10 Commits

Author SHA1 Message Date
Mahesh Doijade
0ec4bd58e5 cudaCompressibleMemory: when unable to allocate compressible memory waive the execution 2020-06-01 20:22:22 +05:30
Mahesh Doijade
908dddb207 cudaCompressibleMemory: refactor and refine, and use only cuMemMap API for non-compressible allocation.
hence, remove use of cudaMalloc.
2020-05-27 19:00:33 +05:30
Mahesh Doijade
b60c149b17 Correct the runtime API section in documentation when sample has both driver as well as runtime API 2020-05-22 22:33:02 +05:30
Mahesh Doijade
56b5b4c767 -- add cudaCompressibleMemory sample
-- fix simpleD3D12 build issue
-- fix simpleVulkanMMAP shader file locating issue
2020-05-22 22:07:30 +05:30
Mahesh Doijade
53e9e413a8 Update VS project files to use cuda 11.0 props/targets 2020-05-19 18:16:05 +05:30
Mahesh Doijade
4f6e02970b Add and update samples for cuda 11.0 support 2020-05-18 22:22:06 +05:30
Mahesh Doijade
6be514679b Add and update samples for cuda 10.2 support 2019-10-23 19:00:39 +05:30
Mahesh Doijade
489d9f7b1f conjugateGradientCudaGraphs: Add cusparseScsrmv & cublasSdot instead of
gpuSpMV & gpuDotProduct as they are graph compliant
2019-08-19 15:47:24 +05:30
Mahesh Doijade
c8483e0798 Add release notes for cuda 10.1 update 2 support 2019-08-05 18:58:54 +05:30
Mahesh Doijade
58d0ceff18 Add and update samples for cuda 10.1 update 2 support 2019-08-02 23:15:47 +05:30
591 changed files with 118324 additions and 2961 deletions

View File

@@ -0,0 +1,142 @@
FreeImage Public License - Version 1.0
---------------------------------------------
1. Definitions.
1.1. "Contributor" means each entity that creates or contributes to the creation of Modifications.
1.2. "Contributor Version" means the combination of the Original Code, prior Modifications used by a Contributor, and the Modifications made by that particular Contributor.
1.3. "Covered Code" means the Original Code or Modifications or the combination of the Original Code and Modifications, in each case including portions thereof.
1.4. "Electronic Distribution Mechanism" means a mechanism generally accepted in the software development community for the electronic transfer of data.
1.5. "Executable" means Covered Code in any form other than Source Code.
1.6. "Initial Developer" means the individual or entity identified as the Initial Developer in the Source Code notice required by Exhibit A.
1.7. "Larger Work" means a work which combines Covered Code or portions thereof with code not governed by the terms of this License.
1.8. "License" means this document.
1.9. "Modifications" means any addition to or deletion from the substance or structure of either the Original Code or any previous Modifications. When Covered Code is released as a series of files, a
Modification is:
A. Any addition to or deletion from the contents of a file containing Original Code or previous Modifications.
B. Any new file that contains any part of the Original Code or previous Modifications.
1.10. "Original Code" means Source Code of computer software code which is described in the Source Code notice required by Exhibit A as Original Code, and which, at the time of its release under this License is not already Covered Code governed by this License.
1.11. "Source Code" means the preferred form of the Covered Code for making modifications to it, including all modules it contains, plus any associated interface definition files, scripts used to control
compilation and installation of an Executable, or a list of source code differential comparisons against either the Original Code or another well known, available Covered Code of the Contributor's choice. The Source Code can be in a compressed or archival form, provided the appropriate decompression or de-archiving software is widely available for no charge.
1.12. "You" means an individual or a legal entity exercising rights under, and complying with all of the terms of, this License or a future version of this License issued under Section 6.1. For legal entities, "You" includes any entity which controls, is controlled by, or is under common control with You. For purposes of this definition, "control" means (a) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or otherwise, or (b) ownership of fifty percent (50%) or more of the outstanding shares or beneficial ownership of such entity.
2. Source Code License.
2.1. The Initial Developer Grant.
The Initial Developer hereby grants You a world-wide, royalty-free, non-exclusive license, subject to third party intellectual property claims:
(a) to use, reproduce, modify, display, perform, sublicense and distribute the Original Code (or portions thereof) with or without Modifications, or as part of a Larger Work; and
(b) under patents now or hereafter owned or controlled by Initial Developer, to make, have made, use and sell ("Utilize") the Original Code (or portions thereof), but solely to the extent that
any such patent is reasonably necessary to enable You to Utilize the Original Code (or portions thereof) and not to any greater extent that may be necessary to Utilize further Modifications or
combinations.
2.2. Contributor Grant.
Each Contributor hereby grants You a world-wide, royalty-free, non-exclusive license, subject to third party intellectual property claims:
(a) to use, reproduce, modify, display, perform, sublicense and distribute the Modifications created by such Contributor (or portions thereof) either on an unmodified basis, with other Modifications, as Covered Code or as part of a Larger Work; and
(b) under patents now or hereafter owned or controlled by Contributor, to Utilize the Contributor Version (or portions thereof), but solely to the extent that any such patent is reasonably necessary to enable You to Utilize the Contributor Version (or portions thereof), and not to any greater extent that
may be necessary to Utilize further Modifications or combinations.
3. Distribution Obligations.
3.1. Application of License.
The Modifications which You create or to which You contribute are governed by the terms of this License, including without limitation Section 2.2. The Source Code version of Covered Code may be distributed only under the terms of this License or a future version of this License released under Section 6.1, and You must include a copy of this License with every copy of the Source Code You distribute. You may not offer or impose any terms on any Source Code version that alters or
restricts the applicable version of this License or the recipients' rights hereunder. However, You may include an additional document offering the additional rights described in Section 3.5.
3.2. Availability of Source Code.
Any Modification which You create or to which You contribute must be made available in Source Code form under the terms of this License either on the same media as an Executable version or via an accepted Electronic Distribution Mechanism to anyone to whom you made an Executable version available; and if made available via Electronic Distribution Mechanism, must remain available for at least twelve (12) months after the date it initially became available, or at least six (6) months after a subsequent version of that particular Modification has been made available to such recipients. You are responsible for ensuring that the Source Code version remains available even if the Electronic Distribution Mechanism is maintained by a third party.
3.3. Description of Modifications.
You must cause all Covered Code to which you contribute to contain a file documenting the changes You made to create that Covered Code and the date of any change. You must include a prominent statement that the Modification is derived, directly or indirectly, from Original Code provided by the Initial Developer and including the name of the Initial Developer in (a) the Source Code, and (b) in any notice in an Executable version or related documentation in which You describe the origin or ownership of the Covered Code.
3.4. Intellectual Property Matters
(a) Third Party Claims.
If You have knowledge that a party claims an intellectual property right in particular functionality or code (or its utilization under this License), you must include a text file with the source code distribution titled "LEGAL" which describes the claim and the party making the claim in sufficient detail that a recipient will know whom to contact. If you obtain such knowledge after You make Your Modification available as described in Section 3.2, You shall promptly modify the LEGAL file in all copies You make
available thereafter and shall take other steps (such as notifying appropriate mailing lists or newsgroups) reasonably calculated to inform those who received the Covered Code that new knowledge has been obtained.
(b) Contributor APIs.
If Your Modification is an application programming interface and You own or control patents which are reasonably necessary to implement that API, you must also include this information in the LEGAL file.
3.5. Required Notices.
You must duplicate the notice in Exhibit A in each file of the Source Code, and this License in any documentation for the Source Code, where You describe recipients' rights relating to Covered Code. If You created one or more Modification(s), You may add your name as a Contributor to the notice described in Exhibit A. If it is not possible to put such notice in a particular Source Code file due to its
structure, then you must include such notice in a location (such as a relevant directory file) where a user would be likely to look for such a notice. You may choose to offer, and to charge a fee for, warranty, support, indemnity or liability obligations to one or more recipients of Covered Code. However, You may do so only on Your own behalf, and not on behalf of the Initial Developer or any Contributor. You must make it absolutely clear than any such warranty, support, indemnity or
liability obligation is offered by You alone, and You hereby agree to indemnify the Initial Developer and every Contributor for any liability incurred by the Initial Developer or such Contributor as a result of
warranty, support, indemnity or liability terms You offer.
3.6. Distribution of Executable Versions.
You may distribute Covered Code in Executable form only if the requirements of Section 3.1-3.5 have been met for that Covered Code, and if You include a notice stating that the Source Code version of the Covered Code is available under the terms of this License, including a description of how and where You have fulfilled the obligations of Section 3.2. The notice must be conspicuously included in any notice in an Executable version, related documentation or collateral in which You
describe recipients' rights relating to the Covered Code. You may distribute the Executable version of Covered Code under a license of Your choice, which may contain terms different from this License,
provided that You are in compliance with the terms of this License and that the license for the Executable version does not attempt to limit or alter the recipient's rights in the Source Code version from the rights set forth in this License. If You distribute the Executable version under a different license You must make it absolutely clear that any terms which differ from this License are offered by You alone, not by the Initial Developer or any Contributor. You hereby agree to indemnify the Initial Developer and every Contributor for any liability incurred by the Initial Developer or such Contributor as a result of any such terms You offer.
3.7. Larger Works.
You may create a Larger Work by combining Covered Code with other code not governed by the terms of this License and distribute the Larger Work as a single product. In such a case, You must make sure the requirements of this License are fulfilled for the Covered Code.
4. Inability to Comply Due to Statute or Regulation.
If it is impossible for You to comply with any of the terms of this License with respect to some or all of the Covered Code due to statute or regulation then You must: (a) comply with the terms of this License to the maximum extent possible; and (b) describe the limitations and the code they affect. Such description must be included in the LEGAL file described in Section 3.4 and must be included with all distributions of the Source Code. Except to the extent prohibited by statute or regulation, such description must be sufficiently detailed for a recipient of ordinary skill to be able to understand it.
5. Application of this License.
This License applies to code to which the Initial Developer has attached the notice in Exhibit A, and to related Covered Code.
6. Versions of the License.
6.1. New Versions.
Floris van den Berg may publish revised and/or new versions of the License from time to time. Each version will be given a distinguishing version number.
6.2. Effect of New Versions.
Once Covered Code has been published under a particular version of the License, You may always continue to use it under the terms of that version. You may also choose to use such Covered Code under the terms of any subsequent version of the License published by Floris van den Berg
No one other than Floris van den Berg has the right to modify the terms applicable to Covered Code created under this License.
6.3. Derivative Works.
If you create or use a modified version of this License (which you may only do in order to apply it to code which is not already Covered Code governed by this License), you must (a) rename Your license so that the phrases "FreeImage", `FreeImage Public License", "FIPL", or any confusingly similar phrase do not appear anywhere in your license and (b) otherwise make it clear that your version of the license contains terms which differ from the FreeImage Public License. (Filling in the name of the Initial Developer, Original Code or Contributor in the notice described in Exhibit A shall not of themselves be deemed to be modifications of this License.)
7. DISCLAIMER OF WARRANTY.
COVERED CODE IS PROVIDED UNDER THIS LICENSE ON AN "AS IS" BASIS, WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, WITHOUT LIMITATION, WARRANTIES THAT THE COVERED CODE IS FREE OF DEFECTS, MERCHANTABLE, FIT FOR A PARTICULAR PURPOSE OR NON-INFRINGING. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE COVERED CODE IS WITH YOU. SHOULD ANY COVERED CODE PROVE DEFECTIVE IN ANY RESPECT, YOU (NOT THE INITIAL DEVELOPER OR ANY OTHER CONTRIBUTOR) ASSUME THE COST OF ANY NECESSARY SERVICING, REPAIR OR CORRECTION. THIS DISCLAIMER OF WARRANTY CONSTITUTES AN ESSENTIAL PART OF THIS LICENSE. NO USE OF ANY COVERED CODE IS AUTHORIZED HEREUNDER EXCEPT UNDER THIS DISCLAIMER.
8. TERMINATION.
This License and the rights granted hereunder will terminate automatically if You fail to comply with terms herein and fail to cure such breach within 30 days of becoming aware of the breach. All sublicenses to the Covered Code which are properly granted shall survive any termination of this License. Provisions which, by their nature, must remain in effect beyond the termination of this License shall survive.
9. LIMITATION OF LIABILITY.
UNDER NO CIRCUMSTANCES AND UNDER NO LEGAL THEORY, WHETHER TORT (INCLUDING NEGLIGENCE), CONTRACT, OR OTHERWISE, SHALL THE INITIAL DEVELOPER, ANY OTHER CONTRIBUTOR, OR ANY DISTRIBUTOR OF COVERED CODE, OR ANY SUPPLIER OF ANY OF SUCH PARTIES, BE LIABLE TO YOU OR ANY OTHER PERSON FOR ANY INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES OF ANY CHARACTER INCLUDING, WITHOUT LIMITATION, DAMAGES FOR LOSS OF GOODWILL, WORK STOPPAGE, COMPUTER FAILURE OR MALFUNCTION, OR ANY AND ALL OTHER COMMERCIAL DAMAGES OR LOSSES, EVEN IF SUCH PARTY SHALL HAVE BEEN INFORMED OF THE POSSIBILITY OF SUCH DAMAGES. THIS LIMITATION OF LIABILITY SHALL NOT APPLY TO LIABILITY FOR DEATH OR PERSONAL INJURY RESULTING FROM SUCH PARTY'S NEGLIGENCE TO THE EXTENT APPLICABLE LAW PROHIBITS SUCH LIMITATION. SOME JURISDICTIONS DO NOT ALLOW THE
EXCLUSION OR LIMITATION OF INCIDENTAL OR CONSEQUENTIAL DAMAGES, SO THAT EXCLUSION AND LIMITATION MAY NOT APPLY TO YOU.
10. U.S. GOVERNMENT END USERS.
The Covered Code is a "commercial item," as that term is defined in 48 C.F.R. 2.101 (Oct. 1995), consisting of "commercial computer software" and "commercial computer software documentation," as such terms are used in 48 C.F.R. 12.212 (Sept. 1995). Consistent with 48 C.F.R. 12.212 and 48 C.F.R. 227.7202-1 through 227.7202-4 (June 1995), all U.S. Government End Users acquire Covered Code with only those rights set forth herein.
11. MISCELLANEOUS.
This License represents the complete agreement concerning subject matter hereof. If any provision of this License is held to be unenforceable, such provision shall be reformed only to the extent necessary to make it enforceable. This License shall be governed by Dutch law provisions (except to the extent applicable law, if any, provides otherwise), excluding its conflict-of-law provisions. With respect to disputes in which at least one party is a citizen of, or an entity chartered or registered to do business in, the The Netherlands: (a) unless otherwise agreed in writing, all disputes relating to this License (excepting any dispute relating to intellectual property rights) shall be subject to final and binding arbitration, with the losing party paying all costs of arbitration; (b) any arbitration relating to this Agreement shall be held in Almelo, The Netherlands; and (c) any litigation relating to this Agreement shall be subject to the jurisdiction of the court of Almelo, The Netherlands with the losing party responsible for costs, including without limitation, court costs and reasonable attorneys fees and expenses. Any law or regulation which provides that the language of a contract shall be construed against the drafter shall not apply to this License.
12. RESPONSIBILITY FOR CLAIMS.
Except in cases where another Contributor has failed to comply with Section 3.4, You are responsible for damages arising, directly or indirectly, out of Your utilization of rights under this License, based
on the number of copies of Covered Code you made available, the revenues you received from utilizing such rights, and other relevant factors. You agree to work with affected parties to distribute
responsibility on an equitable basis.
EXHIBIT A.
"The contents of this file are subject to the FreeImage Public License Version 1.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://home.wxs.nl/~flvdberg/freeimage-license.txt
Software distributed under the License is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See the License for the specific language governing rights and limitations under the License.

