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

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

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

viii .

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

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

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

2 .

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

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

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

an allocatable array. from when it is allocated until it is deallocated. If declared in a module. if the allocation in page-locked memory fails. Some operating systems or installations may restrict the use. it will be allocated in host pagelocked memory. or by other device subprograms to which it is passed as an argument.4. It can be read or written by all threads in the block. and has a lifetime of the thread block. though a write in one thread is only guaranteed to be visible to other threads after the next call to the SYNCTHREADS() intrinsic. Device constant variables are allocated in the device constant memory space. A pinned variable must be an allocatable array.Datatypes in Device Subprograms • If declared in a host subprogram. Device constant data may not be assigned or modified in any device subprogram. or an assumed-shape dummy array.4. or size of page-locked memory.1.8 equivalent to real(kind=8) 6 . and may only be accessed within that subprogram. Other device variables have a lifetime of the entire application. Attributes(pinned) A variable with the pinned attributes is called a pinned variable. A device array may be an explicit-shape array. 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. An allocatable device variable has a dynamic lifetime. the variable may be accessed by that subprogram or subprograms contained in that subprogram. the variable will be allocated in the normal host paged memory. Attributes(shared) A variable with the shared attributed is called a device shared variable or a shared variable.8 1. the variable may be accessed by any subprogram in that module and by any subprogram that uses the module.8 4. When a pinned variable is allocated. and have a lifetime of the entire application.2. A shared variable may only be declared in a device subprogram. A shared variable is allocated in the device shared memory for a thread block. Attributes(constant) A variable with the constant attributes is called a device constant variable. Device constant variables may not be allocatable. A shared variable may not be data initialized. but may be modified in host subprograms. Datatypes in Device Subprograms The following intrinsic datatypes are allowed in device subprograms and device data: Table 2.2. availability. Intrinsic Datatypes Type integer logical real double precision Type Kind 1.

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

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

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

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

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

12 .

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

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

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

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

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

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

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

and the associated stream. or from one device array to another.1).” on page 32.1. and their performance is dependent on the size of the slice. block.1. stream >>> • grid is an integer. The execution configuration is specified after the subroutine name in the call statement. the number of threads per thread block is block%x*block%y*block%z. which must be less than the maximum supported by the device. device to host. bytes. as well as the amount of shared memory to use for assumed-size shared memory arrays. Fortran interfaces to the CUDA memory management runtime routines are provided. The cudaMemcpy function can be used to copy data between the host and the GPU: real. Invoking a kernel subroutine A call to a kernel subroutine must give the execution configuration for the call. it has the form: <<< grid. and the implementation. cur. 20 . not bytes. or both: c = adev + bdev adev = adev + a b = sqrt(adev) Elemental transfers are supported by the language but will perform poorly. Counts expressed in arguments to the Fortran runtime routines are expressed in terms of data type elements. If block is an integer. The execution configuration gives the size and shape of the grid and thread blocks that execute the function. For a complete list of memory management runtime routines. Array slices are also supported. the value of grid%z must be one. If grid is an integer. it is converted to dim3(grid. 512) For those familiar with the CUDA C routines. The product grid%x*grid%y gives the number of thread blocks to launch. bl • block is an integer. or of type(dim3). Data Transfer Using Runtime Routines For programmers comfortable with the CUDA C programming environment. These functions can transfer data either from the host to device.1).Invoking a kernel subroutine The following expressions are not legal as they either promote a false impression of where the actual computation occurs. the kind parameter to the Memcpy routines is optional in Fortran since the attributes of the arrays are explicitly declared. If it is type(dim3). it is converted to dim3(block. refer to “Memory Management. If it is type(dim3). device :: wrk(1024) real cur(512) istat = cudaMemcpy(wrk. or of type(dim3). amount of contiguous data in the slices. or would be more efficient written in another way.

