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

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

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

viii .

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

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

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

2 .

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

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

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

A shared variable is allocated in the device shared memory for a thread block. An allocatable device variable has a dynamic lifetime. Attributes(pinned) A variable with the pinned attributes is called a pinned variable.8 4. and may only be accessed within that subprogram. Device constant variables may not be allocatable. the variable may be accessed by any subprogram in that module and by any subprogram that uses the module. from when it is allocated until it is deallocated.4.2.1. Other device variables have a lifetime of the entire application. the variable will be allocated in the normal host paged memory. Datatypes in Device Subprograms The following intrinsic datatypes are allowed in device subprograms and device data: Table 2. an allocatable array. it will be allocated in host pagelocked memory. Intrinsic Datatypes Type integer logical real double precision Type Kind 1. When a pinned variable is allocated. A shared variable may only be declared in a device subprogram. but may be modified in host subprograms. or by other device subprograms to which it is passed as an argument. Device constant variables are allocated in the device constant memory space. A pinned variable must be an allocatable array. availability. or an assumed-shape dummy array. Device constant data may not be assigned or modified in any device subprogram. If declared in a module. though a write in one thread is only guaranteed to be visible to other threads after the next call to the SYNCTHREADS() intrinsic. A device array may be an explicit-shape array. or size of page-locked memory. and have a lifetime of the entire application.4. Some operating systems or installations may restrict the use. It can be read or written by all threads in the block. the variable may be accessed by that subprogram or subprograms contained in that subprogram.Datatypes in Device Subprograms • If declared in a host subprogram. A shared variable may not be data initialized. if the allocation in page-locked memory fails.8 equivalent to real(kind=8) 6 . Attributes(constant) A variable with the constant attributes is called a device constant variable. 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.2. Attributes(shared) A variable with the shared attributed is called a device shared variable or a shared variable. and has a lifetime of the thread block.8 1.

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

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

j) end do end do 9 . The inner i dimension is run sequentially on each thread. that is.j) +c(i.j) +c(i. n a(i. the directive defines a two-dimensional thread block of size 32x4. block >>> The compiler maps the launch configuration specified by the grid and block values onto the outermost n loops. m do i = 1. In this case. (256.j) +c(i. Programming Guide Syntax The general form of the kernel directive is: !$cuf kernel do[(n)] <<< grid. ThreadIdx%y runs from 1 to 4 and is mapped onto the outer j loop. using asterisks tells the compiler to choose a thread block shape and/or compute the grid shape from the thread block shape and the loop limits.*). The grid shape.Chapter 2. only the outer j loop is run in parallel with a thread block size of 256. m do i = 1. Example 2 !$cuf kernel do <<< *. !$cuf kernel do(2) <<< *. m do i = 1. The grid and block values can be an integer scalar or a parenthesized list. The body of the doubly-nested loop is turned in the kernel body: ThreadIdx%x runs from 1 to 32 and is mapped onto the inner i loop. Loops which are not mapped onto the grid and block values are run sequentially on each thread.j) = b(i. (32.4) >>> do j = 1. You might consider if the code in Example 2 would perform better if the two loops were interchanged. the default value is 1. specified as (*. n a(i. Alternatively. starting at loop n and working out. n a(i. Alternatively. 256 >>> do j = 1.j) end do end do Without an explicit n on the do. you could specify a configuration like the following in which the threads read and write the matrices in coalesced fashion.j) end do end do In this example.j) = b(i. so all iterations are computed. is computed by the compiler and runtime by dividing the loop trip counts n and m by the thread block size. Example 1 !$cuf kernel do(2) <<< (*. the schedule applies just to the outermost loop.1) >>> do j = 1.j) = b(i.*).

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

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

12 .

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

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

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

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

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

Allocating Device and Pinned Arrays real. 100) … istat = cudaFree(v) For a complete list of the memory management runtime routines.stat=istat) … if(allocated(b)) deallocate(b) Scalar variables can be allocated on the device using the Fortran 2003 allocatable scalar feature. These arrays will also have a lifetime of the subprogram as is usual with the Fortran language: subroutine vfunc(a. refer to“Memory Management. Fortran interfaces to the CUDA memory management runtime routines are provided. pinned allocatable arrays work and act like any other allocatable array on the host. pinned :: p(:) allocate(p(5000). allocatable. allocatable. or if it is exhausted. device :: adev(n) real. These functions return memory which will bypass certain Fortran allocatable properties such as automatic deallocation.c. If no such memory space is available.n) real. the compiler allocates the array in normal paged host memory. device :: b(:) allocate(b(5024). 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. To use these.” on page 32. and thus the arrays are treated more like C malloc’ed areas. real. 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. The cudaMalloc function can be used to allocate single-dimensional arrays of the supported intrinsic datatypes. allocatable. Mixing standard Fortran allocate/deallocate with the runtime Malloc/Free for a given array is not supported. device :: v(:) istat = cudaMalloc(v. and cudaFree can be used to free it: real. Allocating Pinned Memory Allocatable arrays with the pinned attribute are dynamically allocated using the Allocate statement. Otherwise. The compiler will generate code to allocate the array in host page-locked memory. allocatable. if available.stat=istat) … if(allocated(p)) deallocate(p) 18 . declare and initialize the scalar on the host as: integer.

