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

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

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

viii .

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

x .pgroup. Zero or more of the preceding item may occur. FORTRAN Fortran language statements are shown in the text of this guide using a reduced fixed point size.Terminology Constant Width is used for filenames. 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. Geneva. 2009. NVIDIA. In this example. square brackets are required to specify a p/t-set. ellipsis indicate a repetition. Information Technology – Programming Languages – FORTRAN. June 2010.com/doc/pgiug. multiple filenames are allowed. Available online at http://www. filename . • NVIDIA CUDA Programming Guide.htm Related Publications The following documents contain additional information related to CUDA Fortran programming. • NVIDIA CUDA Compute Unified Device Architecture Reference Manual. Available online at http:// www. directories. examples. In this case. The Portland Group. 7/21/2010.pdf. Version 3. PGI provides a glossary of terms which you can access at www.pgroup. square brackets indicate optional items. Available online at http:// www. • ISO/IEC 1539-1:1997. you must select either item2 or item3.nvidia. In this case item1 is optional. and for language statements in the text. In the context of p/tsets.1.nvidia. [ item1 ] in general.. Terminology If there are terms in this guide with which you are unfamiliar. arguments. including assembly language statements. 1997 (Fortran 95).9. Release 10. • PGI Users Guide.com/cuda. { item2 | item 3} braces indicate that a selection is required. September. Version 3.1. options..com/cuda. NVIDIA. There are a wide variety of releases and distributions of each of these types of operating systems. Bold is used for commands.1. C/C++ C/C++ language statements are shown in the test of this guide using a reduced fixed point size.com/support/definitions.

Compared to the PGI Accelerator directives-based model and compilers.Chapter 1. Each block is partitioned into fine grain threads. but are general enough to be useful in many data-parallel. A properly designed CUDA program will run on any CUDA-enabled GPU. NVIDIA introduced CUDA™. PGI 2011 includes support for CUDA Fortran on Linux. highly parallel computational units with very high memory bandwidth. shared memory and barrier synchronization within thread groups. a small set of extensions to Fortran that supports and is built upon the CUDA computing architecture. 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. here called CUDA C. and tremendous potential for many applications. 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. and coordinated independent thread groups organized into a grid. The programming model supports four key abstractions: cooperating threads organized into thread groups. GPU designs are optimized for the computations found in graphics rendering. Introduction Welcome to Release 2011 of PGI CUDA Fortran. allowing direct programming of the GPU from a high level language. regardless of the number of available processor cores. Mac OS X and Windows. 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 . A CUDA programmer must partition the program into coarse grain blocks that can be executed in parallel. Graphic processing units or GPUs have evolved into programmable. CUDA Fortran is an analog to NVIDIA's CUDA C compiler. a general purpose parallel programming architecture. with compilers and libraries to support the programming of NVIDIA GPUs. compute-intensive programs. which can cooperate using shared memory and barrier synchronization. CUDA comes with an extended C compiler. CUDA Fortran includes a Fortran 2003 compiler and tool chain for programming NVIDIA GPUs using Fortran.

2 .

device. each with 64 threads. x. 64>>>( 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. y ) real. For a reference for CUDA Fortran. x. value :: n. dimension(*) :: x. y ) real.Chapter 2. each instance will be executed by a different CUDA thread. “Reference. 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. 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. value :: a integer.” on page 13. Such a subroutine is called a device kernel or kernel. 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. 3 . each thread performs one iteration of the common SAXPY loop operation. y real :: a integer :: n ! call the kernel call ksaxpy<<<n/64. a. dimension(*) :: x. refer to Chapter 3. x.y real. y ) end subroutine In this case. Programming Guide This chapter introduces the CUDA programming model through examples written in CUDA Fortran. and a local thread index within its thread block. and each thread has a global thread block index. the call to the kernel ksaxpy specifies n/64 thread blocks. and a thread index accessed through threadidx. A kernel is defined using the attributes(global) specifier on the subroutine statement. a. In this example. The CUDA threads are organized into thread blocks. a.

