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

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

viii .

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

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

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

2 .

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

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

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

A pinned variable must be an allocatable array.1.4. 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. or by other device subprograms to which it is passed as an argument.2. A shared variable may only be declared in a device subprogram. If declared in a module. and has a lifetime of the thread block. Device constant data may not be assigned or modified in any device subprogram. Intrinsic Datatypes Type integer logical real double precision Type Kind 1. and may only be accessed within that subprogram. An allocatable device variable has a dynamic lifetime.8 1.Datatypes in Device Subprograms • If declared in a host subprogram. Attributes(pinned) A variable with the pinned attributes is called a pinned variable. Attributes(constant) A variable with the constant attributes is called a device constant variable.8 4.8 equivalent to real(kind=8) 6 . Device constant variables are allocated in the device constant memory space. if the allocation in page-locked memory fails.4. A device array may be an explicit-shape array. an allocatable array. and have a lifetime of the entire application. it will be allocated in host pagelocked memory. Other device variables have a lifetime of the entire application. A shared variable may not be data initialized. Device constant variables may not be allocatable. 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. Datatypes in Device Subprograms The following intrinsic datatypes are allowed in device subprograms and device data: Table 2. A shared variable is allocated in the device shared memory for a thread block. Attributes(shared) A variable with the shared attributed is called a device shared variable or a shared variable. but may be modified in host subprograms. When a pinned variable is allocated. or size of page-locked memory. the variable may be accessed by any subprogram in that module and by any subprogram that uses the module. It can be read or written by all threads in the block. though a write in one thread is only guaranteed to be visible to other threads after the next call to the SYNCTHREADS() intrinsic. Some operating systems or installations may restrict the use. availability. or an assumed-shape dummy array.2. from when it is allocated until it is deallocated.

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

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

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

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

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

12 .

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

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

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

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

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

The cudaMalloc function can be used to allocate single-dimensional arrays of the supported intrinsic datatypes. allocatable. Allocating Pinned Memory Allocatable arrays with the pinned attribute are dynamically allocated using the Allocate statement. declare and initialize the scalar on the host as: integer. the compiler allocates the array in normal paged host memory. allocatable. device :: v(:) istat = cudaMalloc(v. These arrays will also have a lifetime of the subprogram as is usual with the Fortran language: subroutine vfunc(a. allocatable. allocatable.n) real.stat=istat) … if(allocated(b)) deallocate(b) Scalar variables can be allocated on the device using the Fortran 2003 allocatable scalar feature. To use these. device :: b(:) allocate(b(5024). device :: adev(n) real. device :: atmp(4) … end subroutine vfunc ! adev and atmp are deallocated Allocating Device Memory Using Runtime Routines For programmers comfortable with the CUDA C programming environment.stat=istat) … if(allocated(p)) deallocate(p) 18 . The compiler will generate code to allocate the array in host page-locked memory. Mixing standard Fortran allocate/deallocate with the runtime Malloc/Free for a given array is not supported.Allocating Device and Pinned Arrays real. and thus the arrays are treated more like C malloc’ed areas. and cudaFree can be used to free it: real. Otherwise. These functions return memory which will bypass certain Fortran allocatable properties such as automatic deallocation. pinned allocatable arrays work and act like any other allocatable array on the host. Fortran interfaces to the CUDA memory management runtime routines are provided. 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. 100) … istat = cudaFree(v) For a complete list of the memory management runtime routines. or if it is exhausted. If no such memory space is available. if available. real.c.” on page 32. pinned :: p(:) allocate(p(5000). refer to“Memory Management.

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

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

y. Device code Datatypes allowed Variables and arrays with the device.8 1 (default) Additionally. if present.4(default). The system module cudafor includes definitions of the derived type dim3.1. such as the function arguments and execution configuration arguments. defined as type(dim3) integer(kind=4) :: x. a kernel subroutine attributes(global) subroutine sub( a ) can be called like: call sub <<< DG. are limited to the types described in this section.8 4(default). If not specified.” on page 16.z end type 21 .2. where the members of the derived type have one of the allowed intrinsic datatypes. constant. or a value returned by a call to cudaStreamCreate.2.Chapter 3. the value zero is used. Reference • bytes is optional.5 on page 41. and specifies the number of bytes of shared memory to be allocated for each thread block to use for assumed-size shared memory arrays. Table 3. or if bytes is greater than the shared memory available. For instance. See Section 4. or declared in device subprograms.8 equivalent to real(kind=8) 4(default). they may be of derived type. Device Code Intrinsic Datatypes Type integer logical real double precision complex character(len=1) Type Kind 1. or another allowed derived type. if present. refer to “Shared data. For more information. It specifies the stream to which this call is enqueued. DB >>> ( A ) The function call will fail if the grid or block arguments are greater than the maximum sizes allowed. • stream is optional. and have a value of zero.4(default). They may have any of the intrinsic datatypes in the following table.8 1. it must be an integer. or shared attributes. it must be a scalar integer. allowing for static shared memory declared in the kernel and for other dedicated uses.