File diff suppressed because it is too large Load Diff

Binary file not shown.

Binary file not shown.

Binary file not shown.

Binary file not shown.

Binary file not shown.

Binary file not shown.

Binary file not shown.

Binary file not shown.

Binary file not shown.

197
Common/UtilNPP/Exceptions.h Normal file
View File

@@ -0,0 +1,197 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#ifndef NV_UTIL_NPP_EXCEPTIONS_H
#define NV_UTIL_NPP_EXCEPTIONS_H
#include <string>
#include <sstream>
#include <iostream>
/// All npp related C++ classes are put into the npp namespace.
namespace npp
{
/// Exception base class.
/// This exception base class will be used for everything C++ throught
/// the NPP project.
/// The exception contains a string message, as well as data fields for a string
/// containing the name of the file as well as the line number where the exception was thrown.
/// The easiest way of throwing exceptions and providing filename and line number is
/// to use one of the ASSERT macros defined for that purpose.
class Exception
{
public:
/// Constructor.
/// \param rMessage A message with information as to why the exception was thrown.
/// \param rFileName The name of the file where the exception was thrown.
/// \param nLineNumber Line number in the file where the exception was thrown.
explicit
Exception(const std::string &rMessage = "", const std::string &rFileName = "", unsigned int nLineNumber = 0)
: sMessage_(rMessage), sFileName_(rFileName), nLineNumber_(nLineNumber)
{ };
Exception(const Exception &rException)
: sMessage_(rException.sMessage_), sFileName_(rException.sFileName_), nLineNumber_(rException.nLineNumber_)
{ };
virtual
~Exception()
{ };
/// Get the exception's message.
const
std::string &
message()
const
{
return sMessage_;
}
/// Get the exception's file info.
const
std::string &
fileName()
const
{
return sFileName_;
}
/// Get the exceptions's line info.
unsigned int
lineNumber()
const
{
return nLineNumber_;
}
/// Create a clone of this exception.
/// This creates a new Exception object on the heap. It is
/// the responsibility of the user of this function to free this memory
/// (delete x).
virtual
Exception *
clone()
const
{
return new Exception(*this);
}
/// Create a single string with all the exceptions information.
/// The virtual toString() method is used by the operator<<()
/// so that all exceptions derived from this base-class can print
/// their full information correctly even if a reference to their
/// exact type is not had at the time of printing (i.e. the basic
/// operator<<() is used).
virtual
std::string
toString()
const
{
std::ostringstream oOutputString;
oOutputString << fileName() << ":" << lineNumber() << ": " << message();
return oOutputString.str();
}
private:
std::string sMessage_; ///< Message regarding the cause of the exception.
std::string sFileName_; ///< Name of the file where the exception was thrown.
unsigned int nLineNumber_; ///< Line number in the file where the exception was thrown
};
/// Output stream inserter for Exception.
/// \param rOutputStream The stream the exception information is written to.
/// \param rException The exception that's being written.
/// \return Reference to the output stream being used.
std::ostream &
operator << (std::ostream &rOutputStream, const Exception &rException)
{
rOutputStream << rException.toString();
return rOutputStream;
}
/// Basic assert macro.
/// This macro should be used to enforce any kind of pre or post conditions.
/// Unlike the C-runtime assert macro, this macro does not abort execution, but throws
/// a C++ exception. The exception is automatically filled with information about the failing
/// condition, the filename and line number where the exception was thrown.
/// \note The macro is written in such a way that omitting a semicolon after its usage
/// causes a compiler error. The correct way to invoke this macro is:
/// NPP_ASSERT(n < MAX);
#define NPP_ASSERT(C) do {if (!(C)) throw npp::Exception(#C " assertion faild!", __FILE__, __LINE__);} while(false)
// ASSERT macro.
// Same functionality as the basic assert macro with the added ability to pass
// a message M. M should be a string literal.
// Note: Never use code inside ASSERT() that causes a side-effect ASSERT macros may get compiled
// out in release mode.
#define NPP_ASSERT_MSG(C, M) do {if (!(C)) throw npp::Exception(#C " assertion faild! Message: " M, __FILE__, __LINE__);} while(false)
#ifdef _DEBUG
/// Basic debug assert macro.
/// This macro is identical in every respect to NPP_ASSERT(C) but it does get compiled to a
/// no-op in release builds. It is therefor of utmost importance to not put statements into
/// this macro that cause side effects required for correct program execution.
#define NPP_DEBUG_ASSERT(C) do {if (!(C)) throw npp::Exception(#C " debug assertion faild!", __FILE__, __LINE__);} while(false)
#else
#define NPP_DEBUG_ASSERT(C)
#endif
/// ASSERT for null-pointer test.
/// It is safe to put code with side effects into this macro. Also: This macro never
/// gets compiled to a no-op because resource allocation may fail based on external causes not under
/// control of a software developer.
#define NPP_ASSERT_NOT_NULL(P) do {if ((P) == 0) throw npp::Exception(#P " not null assertion faild!", __FILE__, __LINE__);} while(false)
/// Macro for flagging methods as not implemented.
/// The macro throws an exception with a message that an implementation was missing
#define NPP_NOT_IMPLEMENTED() do {throw npp::Exception("Implementation missing!", __FILE__, __LINE__);} while(false)
/// Macro for checking error return code of CUDA (runtime) calls.
/// This macro never gets disabled.
#define NPP_CHECK_CUDA(S) do {cudaError_t eCUDAResult; \
eCUDAResult = S; \
if (eCUDAResult != cudaSuccess) std::cout << "NPP_CHECK_CUDA - eCUDAResult = " << eCUDAResult << std::endl; \
NPP_ASSERT(eCUDAResult == cudaSuccess);} while (false)
/// Macro for checking error return code for NPP calls.
#define NPP_CHECK_NPP(S) do {NppStatus eStatusNPP; \
eStatusNPP = S; \
if (eStatusNPP != NPP_SUCCESS) std::cout << "NPP_CHECK_NPP - eStatusNPP = " << _cudaGetErrorEnum(eStatusNPP) << "("<< eStatusNPP << ")" << std::endl; \
NPP_ASSERT(eStatusNPP == NPP_SUCCESS);} while (false)
/// Macro for checking error return codes from cuFFT calls.
#define NPP_CHECK_CUFFT(S) do {cufftResult eCUFFTResult; \
eCUFFTResult = S; \
if (eCUFFTResult != NPP_SUCCESS) std::cout << "NPP_CHECK_CUFFT - eCUFFTResult = " << eCUFFTResult << std::endl; \
NPP_ASSERT(eCUFFTResult == CUFFT_SUCCESS);} while (false)
} // npp namespace
#endif // NV_UTIL_NPP_EXCEPTIONS_H

155
Common/UtilNPP/Image.h Normal file
View File

@@ -0,0 +1,155 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#ifndef NV_UTIL_NPP_IMAGE_H
#define NV_UTIL_NPP_IMAGE_H
#include <cstddef>
namespace npp
{
class Image
{
public:
struct Size
{
unsigned int nWidth;
unsigned int nHeight;
Size() : nWidth(0), nHeight(0)
{ };
Size(unsigned int nWidthNew, unsigned nHeightNew) : nWidth(nWidthNew), nHeight(nHeightNew)
{ };
Size(const Size &rSize) : nWidth(rSize.nWidth), nHeight(rSize.nHeight)
{ };
Size &
operator= (const Size &rSize)
{
if (&rSize == this)
{
return *this;
}
nWidth = rSize.nWidth;
nHeight = rSize.nHeight;
return *this;
}
void
swap(Size &rSize)
{
unsigned int nTemp;
nTemp = nWidth;
nWidth = rSize.nWidth;
rSize.nWidth = nTemp;
nTemp = nHeight;
nHeight = rSize.nHeight;
rSize.nHeight = nTemp;
}
};
Image()
{ };
Image(unsigned int nWidth, unsigned int nHeight) : oSize_(nWidth, nHeight)
{ };
Image(const Image::Size &rSize) : oSize_(rSize)
{ };
Image(const Image &rImage) : oSize_(rImage.oSize_)
{ };
virtual
~Image()
{ };
Image &
operator= (const Image &rImage)
{
if (&rImage == this)
{
return *this;
}
oSize_ = rImage.oSize_;
return *this;
};
unsigned int
width()
const
{
return oSize_.nWidth;
}
unsigned int
height()
const
{
return oSize_.nHeight;
}
Size
size()
const
{
return oSize_;
}
void
swap(Image &rImage)
{
oSize_.swap(rImage.oSize_);
}
private:
Size oSize_;
};
bool
operator== (const Image::Size &rFirst, const Image::Size &rSecond)
{
return rFirst.nWidth == rSecond.nWidth && rFirst.nHeight == rSecond.nHeight;
}
bool
operator!= (const Image::Size &rFirst, const Image::Size &rSecond)
{
return rFirst.nWidth != rSecond.nWidth || rFirst.nHeight != rSecond.nHeight;
}
} // npp namespace
#endif // NV_UTIL_NPP_IMAGE_H

View File

@@ -0,0 +1,80 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#ifndef NV_UTIL_NPP_IMAGE_ALLOCATORS_CPU_H
#define NV_UTIL_NPP_IMAGE_ALLOCATORS_CPU_H
#include "Exceptions.h"
namespace npp
{
template <typename D, size_t N>
class ImageAllocatorCPU
{
public:
static
D *
Malloc2D(unsigned int nWidth, unsigned int nHeight, unsigned int *pPitch)
{
NPP_ASSERT(nWidth * nHeight > 0);
D *pResult = new D[nWidth * N * nHeight];
*pPitch = nWidth * sizeof(D) * N;
return pResult;
};
static
void
Free2D(D *pPixels)
{
delete[] pPixels;
};
static
void
Copy2D(D *pDst, size_t nDstPitch, const D *pSrc, size_t nSrcPitch, size_t nWidth, size_t nHeight)
{
const void *pSrcLine = pSrc;
void *pDstLine = pDst;
for (size_t iLine = 0; iLine < nHeight; ++iLine)
{
// copy one line worth of data
memcpy(pDst, pSrc, nWidth * N * sizeof(D));
// move data pointers to next line
pDst += nDstPitch;
pSrc += nSrcPitch;
}
};
};
} // npp namespace
#endif // NV_UTIL_NPP_IMAGE_ALLOCATORS_CPU_H

File diff suppressed because it is too large Load Diff

149
Common/UtilNPP/ImageIO.h Normal file
View File

@@ -0,0 +1,149 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#ifndef NV_UTIL_NPP_IMAGE_IO_H
#define NV_UTIL_NPP_IMAGE_IO_H
#include "ImagesCPU.h"
#include "ImagesNPP.h"
#include "FreeImage.h"
#include "Exceptions.h"
#include <string>
#include "string.h"
// Error handler for FreeImage library.
// In case this handler is invoked, it throws an NPP exception.
void
FreeImageErrorHandler(FREE_IMAGE_FORMAT oFif, const char *zMessage)
{
throw npp::Exception(zMessage);
}
namespace npp
{
// Load a gray-scale image from disk.
void
loadImage(const std::string &rFileName, ImageCPU_8u_C1 &rImage)
{
// set your own FreeImage error handler
FreeImage_SetOutputMessage(FreeImageErrorHandler);
FREE_IMAGE_FORMAT eFormat = FreeImage_GetFileType(rFileName.c_str());
// no signature? try to guess the file format from the file extension
if (eFormat == FIF_UNKNOWN)
{
eFormat = FreeImage_GetFIFFromFilename(rFileName.c_str());
}
NPP_ASSERT(eFormat != FIF_UNKNOWN);
// check that the plugin has reading capabilities ...
FIBITMAP *pBitmap;
if (FreeImage_FIFSupportsReading(eFormat))
{
pBitmap = FreeImage_Load(eFormat, rFileName.c_str());
}
NPP_ASSERT(pBitmap != 0);
// make sure this is an 8-bit single channel image
NPP_ASSERT(FreeImage_GetColorType(pBitmap) == FIC_MINISBLACK);
NPP_ASSERT(FreeImage_GetBPP(pBitmap) == 8);
// create an ImageCPU to receive the loaded image data
ImageCPU_8u_C1 oImage(FreeImage_GetWidth(pBitmap), FreeImage_GetHeight(pBitmap));
// Copy the FreeImage data into the new ImageCPU
unsigned int nSrcPitch = FreeImage_GetPitch(pBitmap);
const Npp8u *pSrcLine = FreeImage_GetBits(pBitmap) + nSrcPitch * (FreeImage_GetHeight(pBitmap) -1);
Npp8u *pDstLine = oImage.data();
unsigned int nDstPitch = oImage.pitch();
for (size_t iLine = 0; iLine < oImage.height(); ++iLine)
{
memcpy(pDstLine, pSrcLine, oImage.width() * sizeof(Npp8u));
pSrcLine -= nSrcPitch;
pDstLine += nDstPitch;
}
// swap the user given image with our result image, effecively
// moving our newly loaded image data into the user provided shell
oImage.swap(rImage);
}
// Save an gray-scale image to disk.
void
saveImage(const std::string &rFileName, const ImageCPU_8u_C1 &rImage)
{
// create the result image storage using FreeImage so we can easily
// save
FIBITMAP *pResultBitmap = FreeImage_Allocate(rImage.width(), rImage.height(), 8 /* bits per pixel */);
NPP_ASSERT_NOT_NULL(pResultBitmap);
unsigned int nDstPitch = FreeImage_GetPitch(pResultBitmap);
Npp8u *pDstLine = FreeImage_GetBits(pResultBitmap) + nDstPitch * (rImage.height()-1);
const Npp8u *pSrcLine = rImage.data();
unsigned int nSrcPitch = rImage.pitch();
for (size_t iLine = 0; iLine < rImage.height(); ++iLine)
{
memcpy(pDstLine, pSrcLine, rImage.width() * sizeof(Npp8u));
pSrcLine += nSrcPitch;
pDstLine -= nDstPitch;
}
// now save the result image
bool bSuccess;
bSuccess = FreeImage_Save(FIF_PGM, pResultBitmap, rFileName.c_str(), 0) == TRUE;
NPP_ASSERT_MSG(bSuccess, "Failed to save result image.");
}
// Load a gray-scale image from disk.
void
loadImage(const std::string &rFileName, ImageNPP_8u_C1 &rImage)
{
ImageCPU_8u_C1 oImage;
loadImage(rFileName, oImage);
ImageNPP_8u_C1 oResult(oImage);
rImage.swap(oResult);
}
// Save an gray-scale image to disk.
void
saveImage(const std::string &rFileName, const ImageNPP_8u_C1 &rImage)
{
ImageCPU_8u_C1 oHostImage(rImage.size());
// copy the device result data
rImage.copyTo(oHostImage.data(), oHostImage.pitch());
saveImage(rFileName, oHostImage);
}
}
#endif // NV_UTIL_NPP_IMAGE_IO_H