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

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

Attributes(pinned) A variable with the pinned attributes is called a pinned variable. Device constant variables are allocated in the device constant memory space.2. Attributes(constant) A variable with the constant attributes is called a device constant variable. Some operating systems or installations may restrict the use. It can be read or written by all threads in the block. A shared variable may only be declared in a device subprogram.4. the variable will be allocated in the normal host paged memory. and have a lifetime of the entire application. if the allocation in page-locked memory fails. A device array may be an explicit-shape array. an allocatable array.8 1. the variable may be accessed by that subprogram or subprograms contained in that subprogram. Other device variables have a lifetime of the entire application. availability. from when it is allocated until it is deallocated.2.4. or by other device subprograms to which it is passed as an argument. Intrinsic Datatypes Type integer logical real double precision Type Kind 1. Device constant variables may not be allocatable. Datatypes in Device Subprograms The following intrinsic datatypes are allowed in device subprograms and device data: Table 2. or an assumed-shape dummy array. but may be modified in host subprograms. or size of page-locked memory. and may only be accessed within that subprogram. and has a lifetime of the thread block. though a write in one thread is only guaranteed to be visible to other threads after the next call to the SYNCTHREADS() intrinsic. the variable may be accessed by any subprogram in that module and by any subprogram that uses the module. 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. Device constant data may not be assigned or modified in any device subprogram.8 equivalent to real(kind=8) 6 . When a pinned variable is allocated.1. If declared in a module. Attributes(shared) A variable with the shared attributed is called a device shared variable or a shared variable. A pinned variable must be an allocatable array.8 4. A shared variable is allocated in the device shared memory for a thread block. An allocatable device variable has a dynamic lifetime.Datatypes in Device Subprograms • If declared in a host subprogram. it will be allocated in host pagelocked memory. A shared variable may not be data initialized.

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

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

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

• The body of the loops may contain assignment statements. • Subroutine and function calls to attributes(device) subprograms are allowed if they are in the same module as the code containing the directive. IF statements. (256. gridDim%x is equal to m. To "unroll" more work into each thread. atomic functions. when a reduction has a scalar target. In CUDA built-in language. One thread block is created for each j. Specifically. 10 . Restrictions on the CUF kernel directive The following restrictions apply to CUF kernel directives: • If the directive specifies n dimensions. one for each thread. as illustrated here: !$cuf kernel do(2) <<< (1. • CUDA Fortran intrinsic functions are allowed. etc. specify non-asterisk values for the grid. • Arrays used or assigned in the loop must have the device attribute. m do i = 1. the 256 threads in each block each do one element of the matrix addition. 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. are not. Except in the case of reductions.*). and increment must be invariant with respect to any other loop in the kernel do. upper limit. because the compiler skips over the constant 1 in the i loop grid size.Kernel Loop Directive Example 3 In Example 2. • The tightly-nested DO loops must have invariant loop limits: the lower limit. in concert. the compiler generates a correct sequence of synchronized operations to produce one copy either in device global memory or on the host. and GOTO 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. it must be followed by at least that many tightly-nested DO loops.j) = b(i. • Implicit loops and F90 array syntax are not allowed within the directive loops. or the compiler will make a device copy of that variable live for the duration of the loops. n a(i. loops. but the device-specific intrinsics such as syncthreads. • Scalars used or assigned in the loop must either have the device attribute. if they are allowed in device code. the j loop is mapped onto the grid x-dimension. • Only CUDA Fortran dataypes are allowed within the loops.1) >>> do j = 1. incrementing by 256.j) end do end do Now the threads in a thread block handle all values in the i direction.j) +c(i. 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.

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

12 .

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

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

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

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

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

allocatable. If no such memory space is available. 100) … istat = cudaFree(v) For a complete list of the memory management runtime routines. device :: adev(n) 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. These arrays will also have a lifetime of the subprogram as is usual with the Fortran language: subroutine vfunc(a. real. the compiler allocates the array in normal paged host memory.Allocating Device and Pinned Arrays real. and thus the arrays are treated more like C malloc’ed areas.c.” on page 32. or if it is exhausted. To use these. 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. The cudaMalloc function can be used to allocate single-dimensional arrays of the supported intrinsic datatypes. Otherwise. allocatable. pinned :: p(:) allocate(p(5000). if available. device :: v(:) istat = cudaMalloc(v. declare and initialize the scalar on the host as: integer.n) real.stat=istat) … if(allocated(p)) deallocate(p) 18 . 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. 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. The compiler will generate code to allocate the array in host page-locked memory. pinned allocatable arrays work and act like any other allocatable array on the host. and cudaFree can be used to free it: real. allocatable. Allocating Pinned Memory Allocatable arrays with the pinned attribute are dynamically allocated using the Allocate statement. refer to“Memory Management. allocatable. Mixing standard Fortran allocate/deallocate with the runtime Malloc/Free for a given array is not supported.