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

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

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

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

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

The function anythread evaluates its argument for all threads in the current warp. value ) Additional Atomic Update mem = mem + value mem = mem – value 27 .false. They also combine that value with the value of the second argument.8. Atomic functions are only supported by devices with compute capability 1. 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. Arithmetic and Bitwise Atomic Functions These atomic functions read and return the value of the first argument.true.Chapter 3. which must be a variable or array element in shared memory (with the shared attribute) or in device global memory (with the device attribute). Compute capability 1.2 or higher is required if the first argument has the shared attribute. for all threads in the warp. Reference ANYTHREAD The anythread function is a warp-vote operation with a single scalar logical argument: if( anythread(a(i)<0. Example: if( ballot(int_value) ) allneg = . depending on the function. value ) atomicsub( mem. 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. Both arguments must be of type integer(kind=4). These functions are: Table 3.false. Arithmetic and Bitwise Atomic Functions Function atomicadd( mem. only if the value of the argument is . Atomic Functions The atomic functions read and write the value of their first operand. The value of the function is . Note The return value for each of these functions is the first argument.0) ) allneg = . and store the combined value back to the first argument location.true.0. mem.1 and higher. This function is only supported by devices of compute capability 2.

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

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

30 .

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

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

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

dst and src may be of TYPE(C_DEVPTR) or TYPE(C_PTR). kdir may be optional. src. cudaMemset integer function cudaMemset(devptr. width. and height are in terms of bytes. dpitch. spitch. If kdir is specified. in which case the count is in term of bytes. src.1. in which case the count is in term of bytes.” on page 21. kdir. it must be one of the defined enums cudaMemcpyHostToDevice.” on page 20. refer to “Data Transfer Using Runtime Routines. The count is in terms of elements. dst and src may be any device or host array.1. in which case the pitch. This function operates on page-locked host memory only. cudaMemset2D integer function cudaMemset2D(devptr. scalar or array. Or. or cudaMemcpyDeviceToDevice. “Device Code Intrinsic Datatypes.” on page 21. cudaMemcpyAsync integer function cudaMemcpyAsync(dst. dst and src may be any device or host. “Device Code Intrinsic Datatypes. dst and src may be of TYPE(C_DEVPTR) or TYPE(C_PTR). devptr may be any device array of a supported type specified in Table 3. The count is in terms of elements. and height are in terms of elements. count) cudaMemset sets a location or array to the specified value. otherwise the stream argument is optional and defaults to zero. it must be one of the defined enums cudaMemcpyHostToDevice.” on page 21.1. width. or cudaMemcpyDeviceToDevice. Or. width. refer to “Data Transfer Using Runtime Routines.” on page 21.” on page 21. “Device Code Intrinsic Datatypes. stream) cudaMemcpy copies data from one location to another. The copy can be associated with a stream by passing a non-zero stream argument. “Device Code Intrinsic Datatypes. of a supported type specified in Table 3. value. of a supported type specified in Table 3. scalar or array. pitch. The count is in terms of elements. dst and src may be any device or host.1. for more information. The value must match in type and kind. Alternatively. If kdir is specified. devptr may be any device scalar or array of a supported type specified in Table 3. in which case the count is in term of bytes. and the lowest byte of value is used. cudaMemcpyDeviceToHost. cudaMemcpyDeviceToHost. The width 34 . for more information. src. count.1. devptr may be of TYPE(C_DEVPTR). kdir may be optional.” on page 20. Alternatively. cudaMemcpy integer function cudaMemcpy(dst. cudaMemcpy2D integer function cudaMemcpy2D(dst. width. count. value. “Device Code Intrinsic Datatypes.Memory Management cudaFreeArray frees an array that was allocated on the device. The pitch. height) cudaMemset2D sets an array to the specified value. height. and the lowest byte of value is used. kdir) cudaMemcpy copies data from one location to another. The value must match in type and kind. of a supported type specified in Table 3. kdir) cudaMemcpy2D copies data from one location to another. devptr may be of TYPE(C_DEVPTR).