View File

@@ -0,0 +1,171 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#ifndef NV_UTIL_NPP_IMAGE_PACKED_H
#define NV_UTIL_NPP_IMAGE_PACKED_H
#include "Image.h"
#include "Pixel.h"
namespace npp
{
template<typename D, size_t N, class A>
class ImagePacked: public npp::Image
{
public:
typedef npp::Pixel<D, N> tPixel;
typedef D tData;
static const size_t gnChannels = N;
typedef npp::Image::Size tSize;
ImagePacked(): aPixels_(0)
, nPitch_(0)
{
;
}
ImagePacked(unsigned int nWidth, unsigned int nHeight): Image(nWidth, nHeight)
, aPixels_(0)
, nPitch_(0)
{
aPixels_ = A::Malloc2D(width(), height(), &nPitch_);
}
ImagePacked(unsigned int nWidth, unsigned int nHeight, bool bTight): Image(nWidth, nHeight)
, aPixels_(0)
, nPitch_(0)
{
aPixels_ = A::Malloc2D(width(), height(), &nPitch_, bTight);
}
ImagePacked(const tSize &rSize): Image(rSize)
, aPixels_(0)
, nPitch_(0)
{
aPixels_ = A::Malloc2D(width(), height(), &nPitch_);
}
ImagePacked(const ImagePacked<D, N, A> &rImage): Image(rImage)
, aPixels_(0)
, nPitch_(rImage.pitch())
{
aPixels_ = A::Malloc2D(width(), height(), &nPitch_);
A::Copy2D(aPixels_, nPitch_, rImage.pixels(), rImage.pitch(), width(), height());
}
virtual
~ImagePacked()
{
A::Free2D(aPixels_);
}
ImagePacked &
operator= (const ImagePacked<D, N, A> &rImage)
{
// in case of self-assignment
if (&rImage == this)
{
return *this;
}
A::Free2D(aPixels_);
aPixels_ = 0;
nPitch_ = 0;
// assign parent class's data fields (width, height)
Image::operator =(rImage);
aPixels_ = A::Malloc2D(width(), height(), &nPitch_);
A::Copy2D(aPixels_, nPitch_, rImage.data(), rImage.pitch(), width(), height());
return *this;
}
unsigned int
pitch()
const
{
return nPitch_;
}
/// Get a pointer to the pixel array.
/// The result pointer can be offset to pixel at position (x, y) and
/// even negative offsets are allowed.
/// \param nX Horizontal pointer/array offset.
/// \param nY Vertical pointer/array offset.
/// \return Pointer to the pixel array (or first pixel in array with coordinates (nX, nY).
tPixel *
pixels(int nX = 0, int nY = 0)
{
return reinterpret_cast<tPixel *>(reinterpret_cast<unsigned char *>(aPixels_) + nY * pitch() + nX * gnChannels * sizeof(D));
}
const
tPixel *
pixels(int nX = 0, int nY = 0)
const
{
return reinterpret_cast<const tPixel *>(reinterpret_cast<unsigned char *>(aPixels_) + nY * pitch() + nX * gnChannels * sizeof(D));
}
D *
data(int nX = 0, int nY = 0)
{
return reinterpret_cast<D *>(pixels(nX, nY));
}
const
D *
data(int nX = 0, int nY = 0)
const
{
return reinterpret_cast<const D *>(pixels(nX, nY));
}
void
swap(ImagePacked<D, N, A> &rImage)
{
Image::swap(rImage);
tData *aTemp = aPixels_;
aPixels_ = rImage.aPixels_;
rImage.aPixels_ = aTemp;
unsigned int nTemp = nPitch_;
nPitch_ = rImage.nPitch_;
rImage.nPitch_ = nTemp;
}
private:
D *aPixels_;
unsigned int nPitch_;
};
} // npp namespace
#endif // NV_IMAGE_IPP_H

121
Common/UtilNPP/ImagesCPU.h Normal file
View File

@@ -0,0 +1,121 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#ifndef NV_UTIL_NPP_IMAGES_CPU_H
#define NV_UTIL_NPP_IMAGES_CPU_H
#include "ImagePacked.h"
#include "ImageAllocatorsCPU.h"
#include "Exceptions.h"
#include <npp.h>
namespace npp
{
template<typename D, unsigned int N, class A>
class ImageCPU: public npp::ImagePacked<D, N, A>
{
public:
ImageCPU()
{
;
}
ImageCPU(unsigned int nWidth, unsigned int nHeight): ImagePacked<D, N, A>(nWidth, nHeight)
{
;
}
explicit
ImageCPU(const npp::Image::Size &rSize): ImagePacked<D, N, A>(rSize)
{
;
}
ImageCPU(const ImageCPU<D, N, A> &rImage): Image(rImage)
{
;
}
virtual
~ImageCPU()
{
;
}
ImageCPU &
operator= (const ImageCPU<D, N, A> &rImage)
{
ImagePacked<D, N, A>::operator= (rImage);
return *this;
}
npp::Pixel<D, N> &
operator()(unsigned int iX, unsigned int iY)
{
return *ImagePacked<D, N, A>::pixels(iX, iY);
}
npp::Pixel<D, N>
operator()(unsigned int iX, unsigned int iY)
const
{
return *ImagePacked<D, N, A>::pixels(iX, iY);
}
};
typedef ImageCPU<Npp8u, 1, npp::ImageAllocatorCPU<Npp8u, 1> > ImageCPU_8u_C1;
typedef ImageCPU<Npp8u, 2, npp::ImageAllocatorCPU<Npp8u, 2> > ImageCPU_8u_C2;
typedef ImageCPU<Npp8u, 3, npp::ImageAllocatorCPU<Npp8u, 3> > ImageCPU_8u_C3;
typedef ImageCPU<Npp8u, 4, npp::ImageAllocatorCPU<Npp8u, 4> > ImageCPU_8u_C4;
typedef ImageCPU<Npp16u, 1, npp::ImageAllocatorCPU<Npp16u, 1> > ImageCPU_16u_C1;
typedef ImageCPU<Npp16u, 3, npp::ImageAllocatorCPU<Npp16u, 3> > ImageCPU_16u_C3;
typedef ImageCPU<Npp16u, 4, npp::ImageAllocatorCPU<Npp16u, 4> > ImageCPU_16u_C4;
typedef ImageCPU<Npp16s, 1, npp::ImageAllocatorCPU<Npp16s, 1> > ImageCPU_16s_C1;
typedef ImageCPU<Npp16s, 3, npp::ImageAllocatorCPU<Npp16s, 3> > ImageCPU_16s_C3;
typedef ImageCPU<Npp16s, 4, npp::ImageAllocatorCPU<Npp16s, 4> > ImageCPU_16s_C4;
typedef ImageCPU<Npp32s, 1, npp::ImageAllocatorCPU<Npp32s, 1> > ImageCPU_32s_C1;
typedef ImageCPU<Npp32s, 3, npp::ImageAllocatorCPU<Npp32s, 3> > ImageCPU_32s_C3;
typedef ImageCPU<Npp32s, 4, npp::ImageAllocatorCPU<Npp32s, 4> > ImageCPU_32s_C4;
typedef ImageCPU<Npp32f, 1, npp::ImageAllocatorCPU<Npp32f, 1> > ImageCPU_32f_C1;
typedef ImageCPU<Npp32f, 3, npp::ImageAllocatorCPU<Npp32f, 3> > ImageCPU_32f_C3;
typedef ImageCPU<Npp32f, 4, npp::ImageAllocatorCPU<Npp32f, 4> > ImageCPU_32f_C4;
} // npp namespace
#endif // NV_IMAGE_IPP_H

149
Common/UtilNPP/ImagesNPP.h Normal file
View File

@@ -0,0 +1,149 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#ifndef NV_UTIL_NPP_IMAGES_NPP_H
#define NV_UTIL_NPP_IMAGES_NPP_H
#include "Exceptions.h"
#include "ImagePacked.h"
#include "ImageAllocatorsNPP.h"
#include <cuda_runtime.h>
namespace npp
{
// forward declaration
template<typename D, unsigned int N, class A> class ImageCPU;
template<typename D, unsigned int N>
class ImageNPP: public npp::ImagePacked<D, N, npp::ImageAllocator<D, N> >
{
public:
ImageNPP()
{
;
}
ImageNPP(unsigned int nWidth, unsigned int nHeight, bool bTight = false): ImagePacked<D, N, npp::ImageAllocator<D, N> >(nWidth, nHeight, bTight)
{
;
}
ImageNPP(const npp::Image::Size &rSize): ImagePacked<D, N, npp::ImageAllocator<D, N> >(rSize)
{
;
}
ImageNPP(const ImageNPP<D, N> &rImage): Image(rImage)
{
;
}
template<class X>
explicit
ImageNPP(const ImageCPU<D, N, X> &rImage, bool bTight = false): ImagePacked<D, N, npp::ImageAllocator<D, N> >(rImage.width(), rImage.height(), bTight)
{
npp::ImageAllocator<D, N>::HostToDeviceCopy2D(ImagePacked<D, N, npp::ImageAllocator<D, N> >::data(),
ImagePacked<D, N, npp::ImageAllocator<D, N> >::pitch(),
rImage.data(),
rImage.pitch(),
ImagePacked<D, N, npp::ImageAllocator<D, N> >::width(),
ImagePacked<D, N, npp::ImageAllocator<D, N> >::height());
}
virtual
~ImageNPP()
{
;
}
ImageNPP &
operator= (const ImageNPP<D, N> &rImage)
{
ImagePacked<D, N, npp::ImageAllocator<D, N> >::operator= (rImage);
return *this;
}
void
copyTo(D *pData, unsigned int nPitch)
const
{
NPP_ASSERT((ImagePacked<D, N, npp::ImageAllocator<D, N> >::width() * sizeof(npp::Pixel<D, N>) <= nPitch));
npp::ImageAllocator<D, N>::DeviceToHostCopy2D(pData,
nPitch,
ImagePacked<D, N, npp::ImageAllocator<D, N> >::data(),
ImagePacked<D, N, npp::ImageAllocator<D, N> >::pitch(),
ImagePacked<D, N, npp::ImageAllocator<D, N> >::width(),
ImagePacked<D, N, npp::ImageAllocator<D, N> >::height());
}
void
copyFrom(D *pData, unsigned int nPitch)
{
NPP_ASSERT((ImagePacked<D, N, npp::ImageAllocator<D, N> >::width() * sizeof(npp::Pixel<D, N>) <= nPitch));
npp::ImageAllocator<D, N>::HostToDeviceCopy2D(ImagePacked<D, N, npp::ImageAllocator<D, N> >::data(),
ImagePacked<D, N, npp::ImageAllocator<D, N> >::pitch(),
pData,
nPitch,
ImagePacked<D, N, npp::ImageAllocator<D, N> >::width(),
ImagePacked<D, N, npp::ImageAllocator<D, N> >::height());
}
};
typedef ImageNPP<Npp8u, 1> ImageNPP_8u_C1;
typedef ImageNPP<Npp8u, 2> ImageNPP_8u_C2;
typedef ImageNPP<Npp8u, 3> ImageNPP_8u_C3;
typedef ImageNPP<Npp8u, 4> ImageNPP_8u_C4;
typedef ImageNPP<Npp16u, 1> ImageNPP_16u_C1;
typedef ImageNPP<Npp16u, 2> ImageNPP_16u_C2;
typedef ImageNPP<Npp16u, 3> ImageNPP_16u_C3;
typedef ImageNPP<Npp16u, 4> ImageNPP_16u_C4;
typedef ImageNPP<Npp16s, 1> ImageNPP_16s_C1;
typedef ImageNPP<Npp16s, 3> ImageNPP_16s_C3;
typedef ImageNPP<Npp16s, 4> ImageNPP_16s_C4;
typedef ImageNPP<Npp32s, 1> ImageNPP_32s_C1;
typedef ImageNPP<Npp32s, 3> ImageNPP_32s_C3;
typedef ImageNPP<Npp32s, 4> ImageNPP_32s_C4;
typedef ImageNPP<Npp32f, 1> ImageNPP_32f_C1;
typedef ImageNPP<Npp32f, 2> ImageNPP_32f_C2;
typedef ImageNPP<Npp32f, 3> ImageNPP_32f_C3;
typedef ImageNPP<Npp32f, 4> ImageNPP_32f_C4;
typedef ImageNPP<Npp64f, 1> ImageNPP_64f_C1;
typedef ImageNPP<Npp64f, 2> ImageNPP_64f_C2;
typedef ImageNPP<Npp64f, 3> ImageNPP_64f_C3;
typedef ImageNPP<Npp64f, 4> ImageNPP_64f_C4;
} // npp namespace
#endif // NV_UTIL_NPP_IMAGES_NPP_H