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

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

z end type 21 . and have a value of zero. DB >>> ( A ) The function call will fail if the grid or block arguments are greater than the maximum sizes allowed. the value zero is used. If not specified. it must be an integer. such as the function arguments and execution configuration arguments. Device code Datatypes allowed Variables and arrays with the device.Chapter 3.y. are limited to the types described in this section. or shared attributes.8 1 (default) Additionally. constant. allowing for static shared memory declared in the kernel and for other dedicated uses. See Section 4. Reference • bytes is optional.8 1.2. they may be of derived type.1.5 on page 41.2. or declared in device subprograms. For more information. it must be a scalar integer.8 4(default). or a value returned by a call to cudaStreamCreate. Device Code Intrinsic Datatypes Type integer logical real double precision complex character(len=1) Type Kind 1. refer to “Shared data.8 equivalent to real(kind=8) 4(default).4(default). Table 3. The system module cudafor includes definitions of the derived type dim3.” on page 16. where the members of the derived type have one of the allowed intrinsic datatypes. • stream is optional. if present. or another allowed derived type.4(default). For instance. and specifies the number of bytes of shared memory to be allocated for each thread block to use for assumed-size shared memory arrays. or if bytes is greater than the shared memory available. a kernel subroutine attributes(global) subroutine sub( a ) can be called like: call sub <<< DG. It specifies the stream to which this call is enqueued. defined as type(dim3) integer(kind=4) :: x. They may have any of the intrinsic datatypes in the following table. if present.

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

complex integer (integer. complex integer. complex real real.3. real real real Name precision radix range selected_int_kind selected_real_kind tiny Argument Datatypes real. complex real 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 . real real integer.integer) real Fortran Bit Manipulation Intrinsics Table 3.5. Fortran Mathematical Intrinsics Name acos asin atan atan2 cos cosh exp Argument Datatypes real real real (real.real) real. complex real real. Fortran Numeric Inquiry Intrinsics Name bit_size digits epsilon huge maxexponent minexponent Argument Datatypes integer integer. complex Name log log10 sin sinh sqrt tan tanh Argument Datatypes real. real.Chapter 3. complex real real Fortran Numeric Inquiry Intrinsics Table 3.4. real integer. Reference Fortran Mathematical Intrinsics Table 3.

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

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

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

and store the combined value back to the first argument location. Compute capability 1.1 and higher. value ) Additional Atomic Update mem = mem + value mem = mem – value 27 . Reference ANYTHREAD The anythread function is a warp-vote operation with a single scalar logical argument: if( anythread(a(i)<0. Example: if( ballot(int_value) ) allneg = . depending on the function.true.8.0) ) allneg = .0. Atomic Functions The atomic functions read and write the value of their first operand. Note The return value for each of these functions is the first argument. which must be a variable or array element in shared memory (with the shared attribute) or in device global memory (with the device 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. Atomic functions are only supported by devices with compute capability 1. 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. value ) atomicsub( mem.2 or higher is required if the first argument has the shared attribute. for all threads in the warp. These functions are: Table 3. They also combine that value with the value of the second argument.Chapter 3.false.true. mem. only if the value of the argument is . Both arguments must be of type integer(kind=4). This function is only supported by devices of compute capability 2. The value of the function is . Arithmetic and Bitwise Atomic Functions Function atomicadd( mem.false. The function anythread evaluates its argument for all threads in the current warp. Arithmetic and Bitwise Atomic Functions These atomic functions read and return the value of the first argument.

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

• ENTRY statements are not allowed.3 are not supported. • Optional arguments are not allowed. Host code Host subprograms may use intrinsic functions. • Input/Output statements are not allowed: READ. BACKSPACE. where A is a variable or expression. • Floating point exception handling is not supported. • Fortran intrinsic functions not listed in Section 3. • STOP and PAUSE statements are not allowed. INQUIRE. PRINT. NAMELIST. such as the new sizeof intrinsic function. • Alternate return specifications are not allowed. CLOSE. OPEN. character substrings are not supported. WRITE.10. • Automatic arrays must be fixed size. • Recursive subroutines and functions are not allowed. • Subroutine and function calls are supported only if they can be inlined.comp. ENDFILE. Compare and Swap Atomic Function Function atomiccas(mem. • Objects with the Pointer and Allocatable attribute are not allowed.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. returns the number of bytes required to hold the value of A. • Objects with character type must have LEN=1. SIZEOF Intrinsic A call to sizeof(A). integer(kind=4) :: i. REWIND. FORMAT. • Cray pointers are not supported.Chapter 3.6. j j = sizeof(i) ! this assigns the value 4 to j 29 . Reference Table 3.