srcx. dsty. count. cudaMemcpy2DAsync integer function cudaMemcpy2DAsync(dst. in which case the width and height are in term of bytes. refer to “Data Transfer Using Runtime Routines. or cudaMemcpyDeviceToDevice. srca. srca. 35 . height. width. cudaMemcpy2DFromArray integer function cudaMemcpy2DFromArray(dst. kdir cudaMemcpy2DFromArray copies array data to and from the device. dstx. srcx. or cudaMemcpyDeviceToDevice. for more information. srcx. dst and src may be any device or host array. kdir cudaMemcpyToArray copies array data to and from the device. kdir. count. dsty. dpitch. of a supported type specified in Table 3.” on page 20. width. it must be one of the defined enums cudaMemcpyHostToDevice.” on page 21. in which case the width and height are in term of bytes. kdir may be optional. kdir cudaMemcpyFromArray copies array data to and from the device. srcy. dst and src may be of TYPE(C_DEVPTR) or TYPE(C_PTR).” on page 20.1. for more information. height. srcy. cudaMemcpyToArray integer function cudaMemcpyToArray(dsta. kdir) type(cudaArrayPtr) :: dsta integer :: dstx. src. it must be one of the defined enums cudaMemcpyHostToDevice. dstx. dpitch. otherwise the stream argument is optional and defaults to zero. If kdir is specified. dst and src may be of TYPE(C_DEVPTR) or TYPE(C_PTR). Alternatively. kdir) type(cudaArrayPtr) :: srca integer :: dstx. This function operates on page-locked host memory only. spitch. kdir) type(cudaArrayPtr) :: dsta integer :: dstx. cudaMemcpyDeviceToHost. height. width. Runtime APIs and height are in terms of elements. cudaMemcpy2DToArray integer function cudaMemcpy2DToArray(dsta. refer to “Data Transfer Using Runtime Routines. count. If kdir is specified. spitch. dsty. height. kdir cudaMemcpy2DToArray copies array data to and from the device. cudaMemcpyFromArray integer function cudaMemcpyFromArray(dst. kdir may be optional. kdir) type(cudaArrayPtr) :: srca integer :: dpitch. src. Alternatively.Chapter 4. “Device Code Intrinsic Datatypes. dsty. The width and height are in terms of elements. srcy. The copy can be associated with a stream by passing a non-zero stream argument. count. src. stream) cudaMemcpy2D copies data from one location to another. dsty. cudaMemcpyDeviceToHost. width. width. height. spitch.

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

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

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

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

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

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

42 .

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

k) * Bsub(k.Matrix Multiplication Example real :: Cij ! Get the thread indices tx = threadidx%x ty = threadidx%y ! This thread computes C(i.L) ) ! Copy A and B to the device Adev = A(1:N. 16.:) :: A.Cdev ! dim3 variables to define the grid and block shapes type(dim3) :: dimGrid. 1 ) call mmul_kernel<<<dimGrid. ! Copy the results back and free up memory C(1:N. L/16.ty) = B(kb+tx-1.dimBlock>>>( Adev.j) = Cij end subroutine mmul_kernel ! The host routine to drive the matrix multiplication subroutine mmul( A. B.16 Cij = Cij + Asub(tx.:) = B(1:M.kb+ty-1) Bsub(tx.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. allocatable. the block size do kb = 1.1:L) ! Create the grid and block dimensions dimGrid = dim3( N/16.j) do k = 1. B. dimBlock ! Get the array sizes N = size( A. Cdev(N.1:L) = Cdev deallocate( Adev.j)) i = (blockidx%x-1) * 16 + tx j = (blockidx%y-1) * 16 + ty Cij = 0. C ) real.Bdev. dimension(:.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. 1 ) M = size( A. Cdev ) end subroutine mmul end module mmul_mod N 44 .:) * B(:. 2 ) L = size( B.M).:) :: Adev.L). 1 ) dimBlock = dim3( 16.0 ! Do the k loop in chunks of 16.j) = sum(A(i. C ! allocatable device arrays real. device.1:M) Bdev(:. 16 ! Fill the submatrices ! Each of the 16x16 threads in the thread block ! loads one element of Asub and Bsub Asub(tx. Bdev.ty) = A(i. M. Bdev(M. dimension(:. 2 ) ! Allocate the device arrays allocate( Adev(N. Cdev. Bdev.

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

46 .

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

48 .

Sign up to vote on this title
UsefulNot useful