126
Common/UtilNPP/Pixel.h Normal file
View File

@@ -0,0 +1,126 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#ifndef NV_UTIL_PIXEL_H
#define NV_UTIL_PIXEL_H
#include "Exceptions.h"
namespace npp
{
template <typename D, size_t N>
struct Pixel
{ };
template <typename D>
struct Pixel<D, 1>
{
D x;
const D &
operator[](size_t iChannel)
const
{
NPP_ASSERT(iChannel < 1);
return (&x)[iChannel];
}
D &
operator[](size_t iChannel)
{
NPP_ASSERT(iChannel < 1);
return (&x)[iChannel];
}
};
template <typename D>
struct Pixel<D, 2>
{
D x,y;
const D &
operator[](size_t iChannel)
const
{
NPP_ASSERT(iChannel < 2);
return (&x)[iChannel];
}
D &
operator[](size_t iChannel)
{
NPP_ASSERT(iChannel < 2);
return (&x)[iChannel];
}
};
template <typename D>
struct Pixel<D, 3>
{
D x,y,z;
const D &
operator[](size_t iChannel)
const
{
NPP_ASSERT(iChannel < 3);
return (&x)[iChannel];
}
D &
operator[](size_t iChannel)
{
NPP_ASSERT(iChannel < 3);
return (&x)[iChannel];
}
};
template <typename D>
struct Pixel<D, 4>
{
D x, y, z, w;
const D &
operator[](size_t iChannel)
const
{
NPP_ASSERT(iChannel < 4);
return (&x)[iChannel];
}
D &
operator[](size_t iChannel)
{
NPP_ASSERT(iChannel < 4);
return (&x)[iChannel];
}
};
} // npp namespace
#endif // NV_UTIL_PIXEL_H

168
Common/UtilNPP/Signal.h Normal file
View File

@@ -0,0 +1,168 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#ifndef NV_UTIL_NPP_SIGNAL_H
#define NV_UTIL_NPP_SIGNAL_H
#include <cstring>
namespace npp
{
class Signal
{
public:
Signal() : nSize_(0)
{ };
explicit
Signal(size_t nSize) : nSize_(nSize)
{ };
Signal(const Signal &rSignal) : nSize_(rSignal.nSize_)
{ };
virtual
~Signal()
{ }
Signal &
operator= (const Signal &rSignal)
{
nSize_ = rSignal.nSize_;
return *this;
}
size_t
size()
const
{
return nSize_;
}
void
swap(Signal &rSignal)
{
size_t nTemp = nSize_;
nSize_ = rSignal.nSize_;
rSignal.nSize_ = nTemp;
}
private:
size_t nSize_;
};
template<typename D, class A>
class SignalTemplate: public Signal
{
public:
typedef D tData;
SignalTemplate(): aValues_(0)
{
;
}
SignalTemplate(size_t nSize): Signal(nSize)
, aValues_(0)
{
aValues_ = A::Malloc1D(size());
}
SignalTemplate(const SignalTemplate<D, A> &rSignal): Signal(rSignal)
, aValues_(0)
{
aValues_ = A::Malloc1D(size());
A::Copy1D(aValues_, rSignal.values(), size());
}
virtual
~SignalTemplate()
{
A::Free1D(aValues_);
}
SignalTemplate &
operator= (const SignalTemplate<D, A> &rSignal)
{
// in case of self-assignment
if (&rSignal == this)
{
return *this;
}
A::Free1D(aValues_);
this->aPixels_ = 0;
// assign parent class's data fields (width, height)
Signal::operator =(rSignal);
aValues_ = A::Malloc1D(size());
A::Copy1D(aValues_, rSignal.value(), size());
return *this;
}
/// Get a pointer to the pixel array.
/// The result pointer can be offset to pixel at position (x, y) and
/// even negative offsets are allowed.
/// \param nX Horizontal pointer/array offset.
/// \param nY Vertical pointer/array offset.
/// \return Pointer to the pixel array (or first pixel in array with coordinates (nX, nY).
tData *
values(int i = 0)
{
return aValues_ + i;
}
const
tData *
values(int i = 0)
const
{
return aValues_ + i;
}
void
swap(SignalTemplate<D, A> &rSignal)
{
Signal::swap(rSignal);
tData *aTemp = this->aValues_;
this->aValues_ = rSignal.aValues_;
rSignal.aValues_ = aTemp;
}
private:
D *aValues_;
};
} // npp namespace
#endif // NV_UTIL_NPP_SIGNAL_H

View File

@@ -0,0 +1,66 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#ifndef NV_UTIL_NPP_SIGNAL_ALLOCATORS_CPU_H
#define NV_UTIL_NPP_SIGNAL_ALLOCATORS_CPU_H
#include "Exceptions.h"
namespace npp
{
template <typename D>
class SignalAllocatorCPU
{
public:
static
D *
Malloc1D(unsigned int nSize)
{
return new D[nSize];;
};
static
void
Free1D(D *pPixels)
{
delete[] pPixels;
};
static
void
Copy1D(D *pDst, const D *pSrc, size_t nSize)
{
memcpy(pDst, pSrc, nSize * sizeof(D));
};
};
} // npp namespace
#endif // NV_UTIL_NPP_SIGNAL_ALLOCATORS_CPU_H

View File

@@ -0,0 +1,684 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#ifndef NV_UTIL_NPP_SIGNAL_ALLOCATORS_NPP_H
#define NV_UTIL_NPP_SIGNAL_ALLOCATORS_NPP_H
#include "Exceptions.h"
#include <npps.h>
#include <cuda_runtime.h>
namespace npp
{
template <typename D>
class SignalAllocator
{
};
template<>
class SignalAllocator<Npp8u>
{
public:
static
Npp8u *
Malloc1D(size_t nSize)
{
Npp8u *pResult = nppsMalloc_8u(static_cast<int>(nSize));
NPP_ASSERT(pResult != 0);
return pResult;
};
static
void
Free1D(Npp8u *pValues)
{
nppsFree(pValues);
};
static
void
Copy1D(Npp8u *pDst, const Npp8u *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp8u),cudaMemcpyDeviceToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
HostToDeviceCopy1D(Npp8u *pDst, const Npp8u *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp8u), cudaMemcpyHostToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
DeviceToHostCopy1D(Npp8u *pDst, const Npp8u *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp8u), cudaMemcpyDeviceToHost);
NPP_ASSERT(cudaSuccess == eResult);
};
};
template<>
class SignalAllocator<Npp16s>
{
public:
static
Npp16s *
Malloc1D(size_t nSize)
{
Npp16s *pResult = nppsMalloc_16s(static_cast<int>(nSize));
NPP_ASSERT(pResult != 0);
return pResult;
};
static
void
Free1D(Npp16s *pValues)
{
nppsFree(pValues);
};
static
void
Copy1D(Npp16s *pDst, const Npp16s *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp16s),cudaMemcpyDeviceToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
HostToDeviceCopy1D(Npp16s *pDst, const Npp16s *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp16s), cudaMemcpyHostToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
DeviceToHostCopy1D(Npp16s *pDst, const Npp16s *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp16s), cudaMemcpyDeviceToHost);
NPP_ASSERT(cudaSuccess == eResult);
};
};
template<>
class SignalAllocator<Npp16u>
{
public:
static
Npp16u *
Malloc1D(size_t nSize)
{
Npp16u *pResult = nppsMalloc_16u(static_cast<int>(nSize));
NPP_ASSERT(pResult != 0);
return pResult;
};
static
void
Free1D(Npp16u *pValues)
{
nppsFree(pValues);
};
static
void
Copy1D(Npp16u *pDst, const Npp16u *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp16u),cudaMemcpyDeviceToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
HostToDeviceCopy1D(Npp16u *pDst, const Npp16u *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp16u), cudaMemcpyHostToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
DeviceToHostCopy1D(Npp16u *pDst, const Npp16u *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp16u), cudaMemcpyDeviceToHost);
NPP_ASSERT(cudaSuccess == eResult);
};
};
template<>
class SignalAllocator<Npp16sc>
{
public:
static
Npp16sc *
Malloc1D(size_t nSize)
{
Npp16sc *pResult = nppsMalloc_16sc(static_cast<int>(nSize));
NPP_ASSERT(pResult != 0);
return pResult;
};
static
void
Free1D(Npp16sc *pValues)
{
nppsFree(pValues);
};
static
void
Copy1D(Npp16sc *pDst, const Npp16sc *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp16sc),cudaMemcpyDeviceToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
HostToDeviceCopy1D(Npp16sc *pDst, const Npp16sc *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp16sc), cudaMemcpyHostToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
DeviceToHostCopy1D(Npp16sc *pDst, const Npp16sc *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp16sc), cudaMemcpyDeviceToHost);
NPP_ASSERT(cudaSuccess == eResult);
};
};
template<>
class SignalAllocator<Npp32u>
{
public:
static
Npp32u *
Malloc1D(size_t nSize)
{
Npp32u *pResult = nppsMalloc_32u(static_cast<int>(nSize));
NPP_ASSERT(pResult != 0);
return pResult;
};
static
void
Free1D(Npp32u *pValues)
{
nppsFree(pValues);
};
static
void
Copy1D(Npp32u *pDst, const Npp32u *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp32u),cudaMemcpyDeviceToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
HostToDeviceCopy1D(Npp32u *pDst, const Npp32u *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp32u), cudaMemcpyHostToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
DeviceToHostCopy1D(Npp32u *pDst, const Npp32u *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp32u), cudaMemcpyDeviceToHost);
NPP_ASSERT(cudaSuccess == eResult);
};
};
template<>
class SignalAllocator<Npp32s>
{
public:
static
Npp32s *
Malloc1D(size_t nSize)
{
Npp32s *pResult = nppsMalloc_32s(static_cast<int>(nSize));
NPP_ASSERT(pResult != 0);
return pResult;
};
static
void
Free1D(Npp32s *pValues)
{
nppsFree(pValues);
};
static
void
Copy1D(Npp32s *pDst, const Npp32s *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp32s),cudaMemcpyDeviceToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
HostToDeviceCopy1D(Npp32s *pDst, const Npp32s *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp32s), cudaMemcpyHostToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
DeviceToHostCopy1D(Npp32s *pDst, const Npp32s *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp32s), cudaMemcpyDeviceToHost);
NPP_ASSERT(cudaSuccess == eResult);
};
};
template<>
class SignalAllocator<Npp32sc>
{
public:
static
Npp32sc *
Malloc1D(size_t nSize)
{
Npp32sc *pResult = nppsMalloc_32sc(static_cast<int>(nSize));
NPP_ASSERT(pResult != 0);
return pResult;
};
static
void
Free1D(Npp32sc *pValues)
{
nppsFree(pValues);
};
static
void
Copy1D(Npp32sc *pDst, const Npp32sc *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp32sc),cudaMemcpyDeviceToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
HostToDeviceCopy1D(Npp32sc *pDst, const Npp32sc *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp32sc), cudaMemcpyHostToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
DeviceToHostCopy1D(Npp32sc *pDst, const Npp32sc *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp32sc), cudaMemcpyDeviceToHost);
NPP_ASSERT(cudaSuccess == eResult);
};
};
template<>
class SignalAllocator<Npp32f>
{
public:
static
Npp32f *
Malloc1D(size_t nSize)
{
Npp32f *pResult = nppsMalloc_32f(static_cast<int>(nSize));
NPP_ASSERT(pResult != 0);
return pResult;
};
static
void
Free1D(Npp32f *pValues)
{
nppsFree(pValues);
};
static
void
Copy1D(Npp32f *pDst, const Npp32f *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp32f),cudaMemcpyDeviceToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
HostToDeviceCopy1D(Npp32f *pDst, const Npp32f *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp32f), cudaMemcpyHostToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
DeviceToHostCopy1D(Npp32f *pDst, const Npp32f *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp32f), cudaMemcpyDeviceToHost);
NPP_ASSERT(cudaSuccess == eResult);
};
};
template<>
class SignalAllocator<Npp32fc>
{
public:
static
Npp32fc *
Malloc1D(size_t nSize)
{
Npp32fc *pResult = nppsMalloc_32fc(static_cast<int>(nSize));
NPP_ASSERT(pResult != 0);
return pResult;
};
static
void
Free1D(Npp32fc *pValues)
{
nppsFree(pValues);
};
static
void
Copy1D(Npp32fc *pDst, const Npp32fc *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp32fc),cudaMemcpyDeviceToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
HostToDeviceCopy1D(Npp32fc *pDst, const Npp32fc *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp32fc), cudaMemcpyHostToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
DeviceToHostCopy1D(Npp32fc *pDst, const Npp32fc *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp32fc), cudaMemcpyDeviceToHost);
NPP_ASSERT(cudaSuccess == eResult);
};
};
template<>
class SignalAllocator<Npp64s>
{
public:
static
Npp64s *
Malloc1D(size_t nSize)
{
Npp64s *pResult = nppsMalloc_64s(static_cast<int>(nSize));
NPP_ASSERT(pResult != 0);
return pResult;
};
static
void
Free1D(Npp64s *pValues)
{
nppsFree(pValues);
};
static
void
Copy1D(Npp64s *pDst, const Npp64s *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp64s),cudaMemcpyDeviceToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
HostToDeviceCopy1D(Npp64s *pDst, const Npp64s *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp64s), cudaMemcpyHostToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
DeviceToHostCopy1D(Npp64s *pDst, const Npp64s *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp64s), cudaMemcpyDeviceToHost);
NPP_ASSERT(cudaSuccess == eResult);
};
};
template<>
class SignalAllocator<Npp64sc>
{
public:
static
Npp64sc *
Malloc1D(size_t nSize)
{
Npp64sc *pResult = nppsMalloc_64sc(static_cast<int>(nSize));
NPP_ASSERT(pResult != 0);
return pResult;
};
static
void
Free1D(Npp64sc *pValues)
{
nppsFree(pValues);
};
static
void
Copy1D(Npp64sc *pDst, const Npp64sc *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp64sc),cudaMemcpyDeviceToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
HostToDeviceCopy1D(Npp64sc *pDst, const Npp64sc *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp64sc), cudaMemcpyHostToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
DeviceToHostCopy1D(Npp64sc *pDst, const Npp64sc *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp64sc), cudaMemcpyDeviceToHost);
NPP_ASSERT(cudaSuccess == eResult);
};
};
template<>
class SignalAllocator<Npp64f>
{
public:
static
Npp64f *
Malloc1D(size_t nSize)
{
Npp64f *pResult = nppsMalloc_64f(static_cast<int>(nSize));
NPP_ASSERT(pResult != 0);
return pResult;
};
static
void
Free1D(Npp64f *pValues)
{
nppsFree(pValues);
};
static
void
Copy1D(Npp64f *pDst, const Npp64f *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp64f),cudaMemcpyDeviceToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
HostToDeviceCopy1D(Npp64f *pDst, const Npp64f *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp64f), cudaMemcpyHostToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
DeviceToHostCopy1D(Npp64f *pDst, const Npp64f *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp64f), cudaMemcpyDeviceToHost);
NPP_ASSERT(cudaSuccess == eResult);
};
};
template<>
class SignalAllocator<Npp64fc>
{
public:
static
Npp64fc *
Malloc1D(size_t nSize)
{
Npp64fc *pResult = nppsMalloc_64fc(static_cast<int>(nSize));
NPP_ASSERT(pResult != 0);
return pResult;
};
static
void
Free1D(Npp64fc *pValues)
{
nppsFree(pValues);
};
static
void
Copy1D(Npp64fc *pDst, const Npp64fc *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp64fc),cudaMemcpyDeviceToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
HostToDeviceCopy1D(Npp64fc *pDst, const Npp64fc *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp64fc), cudaMemcpyHostToDevice);
NPP_ASSERT(cudaSuccess == eResult);
};
static
void
DeviceToHostCopy1D(Npp64fc *pDst, const Npp64fc *pSrc, size_t nSize)
{
cudaError_t eResult;
eResult = cudaMemcpy(pDst, pSrc, nSize * sizeof(Npp64fc), cudaMemcpyDeviceToHost);
NPP_ASSERT(cudaSuccess == eResult);
};
};
} // npp namespace
#endif // NV_UTIL_NPP_SIGNAL_ALLOCATORS_NPP_H