30 .

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

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

Both of these features are subject to change when. in which case the count is in bytes. cudaFree integer function cudaFree(devptr) cudaFree deallocates data on the device. cudaMallocPitch may pad the data. proper Fortran pointers for device data are supported. The height is an integer. two-dimensional device array of a supported type specified in Table 3.1. cudaFreeArray integer function cudaFreeArray(carray) type(cudaArrayPtr) :: carray 33 . devptr may be any allocatable device array of a supported type specified in Table 3. 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. The count is in terms of elements.” on page 21. a shape as the third argument. devptr may be of TYPE(C_DEVPTR). as well as consistency checks between Fortran device arrays and host arrays. Similarly. in which case the integer values are expressed in bytes.1.1.Chapter 4. “Device Code Intrinsic Datatypes. height cudaMallocArray allocates a data array on the device. devptr may be of TYPE(C_DEVPTR). width. “Device Code Intrinsic Datatypes. count) cudaMalloc allocates data on the device.” on page 21. devptr may be any allocatable. Or. and the padded width is returned in the variable pitch. should be of benefit. cdesc. height) type(cudaArrayPtr) :: carray type(cudaChannelFormatDesc) :: cdesc integer :: width. and in effect transfer the allocation to the Fortran array. Currently. c_f_pointer will take a TYPE(C_DEVPTR) as the first argument. pitch. cudaMallocPitch integer function cudaMallocPitch(devptr. Under CUDA Fortran. Or.” on page 21. height) cudaMallocPitch allocates data on the device. width. “Device Code Intrinsic Datatypes. 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. Runtime APIs TYPE(C_DEVPTR). an allocatable device array as the second argument. Use the functions in this section for memory management. devptr may also be of TYPE(C_DEVPTR). The width is in terms of number of elements. one-dimensional device array of a supported type specified in Table 3. cudaMallocArray integer function cudaMallocArray(carray. cudaMalloc integer function cudaMalloc(devptr. in the future. devptr may be any allocatable.

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

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

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

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

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

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

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

This function automatically returns cudaErrorInvalidValue if the iversion argument is NULL. intent(in) :: errcode character*(*) :: cudaGetErrorString cudaGetErrorString returns the message string associated with the given error code.Chapter 4. cudaGetLastError integer function cudaGetLastError() cudaGetLastError returns the error code that was most recently returned from any runtime call in this host thread. then it returns 0 as iversion. Version Management Use the functions in this section for version management. 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. 41 . cudaGetErrorString function cudaGetErrorString( errcode ) integer. This function automatically returns cudaErrorInvalidValue if the iversion argument is NULL. cudaRuntimeGetVersion integer function cudaRuntimeGetVersion(iversion) integer :: iversion cudaRuntimeGetVersion returns the version number of the installed CUDA Runtime as iversion.

42 .

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

M).k) * Bsub(k.j) = Cij end subroutine mmul_kernel ! The host routine to drive the matrix multiplication subroutine mmul( A. Bdev. B.Cdev ! dim3 variables to define the grid and block shapes type(dim3) :: dimGrid. the block size do kb = 1. C ) real. dimension(:.16 Cij = Cij + Asub(tx. B.kb+ty-1) Bsub(tx. 1 ) call mmul_kernel<<<dimGrid.j) do k = 1. Bdev. dimBlock ! Get the array sizes N = size( A. 2 ) L = size( B.:) * B(:.ty) = B(kb+tx-1.j)) i = (blockidx%x-1) * 16 + tx j = (blockidx%y-1) * 16 + ty Cij = 0.1:M) Bdev(:.1:L) ! Create the grid and block dimensions dimGrid = dim3( N/16. 1 ) M = size( A.1:L) = Cdev deallocate( Adev. 16.L). allocatable. Cdev ) end subroutine mmul end module mmul_mod N 44 .dimBlock>>>( Adev.0 ! Do the k loop in chunks of 16. L/16. dimension(:. Cdev. 1 ) dimBlock = dim3( 16. ! Copy the results back and free up memory C(1:N.j) ! Wait until all elements are filled call syncthreads() ! Multiply the two submatrices ! Each of the 16x16 threads accumulates the ! dot product for its element of C(i. 16 ! Fill the submatrices ! Each of the 16x16 threads in the thread block ! loads one element of Asub and Bsub Asub(tx.Matrix Multiplication Example real :: Cij ! Get the thread indices tx = threadidx%x ty = threadidx%y ! This thread computes C(i. Bdev(M.:) = B(1:M. device.j) = sum(A(i. 2 ) ! Allocate the device arrays allocate( Adev(N.:) :: Adev. M.L) ) ! Copy A and B to the device Adev = A(1:N. C ! allocatable device arrays real.ty) = A(i. Cdev(N.:) :: A.Bdev.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.

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

46 .

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

48 .

Sign up to vote on this title
UsefulNot useful