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

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

• stream is optional. if present.5 on page 41. or shared attributes. the value zero is used. They may have any of the intrinsic datatypes in the following table.y.4(default). a kernel subroutine attributes(global) subroutine sub( a ) can be called like: call sub <<< DG.2. Reference • bytes is optional.2. Device code Datatypes allowed Variables and arrays with the device.4(default).8 1 (default) Additionally. If not specified. such as the function arguments and execution configuration arguments. refer to “Shared data. or another allowed derived type. Device Code Intrinsic Datatypes Type integer logical real double precision complex character(len=1) Type Kind 1. it must be a scalar integer. where the members of the derived type have one of the allowed intrinsic datatypes. or declared in device subprograms. The system module cudafor includes definitions of the derived type dim3.z end type 21 . Table 3.” on page 16.1. if present. DB >>> ( A ) The function call will fail if the grid or block arguments are greater than the maximum sizes allowed. or if bytes is greater than the shared memory available. are limited to the types described in this section.8 1. and have a value of zero. It specifies the stream to which this call is enqueued.Chapter 3. defined as type(dim3) integer(kind=4) :: x. they may be of derived type. allowing for static shared memory declared in the kernel and for other dedicated uses. constant. For instance. or a value returned by a call to cudaStreamCreate.8 equivalent to real(kind=8) 4(default). For more information. and specifies the number of bytes of shared memory to be allocated for each thread block to use for assumed-size shared memory arrays. See Section 4. it must be an integer.8 4(default).

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

integer) real Fortran Bit Manipulation Intrinsics Table 3. complex real real Fortran Numeric Inquiry Intrinsics Table 3. Reference Fortran Mathematical Intrinsics Table 3.Chapter 3.3. complex integer (integer. complex Name log log10 sin sinh sqrt tan tanh Argument Datatypes real. 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 . Fortran Numeric Inquiry Intrinsics Name bit_size digits epsilon huge maxexponent minexponent Argument Datatypes integer integer.5.real) real. complex integer. Fortran Mathematical Intrinsics Name acos asin atan atan2 cos cosh exp Argument Datatypes real real real (real.4. real. real real real Name precision radix range selected_int_kind selected_real_kind tiny Argument Datatypes real. complex real real. real real integer. complex real real. real integer. complex real real.

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

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

26 .Device code threadfence() is one method to enforce a specific order.0 or higher. 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. The value of the function is .true. only if the value of the argument is .true. Warp-Vote Operations Warp-vote operations are only supported by devices with compute capability 1.true. 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. GPU_TIME The gpu_time intrinsic returns the value of the clock cycle counter on the GPU.” on page 32. 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. The clock frequency can be determined by calling cudaGetDeviceProperties.2 and higher. as described in “cudaGetDeviceProperties. Each of these functions has a single argument. 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. 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) ) allneg = . for all threads in the warp. ALLTHREADS The allthreads function is a warp-vote operation with a single scalar logical argument: if( allthreads(a(i)<0.

which must be a variable or array element in shared memory (with the shared attribute) or in device global memory (with the device attribute).8. value ) Additional Atomic Update mem = mem + value mem = mem – value 27 . Compute capability 1.true.false. Reference ANYTHREAD The anythread function is a warp-vote operation with a single scalar logical argument: if( anythread(a(i)<0.Chapter 3. This function is only supported by devices of compute capability 2. Example: if( ballot(int_value) ) allneg = . These functions are: Table 3. Atomic functions are only supported by devices with compute capability 1. and store the combined value back to the first argument location. mem. Both arguments must be of type integer(kind=4). 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.false. only if the value of the argument is . Arithmetic and Bitwise Atomic Functions These atomic functions read and return the value of the first argument. Arithmetic and Bitwise Atomic Functions Function atomicadd( mem.2 or higher is required if the first argument has the shared attribute. 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.0) ) allneg = . for all threads in the warp. The value of the function is . value ) atomicsub( mem. depending on the function. Atomic Functions The atomic functions read and write the value of their first operand.0. Note The return value for each of these functions is the first argument. They also combine that value with the value of the second argument.true. The function anythread evaluates its argument for all threads in the current warp.1 and higher.

