded after all preceding operations in all streams are complete. intent(in) :: event cudaEventQuery tests whether an event has been recorded. cudaEventDestroy integer function cudaEventDestroy( event ) type(cudaEvent). stream ) type(cudaEvent). 40 . This function is only valid with events recorded on stream zero. intent() :: start. intent(in) :: event cudaEventDestroy destroys the resources associated with an event object. If stream is zero. start. cudaEventElapsedTime integer function cudaEventElapsedTime( time. intent(out) :: event cudaEventCreate creates an event object and assigns the event identifier to its first argument cudaEventRecord integer function cudaEventRecord( event. and cudaErrorNotReady if it has not. end cudaEventElapsedTime computes the elapsed time between two events (in milliseconds). cudaEventQuery integer function cudaEventQuery( event ) type(cudaEvent).

then it returns 0 as iversion. Version Management Use the functions in this section for version management. cudaGetLastError integer function cudaGetLastError() cudaGetLastError returns the error code that was most recently returned from any runtime call in this host thread. intent(in) :: errcode character*(*) :: cudaGetErrorString cudaGetErrorString returns the message string associated with the given error code. If no driver is installed. cudaRuntimeGetVersion integer function cudaRuntimeGetVersion(iversion) integer :: iversion cudaRuntimeGetVersion returns the version number of the installed CUDA Runtime as iversion. cudaGetErrorString function cudaGetErrorString( errcode ) integer. 41 . Runtime APIs Error Handling Use the functions in this section for error handling. This function automatically returns cudaErrorInvalidValue if the iversion argument is NULL.Chapter 4. cudaDriverGetVersion integer function cudaDriverGetVersion(iversion) integer :: iversion cudaDriverGetVersion returns the version number of the installed CUDA driver as iversion. This function automatically returns cudaErrorInvalidValue if the iversion argument is NULL.

42 .

M). The program computes the products by accumulating submatrix products. Each element of the result is the product of one row of A by one column of B.L) integer.com web page at: www.16) ! the value of C(i. M. k. The program caches the submatrices of A and B in the fast shared memory. it reads a block submatrix of A and a block submatrix of B. ty ! submatrices stored in shared memory real.16). tx.L). Bsub(16. For simplicity. The submatrix size is chosen so the number of threads in a block is a multiple of the warp size (32) and is less than the maximum number of threads per thread block (512). the program assumes the matrix sizes are a multiple of 16. shared :: Asub(16. B(M. then moves to the next submatrix of A rowwise and of B columnwise. C(N. L ) real :: A(N.pgroup. as follows: • Each thread block computes one 16x16 submatrix of C. B is MxL. Examples This chapter contains For up-to-date information about the state of the current release. L integer :: i.com/support/index. accumulates the submatrix product. N. • Each thread within the block computes one element of the submatrix. and has not been highly optimized for execution time. B. C. Source Code Listing ! start the module containing the matmul kernel module mmul_mod use cudafor contains ! mmul_kernel computes A*B into C where ! A is NxM.j) being computed 43 . C is then NxL attributes(global) subroutine mmul_kernel( A. visit the frequently asked questions (FAQ) section of the pgroup.Chapter 5.htm Matrix Multiplication Example This example shows a program to compute the product C of two matrices A and B. kb. j. M. value :: N.