107
Common/UtilNPP/SignalsCPU.h Normal file
View File

@@ -0,0 +1,107 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#ifndef NV_UTIL_NPP_SIGNALS_CPU_H
#define NV_UTIL_NPP_SIGNALS_CPU_H
#include "Signal.h"
#include "SignalAllocatorsCPU.h"
#include "Exceptions.h"
#include <npp.h>
namespace npp
{
template<typename D, class A>
class SignalCPU: public npp::SignalTemplate<D, A>
{
public:
typedef typename npp::SignalTemplate<D, A>::tData tData;
SignalCPU()
{
;
}
SignalCPU(size_t nSize): SignalTemplate<D, A>(nSize)
{
;
}
SignalCPU(const SignalCPU<D, A> &rSignal): SignalTemplate<D, A>(rSignal)
{
;
}
virtual
~SignalCPU()
{
;
}
SignalCPU &
operator= (const SignalCPU<D,A> &rSignal)
{
SignalTemplate<D, A>::operator= (rSignal);
return *this;
}
tData &
operator [](unsigned int i)
{
return *SignalTemplate<D, A>::values(i);
}
tData
operator [](unsigned int i)
const
{
return *SignalTemplate<D, A>::values(i);
}
};
typedef SignalCPU<Npp8u, npp::SignalAllocatorCPU<Npp8u> > SignalCPU_8u;
typedef SignalCPU<Npp32s, npp::SignalAllocatorCPU<Npp32s> > SignalCPU_32s;
typedef SignalCPU<Npp16s, npp::SignalAllocatorCPU<Npp16s> > SignalCPU_16s;
typedef SignalCPU<Npp16sc, npp::SignalAllocatorCPU<Npp16sc> > SignalCPU_16sc;
typedef SignalCPU<Npp32sc, npp::SignalAllocatorCPU<Npp32sc> > SignalCPU_32sc;
typedef SignalCPU<Npp32f, npp::SignalAllocatorCPU<Npp32f> > SignalCPU_32f;
typedef SignalCPU<Npp32fc, npp::SignalAllocatorCPU<Npp32fc> > SignalCPU_32fc;
typedef SignalCPU<Npp64s, npp::SignalAllocatorCPU<Npp64s> > SignalCPU_64s;
typedef SignalCPU<Npp64sc, npp::SignalAllocatorCPU<Npp64sc> > SignalCPU_64sc;
typedef SignalCPU<Npp64f, npp::SignalAllocatorCPU<Npp64f> > SignalCPU_64f;
typedef SignalCPU<Npp64fc, npp::SignalAllocatorCPU<Npp64fc> > SignalCPU_64fc;
} // npp namespace
#endif // NV_UTIL_NPP_SIGNALS_CPU_H

113
Common/UtilNPP/SignalsNPP.h Normal file
View File

@@ -0,0 +1,113 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#ifndef NV_UTIL_NPP_SIGNALS_NPP_H
#define NV_UTIL_NPP_SIGNALS_NPP_H
#include "Exceptions.h"
#include "Signal.h"
#include "SignalAllocatorsNPP.h"
#include <cuda_runtime.h>
namespace npp
{
// forward declaration
template<typename D, class A> class SignalCPU;
template<typename D>
class SignalNPP: public npp::SignalTemplate<D, npp::SignalAllocator<D> >
{
public:
SignalNPP()
{
;
}
explicit
SignalNPP(size_t nSize): SignalTemplate<D, npp::SignalAllocator<D> >(nSize)
{
;
}
SignalNPP(const SignalNPP<D> &rSignal): SignalTemplate<D, npp::SignalAllocator<D> >(rSignal)
{
;
}
template<class X>
explicit
SignalNPP(const SignalCPU<D, X> &rSignal): SignalTemplate<D, npp::SignalAllocator<D> >(rSignal.size())
{
npp::SignalAllocator<D>::HostToDeviceCopy1D(SignalTemplate<D, npp::SignalAllocator<D> >::values(),
rSignal.values(), SignalTemplate<D, npp::SignalAllocator<D> >::size());
}
virtual
~SignalNPP()
{
;
}
SignalNPP &
operator= (const SignalNPP<D> &rSignal)
{
SignalTemplate<D, npp::SignalAllocator<D> >::operator= (rSignal);
return *this;
}
void
copyTo(D *pValues)
const
{
npp::SignalAllocator<D>::DeviceToHostCopy1D(pValues, SignalTemplate<D, npp::SignalAllocator<D> >::values(), SignalTemplate<D, npp::SignalAllocator<D> >::size());
}
void
copyFrom(D *pValues)
{
npp::SignalAllocator<D>::HostToDeviceCopy1D(SignalTemplate<D, npp::SignalAllocator<D> >::values(), pValues, SignalTemplate<D, npp::SignalAllocator<D> >::size());
}
};
typedef SignalNPP<Npp8u> SignalNPP_8u;
typedef SignalNPP<Npp16s> SignalNPP_16s;
typedef SignalNPP<Npp16sc> SignalNPP_16sc;
typedef SignalNPP<Npp32s> SignalNPP_32s;
typedef SignalNPP<Npp32sc> SignalNPP_32sc;
typedef SignalNPP<Npp32f> SignalNPP_32f;
typedef SignalNPP<Npp32fc> SignalNPP_32fc;
typedef SignalNPP<Npp64s> SignalNPP_64s;
typedef SignalNPP<Npp64sc> SignalNPP_64sc;
typedef SignalNPP<Npp64f> SignalNPP_64f;
typedef SignalNPP<Npp64fc> SignalNPP_64fc;
} // npp namespace
#endif // NV_UTIL_NPP_SIGNALS_NPP_H

160
Common/dynlink_d3d11.h Normal file
View File

@@ -0,0 +1,160 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
//--------------------------------------------------------------------------------------
// File: dynlink_d3d11.h
//
// Shortcut macros and functions for using DX objects
//
// Copyright (c) Microsoft Corporation. All rights reserved
//--------------------------------------------------------------------------------------
#ifndef _DYNLINK_D3D11_H_
#define _DYNLINK_D3D11_H_
// Standard Windows includes
#include <windows.h>
#include <initguid.h>
#include <assert.h>
#include <wchar.h>
#include <mmsystem.h>
#include <commctrl.h> // for InitCommonControls()
#include <shellapi.h> // for ExtractIcon()
#include <new.h> // for placement new
#include <shlobj.h>
#include <math.h>
#include <limits.h>
#include <stdio.h>
// CRT's memory leak detection
#if defined(DEBUG) || defined(_DEBUG)
#include <crtdbg.h>
#endif
// Direct3D10 includes
#include <dxgi.h>
#include <d3d11.h>
// #include <..\Samples\C++\Effects11\Inc\d3dx11effect.h>
// XInput includes
#include <xinput.h>
// strsafe.h deprecates old unsecure string functions. If you
// really do not want to it to (not recommended), then uncomment the next line
//#define STRSAFE_NO_DEPRECATE
#ifndef STRSAFE_NO_DEPRECATE
#pragma deprecated("strncpy")
#pragma deprecated("wcsncpy")
#pragma deprecated("_tcsncpy")
#pragma deprecated("wcsncat")
#pragma deprecated("strncat")
#pragma deprecated("_tcsncat")
#endif
#pragma warning( disable : 4996 ) // disable deprecated warning
#include <strsafe.h>
#pragma warning( default : 4996 )
typedef HRESULT(WINAPI *LPCREATEDXGIFACTORY)(REFIID, void **);
typedef HRESULT(WINAPI *LPD3D11CREATEDEVICEANDSWAPCHAIN)(__in_opt IDXGIAdapter *pAdapter, D3D_DRIVER_TYPE DriverType, HMODULE Software, UINT Flags, __in_ecount_opt(FeatureLevels) CONST D3D_FEATURE_LEVEL *pFeatureLevels, UINT FeatureLevels, UINT SDKVersion, __in_opt CONST DXGI_SWAP_CHAIN_DESC *pSwapChainDesc, __out_opt IDXGISwapChain **ppSwapChain, __out_opt ID3D11Device **ppDevice, __out_opt D3D_FEATURE_LEVEL *pFeatureLevel, __out_opt ID3D11DeviceContext **ppImmediateContext);
typedef HRESULT(WINAPI *LPD3D11CREATEDEVICE)(IDXGIAdapter *, D3D_DRIVER_TYPE, HMODULE, UINT32, D3D_FEATURE_LEVEL *, UINT, UINT32, ID3D11Device **, D3D_FEATURE_LEVEL *, ID3D11DeviceContext **);
static HMODULE s_hModDXGI = NULL;
static LPCREATEDXGIFACTORY sFnPtr_CreateDXGIFactory = NULL;
static HMODULE s_hModD3D11 = NULL;
static LPD3D11CREATEDEVICE sFnPtr_D3D11CreateDevice = NULL;
static LPD3D11CREATEDEVICEANDSWAPCHAIN sFnPtr_D3D11CreateDeviceAndSwapChain = NULL;
// unload the D3D10 DLLs
static bool dynlinkUnloadD3D11API(void)
{
if (s_hModDXGI)
{
FreeLibrary(s_hModDXGI);
s_hModDXGI = NULL;
}
if (s_hModD3D11)
{
FreeLibrary(s_hModD3D11);
s_hModD3D11 = NULL;
}
return true;
}
// Dynamically load the D3D11 DLLs loaded and map the function pointers
static bool dynlinkLoadD3D11API(void)
{
// If both modules are non-NULL, this function has already been called. Note
// that this doesn't guarantee that all ProcAddresses were found.
if (s_hModD3D11 != NULL && s_hModDXGI != NULL)
{
return true;
}
#if 1
// This may fail if Direct3D 11 isn't installed
s_hModD3D11 = LoadLibrary("d3d11.dll");
if (s_hModD3D11 != NULL)
{
sFnPtr_D3D11CreateDevice = (LPD3D11CREATEDEVICE)GetProcAddress(s_hModD3D11, "D3D11CreateDevice");
sFnPtr_D3D11CreateDeviceAndSwapChain = (LPD3D11CREATEDEVICEANDSWAPCHAIN)GetProcAddress(s_hModD3D11, "D3D11CreateDeviceAndSwapChain");
}
else
{
printf("\nLoad d3d11.dll failed\n");
fflush(0);
}
if (!sFnPtr_CreateDXGIFactory)
{
s_hModDXGI = LoadLibrary("dxgi.dll");
if (s_hModDXGI)
{
sFnPtr_CreateDXGIFactory = (LPCREATEDXGIFACTORY)GetProcAddress(s_hModDXGI, "CreateDXGIFactory1");
}
return (s_hModDXGI != NULL) && (s_hModD3D11 != NULL);
}
return (s_hModD3D11 != NULL);
#else
sFnPtr_D3D11CreateDevice = (LPD3D11CREATEDEVICE)D3D11CreateDeviceAndSwapChain;
sFnPtr_D3D11CreateDeviceAndSwapChain = (LPD3D11CREATEDEVICEANDSWAPCHAIN)D3D11CreateDeviceAndSwapChain;
//sFnPtr_D3DX11CreateEffectFromMemory = ( LPD3DX11CREATEEFFECTFROMMEMORY )D3DX11CreateEffectFromMemory;
sFnPtr_D3DX11CompileFromMemory = (LPD3DX11COMPILEFROMMEMORY)D3DX11CompileFromMemory;
sFnPtr_CreateDXGIFactory = (LPCREATEDXGIFACTORY)CreateDXGIFactory;
return true;
#endif
return true;
}
#endif