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

ENDFILE. where A is a variable or expression. • Objects with character type must have LEN=1. PRINT. • ENTRY statements are not allowed. OPEN.comp. BACKSPACE. • Floating point exception handling is not supported. • Input/Output statements are not allowed: READ. Compare and Swap Atomic Function Function atomiccas(mem.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.10. FORMAT. returns the number of bytes required to hold the value of A. • Alternate return specifications are not allowed. Host code Host subprograms may use intrinsic functions.Chapter 3.3 are not supported. SIZEOF Intrinsic A call to sizeof(A). WRITE. • Objects with the Pointer and Allocatable attribute are not allowed. CLOSE. Reference Table 3. • Cray pointers are not supported. j j = sizeof(i) ! this assigns the value 4 to j 29 . • Fortran intrinsic functions not listed in Section 3. such as the new sizeof intrinsic function. INQUIRE. • STOP and PAUSE statements are not allowed. • Recursive subroutines and functions are not allowed. integer(kind=4) :: i. • Automatic arrays must be fixed size. • Optional arguments are not allowed. REWIND. NAMELIST. character substrings are not supported.6. • Subroutine and function calls are supported only if they can be inlined.

30 .

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

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

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

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

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

which was allocated with cudaMalloc3D to a specified value. 36 . intent(out) :: pitchptr type(cudaExtent). dstx. srca integer :: dstx. cudaMemset3D integer function cudaMemset3D(pitchptr. cudaMalloc3DArray integer function cudaMalloc3DArray(carray. srcx. cext) type(cudaPitchedPtr) :: pitchptr integer :: value type(cudaExtent) :: cext cudaMemset3D sets elements of an array. value. srca. cudaMalloc3D integer function cudaMalloc3D(pitchptr. pitchptr may be any allocatable. dsty. cudaMemcpy2DArrayToArray integer function cudaMemcpy2DArrayToArray(dsta. kdir) type(cudaArrayPtr) :: dsta. pitchptr is a derived type defined in the cudafor module. dsty. three-dimensional device array of a supported type specified in “Datatypes allowed. cext) type(cudaPitchedPtr).Memory Management cudaMemcpyArrayToArray integer function cudaMemcpyArrayToArray(dsta. Alternatively. srcx. srcy. srca integer :: dstx. cdesc. the extents in each dimension specified by cext. srcx. dsty. kdir cudaMemcpyArrayToArray copies array data to and from the device. srca. intent(in) :: cext cudaMalloc3D allocates data on the device. srcy. width. kdir) type(cudaArrayPtr) :: dsta. dstx. count. width. dsty. height.” on page 21. cext) type(cudaArrayPtr) :: carray type(cudaChannelFormatDesc) :: cdesc type(cudaExtent) :: cext cudaMalloc3DArray allocates array data on the device. cext is also a derived type which holds the extents of the allocated array. srcy. srcx. srcy. kdir cudaMemcpy2DArrayToArray copies array data to and from the device. count. height. 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 cudaMemcpyToSymbol copies data from the source to a device area in global or constant memory space referenced by a symbol. The copy can be associated with a stream by passing a non-zero stream argument. cudaMemcpyFromSymbolAsync integer function cudaMemcpyFromSymbolAsync(dst. count. cudaMemcpyToSymbolAsync integer function cudaMemcpyToSymbolAsync(symbol. count. optional :: stream 37 . symbol. count. This function operates on page-locked host memory only. stream) type(cudaSymbol) :: symbol integer :: count. kdir cudaMemcpyFromSymbol copies data from a device area in global or constant memory space referenced by a symbol to a destination on the host. cudaMemcpyFromSymbol integer function cudaMemcpyFromSymbol(dst. offset. offset.Chapter 4. The copy can be associated with a stream by passing a non-zero stream argument. dst may be any host scalar or array of a supported type specified in “Datatypes allowed. offset. offset. optional :: stream cudaMemcpyToSymbol copies data from the source to a device area in global or constant memory space referenced by a symbol. src. stream) type(cudaSymbol) :: symbol integer :: count. offset. kdir. src may be any host scalar or array of a supported type specified in “Datatypes allowed. src may be any host scalar or array of a supported type as specified in “Datatypes allowed. kdir) type(cudaSymbol) :: symbol integer :: count. symbol.” on page 21. The count is in terms of elements. The count is in terms of elements. The count is in terms of elements. offset. 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.” on page 21. kdir) type(cudaSymbol) :: symbol integer :: count. src. kdir. cudaMemcpyToSymbol integer function cudaMemcpyToSymbol(symbol. kdir integer. kdir integer. This function operates on page-locked host memory only. Runtime APIs cudaMemcpy3DAsync integer function cudaMemcpy3D(p. offset. count. offset.” on page 21.