M). C ! allocatable device arrays real. Cdev. device. Bdev. 16. Cdev ) end subroutine mmul end module mmul_mod N 44 .:) = B(1:M. Bdev. B. dimBlock ! Get the array sizes N = size( A. allocatable.j) ! Wait until all elements are filled call syncthreads() ! Multiply the two submatrices ! Each of the 16x16 threads accumulates the ! dot product for its element of C(i. 16 ! Fill the submatrices ! Each of the 16x16 threads in the thread block ! loads one element of Asub and Bsub Asub(tx. Cdev(N.ty) enddo ! Synchronize to make sure all threads are done ! reading the submatrices before overwriting them ! in the next iteration of the kb loop call syncthreads() enddo ! Each of the 16x16 threads stores its element ! to the global C array C(i.:) * B(:.j)) i = (blockidx%x-1) * 16 + tx j = (blockidx%y-1) * 16 + ty Cij = 0. M.kb+ty-1) Bsub(tx.:) :: A. the block size do kb = 1.ty) = A(i.0 ! Do the k loop in chunks of 16.:) :: Adev.L).dimBlock>>>( Adev.1:M) Bdev(:. L/16.L) ) ! Copy A and B to the device Adev = A(1:N. B.16 Cij = Cij + Asub(tx.Matrix Multiplication Example real :: Cij ! Get the thread indices tx = threadidx%x ty = threadidx%y ! This thread computes C(i. 2 ) ! Allocate the device arrays allocate( Adev(N.k) * Bsub(k. 1 ) dimBlock = dim3( 16. 1 ) call mmul_kernel<<<dimGrid.Bdev. ! Copy the results back and free up memory C(1:N.1:L) = Cdev deallocate( Adev. dimension(:.j) = Cij end subroutine mmul_kernel ! The host routine to drive the matrix multiplication subroutine mmul( A. 2 ) L = size( B.1:L) ! Create the grid and block dimensions dimGrid = dim3( N/16. C ) real.j) do k = 1. dimension(:. Bdev(M. 1 ) M = size( A.ty) = B(kb+tx-1.Cdev ! dim3 variables to define the grid and block shapes type(dim3) :: dimGrid.j) = sum(A(i.

C. C. • It accumulates the dot product of its row and column of the submatrices. • It steps through the arrays A and B in blocks of size 16. • Finally. The host subroutine mmul is a wrapper for the kernel routine mmul_kernel. depending on the block and thread index. Because the data copy operations are synchronous. • It fills dimGrid and dimBlock to hold the grid and thread block sizes. MMUL_KERNEL This kernel subroutine has two device memory input arrays. A and B. for which element of C(i. and three scalars giving the array sizes. It performs the following operations: • It determines the thread indices for this thread.j) for a particular value of i and j. • It allocates device memory arrays Adev. 45 .Chapter 5. it stores the computed value into the correct element of C. • It initializes a scalar in which it will accumulate the dot product. This routine computes the dot product of A(i. and Cdev. The thread executing this routine is one of 16x16 threads cooperating in a thread block. and L. it does the following steps: • It loads one element of the submatrices of A and B into shared memory. MMUL This host subroutine has two input arrays.:)*B(:. passed as assumed-shape arrays. Bdev. The routine performs the following operations: • It determines the size of the matrices in N. • It determines the i and j indices. A and B. no extra synchronization is needed between the copy operations and the kernel launch. • It calls mmul_kernel to compute Cdev on the device. • It synchronizes again to make sure all threads are done reading the submatrices before starting the next block. • It copies Cdev back from device memory to C. Examples Source Code Description This source code module mmul_mod has two subroutines. • For each block. one device memory output array. • It synchronizes to make sure both submatrices are loaded by all threads in the block.j) it is computing. • It frees the device memory arrays. and one output array. • It copies the arrays A and B to Adev and Bdev using array assignments. M.

46 .

Two Centerpointe Drive Lake Oswego.com/resources/docs. 47 .pgroup. Log in to the PGI website to access the forum: www.pgroup. Phone support is not currently available.com/userforum/index. Many questions and problems can be resolved at our frequently asked questions (FAQ) site at www.Chapter 6.htm. The forum newsgroups may contain answers to commonly asked questions.com All technical support is by email or submissions using an online form at www.com www. Inc. PGI documentation is available at www.com/support/faq. OR 97035 USA The PGI User Forum is monitored by members of the PGI engineering and support teams as well as other PGI customers.htm or in your local copy of the documentation in the release directory doc/index.pgroup.htm.pgroup.com/support. Contact Information You can contact The Portland Group at: The Portland Group STMicroelectronics.php Or contact us electronically using any of the following means: Fax Sales Support WWW +1-503-682-2637 sales@pgroup.com trs@pgroup.pgroup.

48 .

Sign up to vote on this title
UsefulNot useful