View File

@@ -579,24 +579,12 @@ static const char *_cudaGetErrorEnum(NppStatus error) {
} }
#endif #endif
#ifdef __DRIVER_TYPES_H__
#ifndef DEVICE_RESET
#define DEVICE_RESET cudaDeviceReset();
#endif
#else
#ifndef DEVICE_RESET
#define DEVICE_RESET
#endif
#endif
template <typename T> template <typename T>
void check(T result, char const *const func, const char *const file, void check(T result, char const *const func, const char *const file,
int const line) { int const line) {
if (result) { if (result) {
fprintf(stderr, "CUDA error at %s:%d code=%d(%s) \"%s\" \n", file, line, fprintf(stderr, "CUDA error at %s:%d code=%d(%s) \"%s\" \n", file, line,
static_cast<unsigned int>(result), _cudaGetErrorEnum(result), func); static_cast<unsigned int>(result), _cudaGetErrorEnum(result), func);
DEVICE_RESET
// Make sure we call CUDA Device Reset before exiting
exit(EXIT_FAILURE); exit(EXIT_FAILURE);
} }
} }
@@ -619,7 +607,6 @@ inline void __getLastCudaError(const char *errorMessage, const char *file,
" %s : (%d) %s.\n", " %s : (%d) %s.\n",
file, line, errorMessage, static_cast<int>(err), file, line, errorMessage, static_cast<int>(err),
cudaGetErrorString(err)); cudaGetErrorString(err));
DEVICE_RESET
exit(EXIT_FAILURE); exit(EXIT_FAILURE);
} }
} }
@@ -676,6 +663,7 @@ inline int _ConvertSMVer2Cores(int major, int minor) {
{0x70, 64}, {0x70, 64},
{0x72, 64}, {0x72, 64},
{0x75, 64}, {0x75, 64},
{0x80, 64},
{-1, -1}}; {-1, -1}};
int index = 0; int index = 0;
@@ -696,6 +684,51 @@ inline int _ConvertSMVer2Cores(int major, int minor) {
major, minor, nGpuArchCoresPerSM[index - 1].Cores); major, minor, nGpuArchCoresPerSM[index - 1].Cores);
return nGpuArchCoresPerSM[index - 1].Cores; return nGpuArchCoresPerSM[index - 1].Cores;
} }
inline const char* _ConvertSMVer2ArchName(int major, int minor) {
// Defines for GPU Architecture types (using the SM version to determine
// the GPU Arch name)
typedef struct {
int SM; // 0xMm (hexidecimal notation), M = SM Major version,
// and m = SM minor version
const char* name;
} sSMtoArchName;
sSMtoArchName nGpuArchNameSM[] = {
{0x30, "Kepler"},
{0x32, "Kepler"},
{0x35, "Kepler"},
{0x37, "Kepler"},
{0x50, "Maxwell"},
{0x52, "Maxwell"},
{0x53, "Maxwell"},
{0x60, "Pascal"},
{0x61, "Pascal"},
{0x62, "Pascal"},
{0x70, "Volta"},
{0x72, "Xavier"},
{0x75, "Turing"},
{0x80, "Ampere"},
{-1, "Graphics Device"}};
int index = 0;
while (nGpuArchNameSM[index].SM != -1) {
if (nGpuArchNameSM[index].SM == ((major << 4) + minor)) {
return nGpuArchNameSM[index].name;
}
index++;
}
// If we don't find the values, we default use the previous one
// to run properly
printf(
"MapSMtoArchName for SM %d.%d is undefined."
" Default to use %s\n",
major, minor, nGpuArchNameSM[index - 1].name);
return nGpuArchNameSM[index - 1].name;
}
// end of GPU Architecture definitions // end of GPU Architecture definitions
#ifdef __CUDA_RUNTIME_H__ #ifdef __CUDA_RUNTIME_H__
@@ -727,23 +760,24 @@ inline int gpuDeviceInit(int devID) {
return -devID; return -devID;
} }
cudaDeviceProp deviceProp; int computeMode = -1, major = 0, minor = 0;
checkCudaErrors(cudaGetDeviceProperties(&deviceProp, devID)); checkCudaErrors(cudaDeviceGetAttribute(&computeMode, cudaDevAttrComputeMode, devID));
checkCudaErrors(cudaDeviceGetAttribute(&major, cudaDevAttrComputeCapabilityMajor, devID));
if (deviceProp.computeMode == cudaComputeModeProhibited) { checkCudaErrors(cudaDeviceGetAttribute(&minor, cudaDevAttrComputeCapabilityMinor, devID));
if (computeMode == cudaComputeModeProhibited) {
fprintf(stderr, fprintf(stderr,
"Error: device is running in <Compute Mode " "Error: device is running in <Compute Mode "
"Prohibited>, no threads can use cudaSetDevice().\n"); "Prohibited>, no threads can use cudaSetDevice().\n");
return -1; return -1;
} }
if (deviceProp.major < 1) { if (major < 1) {
fprintf(stderr, "gpuDeviceInit(): GPU device does not support CUDA.\n"); fprintf(stderr, "gpuDeviceInit(): GPU device does not support CUDA.\n");
exit(EXIT_FAILURE); exit(EXIT_FAILURE);
} }
checkCudaErrors(cudaSetDevice(devID)); checkCudaErrors(cudaSetDevice(devID));
printf("gpuDeviceInit() CUDA Device [%d]: \"%s\n", devID, deviceProp.name); printf("gpuDeviceInit() CUDA Device [%d]: \"%s\n", devID, _ConvertSMVer2ArchName(major, minor));
return devID; return devID;
} }
@@ -756,7 +790,6 @@ inline int gpuGetMaxGflopsDeviceId() {
int devices_prohibited = 0; int devices_prohibited = 0;
uint64_t max_compute_perf = 0; uint64_t max_compute_perf = 0;
cudaDeviceProp deviceProp;
checkCudaErrors(cudaGetDeviceCount(&device_count)); checkCudaErrors(cudaGetDeviceCount(&device_count));
if (device_count == 0) { if (device_count == 0) {
@@ -770,20 +803,36 @@ inline int gpuGetMaxGflopsDeviceId() {
current_device = 0; current_device = 0;
while (current_device < device_count) { while (current_device < device_count) {
cudaGetDeviceProperties(&deviceProp, current_device); int computeMode = -1, major = 0, minor = 0;
checkCudaErrors(cudaDeviceGetAttribute(&computeMode, cudaDevAttrComputeMode, current_device));
checkCudaErrors(cudaDeviceGetAttribute(&major, cudaDevAttrComputeCapabilityMajor, current_device));
checkCudaErrors(cudaDeviceGetAttribute(&minor, cudaDevAttrComputeCapabilityMinor, current_device));
// If this GPU is not running on Compute Mode prohibited, // If this GPU is not running on Compute Mode prohibited,
// then we can add it to the list // then we can add it to the list
if (deviceProp.computeMode != cudaComputeModeProhibited) { if (computeMode != cudaComputeModeProhibited) {
if (deviceProp.major == 9999 && deviceProp.minor == 9999) { if (major == 9999 && minor == 9999) {
sm_per_multiproc = 1; sm_per_multiproc = 1;
} else { } else {
sm_per_multiproc = sm_per_multiproc =
_ConvertSMVer2Cores(deviceProp.major, deviceProp.minor); _ConvertSMVer2Cores(major, minor);
} }
int multiProcessorCount = 0, clockRate = 0;
uint64_t compute_perf = (uint64_t)deviceProp.multiProcessorCount * checkCudaErrors(cudaDeviceGetAttribute(&multiProcessorCount, cudaDevAttrMultiProcessorCount, current_device));
sm_per_multiproc * deviceProp.clockRate; cudaError_t result = cudaDeviceGetAttribute(&clockRate, cudaDevAttrClockRate, current_device);
if (result != cudaSuccess) {
// If cudaDevAttrClockRate attribute is not supported we
// set clockRate as 1, to consider GPU with most SMs and CUDA Cores.
if(result == cudaErrorInvalidValue) {
clockRate = 1;
}
else {
fprintf(stderr, "CUDA error at %s:%d code=%d(%s) \n", __FILE__, __LINE__,
static_cast<unsigned int>(result), _cudaGetErrorEnum(result));
exit(EXIT_FAILURE);
}
}
uint64_t compute_perf = (uint64_t)multiProcessorCount * sm_per_multiproc * clockRate;
if (compute_perf > max_compute_perf) { if (compute_perf > max_compute_perf) {
max_compute_perf = compute_perf; max_compute_perf = compute_perf;
@@ -808,7 +857,6 @@ inline int gpuGetMaxGflopsDeviceId() {
// Initialization code to find the best CUDA Device // Initialization code to find the best CUDA Device
inline int findCudaDevice(int argc, const char **argv) { inline int findCudaDevice(int argc, const char **argv) {
cudaDeviceProp deviceProp;
int devID = 0; int devID = 0;
// If the command-line has a device number specified, use it // If the command-line has a device number specified, use it
@@ -830,9 +878,12 @@ inline int findCudaDevice(int argc, const char **argv) {
// Otherwise pick the device with highest Gflops/s // Otherwise pick the device with highest Gflops/s
devID = gpuGetMaxGflopsDeviceId(); devID = gpuGetMaxGflopsDeviceId();
checkCudaErrors(cudaSetDevice(devID)); checkCudaErrors(cudaSetDevice(devID));
checkCudaErrors(cudaGetDeviceProperties(&deviceProp, devID)); int major = 0, minor = 0;
printf("GPU Device %d: \"%s\" with compute capability %d.%d\n\n", devID, checkCudaErrors(cudaDeviceGetAttribute(&major, cudaDevAttrComputeCapabilityMajor, devID));
deviceProp.name, deviceProp.major, deviceProp.minor); checkCudaErrors(cudaDeviceGetAttribute(&minor, cudaDevAttrComputeCapabilityMinor, devID));
printf("GPU Device %d: \"%s\" with compute capability %d.%d\n\n",
devID, _ConvertSMVer2ArchName(major, minor), major, minor);
} }
return devID; return devID;
@@ -843,7 +894,6 @@ inline int findIntegratedGPU() {
int device_count = 0; int device_count = 0;
int devices_prohibited = 0; int devices_prohibited = 0;
cudaDeviceProp deviceProp;
checkCudaErrors(cudaGetDeviceCount(&device_count)); checkCudaErrors(cudaGetDeviceCount(&device_count));
if (device_count == 0) { if (device_count == 0) {
@@ -853,17 +903,19 @@ inline int findIntegratedGPU() {
// Find the integrated GPU which is compute capable // Find the integrated GPU which is compute capable
while (current_device < device_count) { while (current_device < device_count) {
cudaGetDeviceProperties(&deviceProp, current_device); int computeMode = -1, integrated = -1;
checkCudaErrors(cudaDeviceGetAttribute(&computeMode, cudaDevAttrComputeMode, current_device));
checkCudaErrors(cudaDeviceGetAttribute(&integrated, cudaDevAttrIntegrated, current_device));
// If GPU is integrated and is not running on Compute Mode prohibited, // If GPU is integrated and is not running on Compute Mode prohibited,
// then cuda can map to GLES resource // then cuda can map to GLES resource
if (deviceProp.integrated && if (integrated && (computeMode != cudaComputeModeProhibited)) {
(deviceProp.computeMode != cudaComputeModeProhibited)) {
checkCudaErrors(cudaSetDevice(current_device)); checkCudaErrors(cudaSetDevice(current_device));
checkCudaErrors(cudaGetDeviceProperties(&deviceProp, current_device));
int major = 0, minor = 0;
checkCudaErrors(cudaDeviceGetAttribute(&major, cudaDevAttrComputeCapabilityMajor, current_device));
checkCudaErrors(cudaDeviceGetAttribute(&minor, cudaDevAttrComputeCapabilityMinor, current_device));
printf("GPU Device %d: \"%s\" with compute capability %d.%d\n\n", printf("GPU Device %d: \"%s\" with compute capability %d.%d\n\n",
current_device, deviceProp.name, deviceProp.major, current_device, _ConvertSMVer2ArchName(major, minor), major, minor);
deviceProp.minor);
return current_device; return current_device;
} else { } else {
@@ -885,19 +937,18 @@ inline int findIntegratedGPU() {
// General check for CUDA GPU SM Capabilities // General check for CUDA GPU SM Capabilities
inline bool checkCudaCapabilities(int major_version, int minor_version) { inline bool checkCudaCapabilities(int major_version, int minor_version) {
cudaDeviceProp deviceProp;
deviceProp.major = 0;
deviceProp.minor = 0;
int dev; int dev;
int major = 0, minor = 0;
checkCudaErrors(cudaGetDevice(&dev)); checkCudaErrors(cudaGetDevice(&dev));
checkCudaErrors(cudaGetDeviceProperties(&deviceProp, dev)); checkCudaErrors(cudaDeviceGetAttribute(&major, cudaDevAttrComputeCapabilityMajor, dev));
checkCudaErrors(cudaDeviceGetAttribute(&minor, cudaDevAttrComputeCapabilityMinor, dev));
if ((deviceProp.major > major_version) || if ((major > major_version) ||
(deviceProp.major == major_version && (major == major_version &&
deviceProp.minor >= minor_version)) { minor >= minor_version)) {
printf(" Device %d: <%16s >, Compute SM %d.%d detected\n", dev, printf(" Device %d: <%16s >, Compute SM %d.%d detected\n", dev,
deviceProp.name, deviceProp.major, deviceProp.minor); _ConvertSMVer2ArchName(major, minor), major, minor);
return true; return true;
} else { } else {
printf( printf(