• stream is optional. if present. Device code Datatypes allowed Variables and arrays with the device.4(default).1. or a value returned by a call to cudaStreamCreate.5 on page 41. are limited to the types described in this section. or if bytes is greater than the shared memory available. For more information.y.z end type 21 .4(default).8 1 (default) Additionally.” on page 16.8 equivalent to real(kind=8) 4(default).2. See Section 4. They may have any of the intrinsic datatypes in the following table. they may be of derived type. it must be an integer. If not specified. refer to “Shared data. and specifies the number of bytes of shared memory to be allocated for each thread block to use for assumed-size shared memory arrays.2. Table 3. defined as type(dim3) integer(kind=4) :: x. where the members of the derived type have one of the allowed intrinsic datatypes. Device Code Intrinsic Datatypes Type integer logical real double precision complex character(len=1) Type Kind 1. allowing for static shared memory declared in the kernel and for other dedicated uses. it must be a scalar integer. constant. For instance.8 4(default). or another allowed derived type.8 1.Chapter 3. if present. the value zero is used. and have a value of zero. It specifies the stream to which this call is enqueued. a kernel subroutine attributes(global) subroutine sub( a ) can be called like: call sub <<< DG. or declared in device subprograms. The system module cudafor includes definitions of the derived type dim3. or shared attributes. DB >>> ( A ) The function call will fail if the grid or block arguments are greater than the maximum sizes allowed. Reference • bytes is optional. such as the function arguments and execution configuration arguments.

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

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

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

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

The value of the function is .2 and higher. THREADFENCE_SYSTEM void threadfence_system() threadfence_system acts as a memory fence. GPU_TIME The gpu_time intrinsic returns the value of the clock cycle counter on the GPU.true.true. Each of these functions has a single argument. for all threads in the warp. 26 . The clock frequency can be determined by calling cudaGetDeviceProperties.Device code threadfence() is one method to enforce a specific order. only if the value of the argument is .” on page 32. ALLTHREADS The allthreads function is a warp-vote operation with a single scalar logical argument: if( allthreads(a(i)<0.0 or higher.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. 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. 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. as described in “cudaGetDeviceProperties. 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.0) ) allneg = .

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

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

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

30 .

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

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

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. cdesc. height) type(cudaArrayPtr) :: carray type(cudaChannelFormatDesc) :: cdesc integer :: width. cudaMallocPitch may pad the data. The count is in terms of elements. cudaFree integer function cudaFree(devptr) cudaFree deallocates data on the device. Use the functions in this section for memory management. width. should be of benefit. devptr may be any allocatable device array of a supported type specified in Table 3. Or. devptr may be of TYPE(C_DEVPTR). cudaMallocArray integer function cudaMallocArray(carray. Both of these features are subject to change when. devptr may be any allocatable. “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. an allocatable device array as the second argument. height cudaMallocArray allocates a data array on the device. Currently. cudaMallocPitch integer function cudaMallocPitch(devptr. as well as consistency checks between Fortran device arrays and host arrays.” on page 21. Similarly.1. and in effect transfer the allocation to the Fortran array.Chapter 4.” on page 21. and the padded width is returned in the variable pitch. one-dimensional device array of a supported type specified in Table 3.1. cudaMalloc integer function cudaMalloc(devptr. height) cudaMallocPitch allocates data on the device. The height is an integer. proper Fortran pointers for device data are supported. devptr may be of TYPE(C_DEVPTR). count) cudaMalloc allocates data on the device. Under CUDA Fortran. pitch. devptr may also be of TYPE(C_DEVPTR). a shape as the third argument. devptr may be any allocatable. Or.” on page 21.1. width. two-dimensional device array of a supported type specified in Table 3. in which case the integer values are expressed in bytes. The width is in terms of number of elements. Runtime APIs TYPE(C_DEVPTR). in the future. c_f_pointer will take a TYPE(C_DEVPTR) as the first argument. “Device Code Intrinsic Datatypes. cudaFreeArray integer function cudaFreeArray(carray) type(cudaArrayPtr) :: carray 33 . “Device Code Intrinsic Datatypes. in which case the count is in bytes.

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

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

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

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

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

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

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

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

42 .

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

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

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

46 .

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

48 .

Sign up to vote on this title
UsefulNot useful