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

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

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

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

shared :: Asub(16.16) ! the value of C(i. M. kb. The program caches the submatrices of A and B in the fast shared memory. 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). Examples This chapter contains For up-to-date information about the state of the current release.L).L) integer.16). accumulates the submatrix product. ty ! submatrices stored in shared memory real. L ) real :: A(N. C(N. The program computes the products by accumulating submatrix products.Chapter 5.com web page at: www. For simplicity. B(M.pgroup. • Each thread within the block computes one element of the submatrix. j. visit the frequently asked questions (FAQ) section of the pgroup. k. Bsub(16.M). as follows: • Each thread block computes one 16x16 submatrix of C. and has not been highly optimized for execution time. C. M. N. it reads a block submatrix of A and a block submatrix of B. 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.com/support/index. tx.j) being computed 43 . then moves to the next submatrix of A rowwise and of B columnwise. B. B is MxL. value :: N. C is then NxL attributes(global) subroutine mmul_kernel( A. L integer :: i. Each element of the result is the product of one row of A by one column of B.htm Matrix Multiplication Example This example shows a program to compute the product C of two matrices A and B. the program assumes the matrix sizes are a multiple of 16.

:) * B(:.dimBlock>>>( Adev.Bdev.L) ) ! Copy A and B to the device Adev = A(1:N. B. dimension(:.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. C ! allocatable device arrays real.L).0 ! Do the k loop in chunks of 16.ty) = A(i. C ) real.j) = sum(A(i.1:L) = Cdev deallocate( Adev.j) do k = 1. Bdev(M.ty) = B(kb+tx-1.:) = B(1:M.k) * Bsub(k.j) = Cij end subroutine mmul_kernel ! The host routine to drive the matrix multiplication subroutine mmul( A.kb+ty-1) Bsub(tx.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. Bdev. B.1:L) ! Create the grid and block dimensions dimGrid = dim3( N/16. dimBlock ! Get the array sizes N = size( A.Cdev ! dim3 variables to define the grid and block shapes type(dim3) :: dimGrid. Bdev.:) :: Adev. the block size do kb = 1.1:M) Bdev(:.16 Cij = Cij + Asub(tx. 16. device. Cdev. 1 ) M = size( A. 2 ) ! Allocate the device arrays allocate( Adev(N. Cdev(N.:) :: A. dimension(:. allocatable. ! Copy the results back and free up memory C(1:N.M). 1 ) call mmul_kernel<<<dimGrid. M. 16 ! Fill the submatrices ! Each of the 16x16 threads in the thread block ! loads one element of Asub and Bsub Asub(tx. Cdev ) end subroutine mmul end module mmul_mod N 44 . L/16.j)) i = (blockidx%x-1) * 16 + tx j = (blockidx%y-1) * 16 + ty Cij = 0.Matrix Multiplication Example real :: Cij ! Get the thread indices tx = threadidx%x ty = threadidx%y ! This thread computes C(i. 2 ) L = size( B. 1 ) dimBlock = dim3( 16.

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

46 .

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

48 .

Sign up to vote on this title
UsefulNot useful