1469
Common/helper_math.h Normal file

File diff suppressed because it is too large Load Diff

View File

@@ -29,150 +29,515 @@
#include <cstdlib> #include <cstdlib>
#include <string> #include <string>
int sharedMemoryCreate(const char *name, size_t sz, sharedMemoryInfo *info) int sharedMemoryCreate(const char *name, size_t sz, sharedMemoryInfo *info) {
{
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64) #if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
info->size = sz; info->size = sz;
info->shmHandle = CreateFileMapping(INVALID_HANDLE_VALUE, info->shmHandle = CreateFileMapping(INVALID_HANDLE_VALUE, NULL,
NULL, PAGE_READWRITE, 0, (DWORD)sz, name);
PAGE_READWRITE, if (info->shmHandle == 0) {
0, return GetLastError();
(DWORD)sz, }
name);
if (info->shmHandle == 0) {
return GetLastError();
}
info->addr = MapViewOfFile(info->shmHandle, FILE_MAP_ALL_ACCESS, 0, 0, sz); info->addr = MapViewOfFile(info->shmHandle, FILE_MAP_ALL_ACCESS, 0, 0, sz);
if (info->addr == NULL) { if (info->addr == NULL) {
return GetLastError(); return GetLastError();
} }
return 0; return 0;
#else #else
int status = 0; int status = 0;
info->size = sz; info->size = sz;
info->shmFd = shm_open(name, O_RDWR | O_CREAT, 0777); info->shmFd = shm_open(name, O_RDWR | O_CREAT, 0777);
if (info->shmFd < 0) { if (info->shmFd < 0) {
return errno; return errno;
} }
status = ftruncate(info->shmFd, sz); status = ftruncate(info->shmFd, sz);
if (status != 0) { if (status != 0) {
return status; return status;
} }
info->addr = mmap(0, sz, PROT_READ | PROT_WRITE, MAP_SHARED, info->shmFd, 0); info->addr = mmap(0, sz, PROT_READ | PROT_WRITE, MAP_SHARED, info->shmFd, 0);
if (info->addr == NULL) { if (info->addr == NULL) {
return errno; return errno;
} }
return 0; return 0;
#endif #endif
} }
int sharedMemoryOpen(const char *name, size_t sz, sharedMemoryInfo *info) int sharedMemoryOpen(const char *name, size_t sz, sharedMemoryInfo *info) {
{
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64) #if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
info->size = sz; info->size = sz;
info->shmHandle = OpenFileMapping(FILE_MAP_ALL_ACCESS, FALSE, name); info->shmHandle = OpenFileMapping(FILE_MAP_ALL_ACCESS, FALSE, name);
if (info->shmHandle == 0) { if (info->shmHandle == 0) {
return GetLastError(); return GetLastError();
} }
info->addr = MapViewOfFile(info->shmHandle, FILE_MAP_ALL_ACCESS, 0, 0, sz); info->addr = MapViewOfFile(info->shmHandle, FILE_MAP_ALL_ACCESS, 0, 0, sz);
if (info->addr == NULL) { if (info->addr == NULL) {
return GetLastError(); return GetLastError();
} }
return 0; return 0;
#else #else
info->size = sz; info->size = sz;
info->shmFd = shm_open(name, O_RDWR, 0777); info->shmFd = shm_open(name, O_RDWR, 0777);
if (info->shmFd < 0) { if (info->shmFd < 0) {
return errno; return errno;
} }
info->addr = mmap(0, sz, PROT_READ | PROT_WRITE, MAP_SHARED, info->shmFd, 0); info->addr = mmap(0, sz, PROT_READ | PROT_WRITE, MAP_SHARED, info->shmFd, 0);
if (info->addr == NULL) { if (info->addr == NULL) {
return errno; return errno;
} }
return 0; return 0;
#endif #endif
} }
void sharedMemoryClose(sharedMemoryInfo *info) void sharedMemoryClose(sharedMemoryInfo *info) {
{
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64) #if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
if (info->addr) { if (info->addr) {
UnmapViewOfFile(info->addr); UnmapViewOfFile(info->addr);
} }
if (info->shmHandle) { if (info->shmHandle) {
CloseHandle(info->shmHandle); CloseHandle(info->shmHandle);
} }
#else #else
if (info->addr) { if (info->addr) {
munmap(info->addr, info->size); munmap(info->addr, info->size);
} }
if (info->shmFd) { if (info->shmFd) {
close(info->shmFd); close(info->shmFd);
} }
#endif #endif
} }
int spawnProcess(Process *process, const char *app, char * const *args) int spawnProcess(Process *process, const char *app, char *const *args) {
{
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64) #if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
STARTUPINFO si = {0}; STARTUPINFO si = {0};
BOOL status; BOOL status;
size_t arglen = 0; size_t arglen = 0;
size_t argIdx = 0; size_t argIdx = 0;
std::string arg_string; std::string arg_string;
memset(process, 0, sizeof(*process)); memset(process, 0, sizeof(*process));
while (*args) { while (*args) {
arg_string.append(*args).append(1, ' '); arg_string.append(*args).append(1, ' ');
args++; args++;
} }
status = CreateProcess(app, LPSTR(arg_string.c_str()), NULL, NULL, FALSE, 0, NULL, NULL, &si, process); status = CreateProcess(app, LPSTR(arg_string.c_str()), NULL, NULL, FALSE, 0,
NULL, NULL, &si, process);
return status ? 0 : GetLastError(); return status ? 0 : GetLastError();
#else #else
*process = fork(); *process = fork();
if (*process == 0) { if (*process == 0) {
if (0 > execvp(app, args)) { if (0 > execvp(app, args)) {
return errno; return errno;
}
} }
else if (*process < 0) { } else if (*process < 0) {
return errno; return errno;
} }
return 0; return 0;
#endif #endif
} }
int waitProcess(Process *process) int waitProcess(Process *process) {
{
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64) #if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
DWORD exitCode; DWORD exitCode;
WaitForSingleObject(process->hProcess, INFINITE); WaitForSingleObject(process->hProcess, INFINITE);
GetExitCodeProcess(process->hProcess, &exitCode); GetExitCodeProcess(process->hProcess, &exitCode);
CloseHandle(process->hProcess); CloseHandle(process->hProcess);
CloseHandle(process->hThread); CloseHandle(process->hThread);
return (int)exitCode; return (int)exitCode;
#else #else
int status = 0; int status = 0;
do { do {
if (0 > waitpid(*process, &status, 0)) { if (0 > waitpid(*process, &status, 0)) {
return errno; return errno;
} }
} while (!WIFEXITED(status)); } while (!WIFEXITED(status));
return WEXITSTATUS(status); return WEXITSTATUS(status);
#endif #endif
} }
#if defined(__linux__)
int ipcCreateSocket(ipcHandle *&handle, const char *name,
const std::vector<Process> &processes) {
int server_fd;
struct sockaddr_un servaddr;
handle = new ipcHandle;
memset(handle, 0, sizeof(*handle));
handle->socket = -1;
handle->socketName = NULL;
// Creating socket file descriptor
if ((server_fd = socket(AF_UNIX, SOCK_DGRAM, 0)) == 0) {
perror("IPC failure: Socket creation failed");
return -1;
}
unlink(name);
bzero(&servaddr, sizeof(servaddr));
servaddr.sun_family = AF_UNIX;
size_t len = strlen(name);
if (len > (sizeof(servaddr.sun_path) - 1)) {
perror("IPC failure: Cannot bind provided name to socket. Name too large");
return -1;
}
strncpy(servaddr.sun_path, name, len);
if (bind(server_fd, (struct sockaddr *)&servaddr, SUN_LEN(&servaddr)) < 0) {
perror("IPC failure: Binding socket failed");
return -1;
}
handle->socketName = new char[strlen(name) + 1];
strcpy(handle->socketName, name);
handle->socket = server_fd;
return 0;
}
int ipcOpenSocket(ipcHandle *&handle) {
int sock = 0;
struct sockaddr_un cliaddr;
handle = new ipcHandle;
memset(handle, 0, sizeof(*handle));
if ((sock = socket(AF_UNIX, SOCK_DGRAM, 0)) < 0) {
perror("IPC failure:Socket creation error");
return -1;
}
bzero(&cliaddr, sizeof(cliaddr));
cliaddr.sun_family = AF_UNIX;
char temp[10];
// Create unique name for the socket.
sprintf(temp, "%u", getpid());
strcpy(cliaddr.sun_path, temp);
if (bind(sock, (struct sockaddr *)&cliaddr, sizeof(cliaddr)) < 0) {
perror("IPC failure: Binding socket failed");
return -1;
}
handle->socket = sock;
handle->socketName = new char[strlen(temp) + 1];
strcpy(handle->socketName, temp);
return 0;
}
int ipcCloseSocket(ipcHandle *handle) {
if (!handle) {
return -1;
}
if (handle->socketName) {
unlink(handle->socketName);
delete[] handle->socketName;
}
close(handle->socket);
delete handle;
return 0;
}
int ipcRecvShareableHandle(ipcHandle *handle, ShareableHandle *shHandle) {
struct msghdr msg = {0};
struct iovec iov[1];
struct cmsghdr cm;
// Union to guarantee alignment requirements for control array
union {
struct cmsghdr cm;
char control[CMSG_SPACE(sizeof(int))];
} control_un;
struct cmsghdr *cmptr;
ssize_t n;
int receivedfd;
char dummy_buffer[1];
ssize_t sendResult;
msg.msg_control = control_un.control;
msg.msg_controllen = sizeof(control_un.control);
iov[0].iov_base = (void *)dummy_buffer;
iov[0].iov_len = sizeof(dummy_buffer);
msg.msg_iov = iov;
msg.msg_iovlen = 1;
if ((n = recvmsg(handle->socket, &msg, 0)) <= 0) {
perror("IPC failure: Receiving data over socket failed");
return -1;
}
if (((cmptr = CMSG_FIRSTHDR(&msg)) != NULL) &&
(cmptr->cmsg_len == CMSG_LEN(sizeof(int)))) {
if ((cmptr->cmsg_level != SOL_SOCKET) || (cmptr->cmsg_type != SCM_RIGHTS)) {
return -1;
}
memmove(&receivedfd, CMSG_DATA(cmptr), sizeof(receivedfd));
*(int *)shHandle = receivedfd;
} else {
return -1;
}
return 0;
}
int ipcRecvDataFromClient(ipcHandle *serverHandle, void *data, size_t size) {
ssize_t readResult;
struct sockaddr_un cliaddr;
socklen_t len = sizeof(cliaddr);
readResult = recvfrom(serverHandle->socket, data, size, 0,
(struct sockaddr *)&cliaddr, &len);
if (readResult == -1) {
perror("IPC failure: Receiving data over socket failed");
return -1;
}
return 0;
}
int ipcSendDataToServer(ipcHandle *handle, const char *serverName,
const void *data, size_t size) {
ssize_t sendResult;
struct sockaddr_un serveraddr;
bzero(&serveraddr, sizeof(serveraddr));
serveraddr.sun_family = AF_UNIX;
strncpy(serveraddr.sun_path, serverName, sizeof(serveraddr.sun_path) - 1);
sendResult = sendto(handle->socket, data, size, 0,
(struct sockaddr *)&serveraddr, sizeof(serveraddr));
if (sendResult <= 0) {
perror("IPC failure: Sending data over socket failed");
}
return 0;
}
int ipcSendShareableHandle(ipcHandle *handle,
const std::vector<ShareableHandle> &shareableHandles,
Process process, int data) {
struct msghdr msg;
struct iovec iov[1];
union {
struct cmsghdr cm;
char control[CMSG_SPACE(sizeof(int))];
} control_un;
struct cmsghdr *cmptr;
ssize_t readResult;
struct sockaddr_un cliaddr;
socklen_t len = sizeof(cliaddr);
// Construct client address to send this SHareable handle to
bzero(&cliaddr, sizeof(cliaddr));
cliaddr.sun_family = AF_UNIX;
char temp[10];
sprintf(temp, "%u", process);
strcpy(cliaddr.sun_path, temp);
len = sizeof(cliaddr);
// Send corresponding shareable handle to the client
int sendfd = (int)shareableHandles[data];
msg.msg_control = control_un.control;
msg.msg_controllen = sizeof(control_un.control);
cmptr = CMSG_FIRSTHDR(&msg);
cmptr->cmsg_len = CMSG_LEN(sizeof(int));
cmptr->cmsg_level = SOL_SOCKET;
cmptr->cmsg_type = SCM_RIGHTS;
memmove(CMSG_DATA(cmptr), &sendfd, sizeof(sendfd));
msg.msg_name = (void *)&cliaddr;
msg.msg_namelen = sizeof(struct sockaddr_un);
iov[0].iov_base = (void *)"";
iov[0].iov_len = 1;
msg.msg_iov = iov;
msg.msg_iovlen = 1;
ssize_t sendResult = sendmsg(handle->socket, &msg, 0);
if (sendResult <= 0) {
perror("IPC failure: Sending data over socket failed");
return -1;
}
return 0;
}
int ipcSendShareableHandles(
ipcHandle *handle, const std::vector<ShareableHandle> &shareableHandles,
const std::vector<Process> &processes) {
// Send all shareable handles to every single process.
for (int i = 0; i < shareableHandles.size(); i++) {
for (int j = 0; j < processes.size(); j++) {
checkIpcErrors(
ipcSendShareableHandle(handle, shareableHandles, processes[j], i));
}
}
return 0;
}
int ipcRecvShareableHandles(ipcHandle *handle,
std::vector<ShareableHandle> &shareableHandles) {
for (int i = 0; i < shareableHandles.size(); i++) {
checkIpcErrors(ipcRecvShareableHandle(handle, &shareableHandles[i]));
}
return 0;
}
int ipcCloseShareableHandle(ShareableHandle shHandle) {
return close(shHandle);
}
#elif defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
// Generic name to build individual Mailslot names by appending process ids.
LPTSTR SlotName = (LPTSTR)TEXT("\\\\.\\mailslot\\sample_mailslot_");
int ipcCreateSocket(ipcHandle *&handle, const char *name,
const std::vector<Process> &processes) {
handle = new ipcHandle;
handle->hMailslot.resize(processes.size());
// Open Mailslots of all clients and store respective handles.
for (int i = 0; i < handle->hMailslot.size(); ++i) {
std::basic_string<TCHAR> childSlotName(SlotName);
char tempBuf[20];
_itoa_s(processes[i].dwProcessId, tempBuf, 10);
childSlotName += TEXT(tempBuf);
HANDLE hFile =
CreateFile(TEXT(childSlotName.c_str()), GENERIC_WRITE, FILE_SHARE_READ,
(LPSECURITY_ATTRIBUTES)NULL, OPEN_EXISTING,
FILE_ATTRIBUTE_NORMAL, (HANDLE)NULL);
if (hFile == INVALID_HANDLE_VALUE) {
printf("IPC failure: Opening Mailslot by CreateFile failed with %d\n",
GetLastError());
return -1;
}
handle->hMailslot[i] = hFile;
}
return 0;
}
int ipcOpenSocket(ipcHandle *&handle) {
handle = new ipcHandle;
HANDLE hSlot;
std::basic_string<TCHAR> clientSlotName(SlotName);
char tempBuf[20];
_itoa_s(GetCurrentProcessId(), tempBuf, 10);
clientSlotName += TEXT(tempBuf);
hSlot = CreateMailslot((LPSTR)clientSlotName.c_str(), 0,
MAILSLOT_WAIT_FOREVER, (LPSECURITY_ATTRIBUTES)NULL);
if (hSlot == INVALID_HANDLE_VALUE) {
printf("IPC failure: CreateMailslot failed for client with %d\n",
GetLastError());
return -1;
}
handle->hMailslot.push_back(hSlot);
return 0;
}
int ipcSendData(HANDLE mailslot, const void *data, size_t sz) {
BOOL result;
DWORD cbWritten;
result = WriteFile(mailslot, data, (DWORD)sz, &cbWritten, (LPOVERLAPPED)NULL);
if (!result) {
printf("IPC failure: WriteFile failed with %d.\n", GetLastError());
return -1;
}
return 0;
}
int ipcRecvData(ipcHandle *handle, void *data, size_t sz) {
DWORD cbRead = 0;
if (!ReadFile(handle->hMailslot[0], data, (DWORD)sz, &cbRead, NULL)) {
printf("IPC failure: ReadFile failed with %d.\n", GetLastError());
return -1;
}
if (sz != (size_t)cbRead) {
printf(
"IPC failure: ReadFile didn't receive the expected number of bytes\n");
return -1;
}
return 0;
}
int ipcSendShareableHandles(
ipcHandle *handle, const std::vector<ShareableHandle> &shareableHandles,
const std::vector<Process> &processes) {
// Send all shareable handles to every single process.
for (int i = 0; i < processes.size(); i++) {
HANDLE hProcess =
OpenProcess(PROCESS_DUP_HANDLE, FALSE, processes[i].dwProcessId);
if (hProcess == INVALID_HANDLE_VALUE) {
printf("IPC failure: OpenProcess failed (%d)\n", GetLastError());
return -1;
}
for (int j = 0; j < shareableHandles.size(); j++) {
HANDLE hDup = INVALID_HANDLE_VALUE;
// Duplicate the handle into the target process's space
if (!DuplicateHandle(GetCurrentProcess(), shareableHandles[j], hProcess,
&hDup, 0, FALSE, DUPLICATE_SAME_ACCESS)) {
printf("IPC failure: DuplicateHandle failed (%d)\n", GetLastError());
return -1;
}
checkIpcErrors(ipcSendData(handle->hMailslot[i], &hDup, sizeof(hDup)));
}
CloseHandle(hProcess);
}
return 0;
}
int ipcRecvShareableHandles(ipcHandle *handle,
std::vector<ShareableHandle> &shareableHandles) {
for (int i = 0; i < shareableHandles.size(); i++) {
checkIpcErrors(
ipcRecvData(handle, &shareableHandles[i], sizeof(shareableHandles[i])));
}
return 0;
}
int ipcCloseSocket(ipcHandle *handle) {
for (int i = 0; i < handle->hMailslot.size(); i++) {
CloseHandle(handle->hMailslot[i]);
}
delete handle;
return 0;
}
int ipcCloseShareableHandle(ShareableHandle shHandle) {
CloseHandle(shHandle);
return 0;
}
#endif

View File

@@ -33,13 +33,26 @@
#define WIN32_LEAN_AND_MEAN #define WIN32_LEAN_AND_MEAN
#endif #endif
#include <windows.h> #include <windows.h>
#include <iostream>
#include <stdio.h>
#include <tchar.h>
#include <strsafe.h>
#include <sddl.h>
#include <aclapi.h>
#include <winternl.h>
#else #else
#include <stdio.h>
#include <fcntl.h> #include <fcntl.h>
#include <sys/mman.h> #include <sys/mman.h>
#include <unistd.h> #include <unistd.h>
#include <errno.h> #include <errno.h>
#include <sys/wait.h> #include <sys/wait.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <memory.h>
#include <sys/un.h>
#endif #endif
#include <vector>
typedef struct sharedMemoryInfo_st { typedef struct sharedMemoryInfo_st {
void *addr; void *addr;
@@ -68,4 +81,40 @@ int spawnProcess(Process *process, const char *app, char * const *args);
int waitProcess(Process *process); int waitProcess(Process *process);
#define checkIpcErrors(ipcFuncResult) \
if (ipcFuncResult == -1) { fprintf(stderr, "Failure at %u %s\n", __LINE__, __FILE__); exit(EXIT_FAILURE); }
#if defined(__linux__)
struct ipcHandle_st {
int socket;
char *socketName;
};
typedef int ShareableHandle;
#elif defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
struct ipcHandle_st {
std::vector<HANDLE> hMailslot; // 1 Handle in case of child and `num children` Handles for parent.
};
typedef HANDLE ShareableHandle;
#endif
typedef struct ipcHandle_st ipcHandle;
int
ipcCreateSocket(ipcHandle *&handle, const char *name, const std::vector<Process>& processes);
int
ipcOpenSocket(ipcHandle *&handle);
int
ipcCloseSocket(ipcHandle *handle);
int
ipcRecvShareableHandles(ipcHandle *handle, std::vector<ShareableHandle>& shareableHandles);
int
ipcSendShareableHandles(ipcHandle *handle, const std::vector<ShareableHandle>& shareableHandles, const std::vector<Process>& processes);
int
ipcCloseShareableHandle(ShareableHandle shHandle);
#endif // HELPER_MULTIPROCESS_H #endif // HELPER_MULTIPROCESS_H

428
Common/helper_nvJPEG.hxx Normal file
View File

@@ -0,0 +1,428 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
// This sample needs at least CUDA 10.1.
// It demonstrates usages of the nvJPEG library
#ifndef NV_JPEG_EXAMPLE
#define NV_JPEG_EXAMPLE
#ifdef _WIN64
#include <windows.h>
#endif
#include "cuda_runtime.h"
#include "nvjpeg.h"
#include "helper_cuda.h"
#include "helper_timer.h"
#include <cstdlib>
#include <fstream>
#include <iostream>
#include <sstream>
#include <string>
#include <vector>
#include <string.h> // strcmpi
#ifndef _WIN64
#include <sys/time.h> // timings
#include <dirent.h> // linux dir traverse
#include <unistd.h>
#endif
#include <sys/stat.h>
#include <sys/types.h>
// write bmp, input - RGB, device
int writeBMP(const char *filename, const unsigned char *d_chanR, int pitchR,
const unsigned char *d_chanG, int pitchG,
const unsigned char *d_chanB, int pitchB, int width, int height) {
unsigned int headers[13];
FILE *outfile;
int extrabytes;
int paddedsize;
int x;
int y;
int n;
int red, green, blue;
std::vector<unsigned char> vchanR(height * width);
std::vector<unsigned char> vchanG(height * width);
std::vector<unsigned char> vchanB(height * width);
unsigned char *chanR = vchanR.data();
unsigned char *chanG = vchanG.data();
unsigned char *chanB = vchanB.data();
checkCudaErrors(cudaMemcpy2D(chanR, (size_t)width, d_chanR, (size_t)pitchR,
width, height, cudaMemcpyDeviceToHost));
checkCudaErrors(cudaMemcpy2D(chanG, (size_t)width, d_chanG, (size_t)pitchR,
width, height, cudaMemcpyDeviceToHost));
checkCudaErrors(cudaMemcpy2D(chanB, (size_t)width, d_chanB, (size_t)pitchR,
width, height, cudaMemcpyDeviceToHost));
extrabytes =
4 - ((width * 3) % 4); // How many bytes of padding to add to each
// horizontal line - the size of which must
// be a multiple of 4 bytes.
if (extrabytes == 4) extrabytes = 0;
paddedsize = ((width * 3) + extrabytes) * height;
// Headers...
// Note that the "BM" identifier in bytes 0 and 1 is NOT included in these
// "headers".
headers[0] = paddedsize + 54; // bfSize (whole file size)
headers[1] = 0; // bfReserved (both)
headers[2] = 54; // bfOffbits
headers[3] = 40; // biSize
headers[4] = width; // biWidth
headers[5] = height; // biHeight
// Would have biPlanes and biBitCount in position 6, but they're shorts.
// It's easier to write them out separately (see below) than pretend
// they're a single int, especially with endian issues...
headers[7] = 0; // biCompression
headers[8] = paddedsize; // biSizeImage
headers[9] = 0; // biXPelsPerMeter
headers[10] = 0; // biYPelsPerMeter
headers[11] = 0; // biClrUsed
headers[12] = 0; // biClrImportant
if (!(outfile = fopen(filename, "wb"))) {
std::cerr << "Cannot open file: " << filename << std::endl;
return 1;
}
//
// Headers begin...
// When printing ints and shorts, we write out 1 character at a time to avoid
// endian issues.
//
fprintf(outfile, "BM");
for (n = 0; n <= 5; n++) {
fprintf(outfile, "%c", headers[n] & 0x000000FF);
fprintf(outfile, "%c", (headers[n] & 0x0000FF00) >> 8);
fprintf(outfile, "%c", (headers[n] & 0x00FF0000) >> 16);
fprintf(outfile, "%c", (headers[n] & (unsigned int)0xFF000000) >> 24);
}
// These next 4 characters are for the biPlanes and biBitCount fields.
fprintf(outfile, "%c", 1);
fprintf(outfile, "%c", 0);
fprintf(outfile, "%c", 24);
fprintf(outfile, "%c", 0);
for (n = 7; n <= 12; n++) {
fprintf(outfile, "%c", headers[n] & 0x000000FF);
fprintf(outfile, "%c", (headers[n] & 0x0000FF00) >> 8);
fprintf(outfile, "%c", (headers[n] & 0x00FF0000) >> 16);
fprintf(outfile, "%c", (headers[n] & (unsigned int)0xFF000000) >> 24);
}
//
// Headers done, now write the data...
//
for (y = height - 1; y >= 0;
y--) // BMP image format is written from bottom to top...
{
for (x = 0; x <= width - 1; x++) {
red = chanR[y * width + x];
green = chanG[y * width + x];
blue = chanB[y * width + x];
if (red > 255) red = 255;
if (red < 0) red = 0;
if (green > 255) green = 255;
if (green < 0) green = 0;
if (blue > 255) blue = 255;
if (blue < 0) blue = 0;
// Also, it's written in (b,g,r) format...
fprintf(outfile, "%c", blue);
fprintf(outfile, "%c", green);
fprintf(outfile, "%c", red);
}
if (extrabytes) // See above - BMP lines must be of lengths divisible by 4.
{
for (n = 1; n <= extrabytes; n++) {
fprintf(outfile, "%c", 0);
}
}
}
fclose(outfile);
return 0;
}
// write bmp, input - RGB, device
int writeBMPi(const char *filename, const unsigned char *d_RGB, int pitch,
int width, int height) {
unsigned int headers[13];
FILE *outfile;
int extrabytes;
int paddedsize;
int x;
int y;
int n;
int red, green, blue;
std::vector<unsigned char> vchanRGB(height * width * 3);
unsigned char *chanRGB = vchanRGB.data();
checkCudaErrors(cudaMemcpy2D(chanRGB, (size_t)width * 3, d_RGB, (size_t)pitch,
width * 3, height, cudaMemcpyDeviceToHost));
extrabytes =
4 - ((width * 3) % 4); // How many bytes of padding to add to each
// horizontal line - the size of which must
// be a multiple of 4 bytes.
if (extrabytes == 4) extrabytes = 0;
paddedsize = ((width * 3) + extrabytes) * height;
// Headers...
// Note that the "BM" identifier in bytes 0 and 1 is NOT included in these
// "headers".
headers[0] = paddedsize + 54; // bfSize (whole file size)
headers[1] = 0; // bfReserved (both)
headers[2] = 54; // bfOffbits
headers[3] = 40; // biSize
headers[4] = width; // biWidth
headers[5] = height; // biHeight
// Would have biPlanes and biBitCount in position 6, but they're shorts.
// It's easier to write them out separately (see below) than pretend
// they're a single int, especially with endian issues...
headers[7] = 0; // biCompression
headers[8] = paddedsize; // biSizeImage
headers[9] = 0; // biXPelsPerMeter
headers[10] = 0; // biYPelsPerMeter
headers[11] = 0; // biClrUsed
headers[12] = 0; // biClrImportant
if (!(outfile = fopen(filename, "wb"))) {
std::cerr << "Cannot open file: " << filename << std::endl;
return 1;
}
//
// Headers begin...
// When printing ints and shorts, we write out 1 character at a time to avoid
// endian issues.
//
fprintf(outfile, "BM");
for (n = 0; n <= 5; n++) {
fprintf(outfile, "%c", headers[n] & 0x000000FF);
fprintf(outfile, "%c", (headers[n] & 0x0000FF00) >> 8);
fprintf(outfile, "%c", (headers[n] & 0x00FF0000) >> 16);
fprintf(outfile, "%c", (headers[n] & (unsigned int)0xFF000000) >> 24);
}
// These next 4 characters are for the biPlanes and biBitCount fields.
fprintf(outfile, "%c", 1);
fprintf(outfile, "%c", 0);
fprintf(outfile, "%c", 24);
fprintf(outfile, "%c", 0);
for (n = 7; n <= 12; n++) {
fprintf(outfile, "%c", headers[n] & 0x000000FF);
fprintf(outfile, "%c", (headers[n] & 0x0000FF00) >> 8);
fprintf(outfile, "%c", (headers[n] & 0x00FF0000) >> 16);
fprintf(outfile, "%c", (headers[n] & (unsigned int)0xFF000000) >> 24);
}
//
// Headers done, now write the data...
//
for (y = height - 1; y >= 0;
y--) // BMP image format is written from bottom to top...
{
for (x = 0; x <= width - 1; x++) {
red = chanRGB[(y * width + x) * 3];
green = chanRGB[(y * width + x) * 3 + 1];
blue = chanRGB[(y * width + x) * 3 + 2];
if (red > 255) red = 255;
if (red < 0) red = 0;
if (green > 255) green = 255;
if (green < 0) green = 0;
if (blue > 255) blue = 255;
if (blue < 0) blue = 0;
// Also, it's written in (b,g,r) format...
fprintf(outfile, "%c", blue);
fprintf(outfile, "%c", green);
fprintf(outfile, "%c", red);
}
if (extrabytes) // See above - BMP lines must be of lengths divisible by 4.
{
for (n = 1; n <= extrabytes; n++) {
fprintf(outfile, "%c", 0);
}
}
}
fclose(outfile);
return 0;
}
int inputDirExists(const char *pathname) {
struct stat info;
if (stat(pathname, &info) != 0) {
return 0; // Directory does not exists
} else if (info.st_mode & S_IFDIR) {
// is a directory
return 1;
} else {
// is not a directory
return 0;
}
}
int readInput(const std::string &sInputPath,
std::vector<std::string> &filelist) {
int error_code = 1;
#ifndef _WIN64
struct stat s;
if (stat(sInputPath.c_str(), &s) == 0) {
if (s.st_mode & S_IFREG) {
filelist.push_back(sInputPath);
} else if (s.st_mode & S_IFDIR) {
// processing each file in directory
DIR *dir_handle;
struct dirent *dir;
dir_handle = opendir(sInputPath.c_str());
std::vector<std::string> filenames;
if (dir_handle) {
error_code = 0;
while ((dir = readdir(dir_handle)) != NULL) {
std::string sFileName = sInputPath + dir->d_name;
if (inputDirExists(sFileName.c_str()))
{
std::string sname = dir->d_name;
if (sname != "." && sname != "..") {
readInput(sInputPath + sname + "/", filelist);
}
}
else
{
filelist.push_back(sFileName);
}
}
closedir(dir_handle);
} else {
std::cout << "Cannot open input directory: " << sInputPath << std::endl;
return error_code;
}
} else {
std::cout << "Cannot open input: " << sInputPath << std::endl;
return error_code;
}
} else {
std::cout << "Cannot find input path " << sInputPath << std::endl;
return error_code;
}
#else
std::string search_path = sInputPath + "/*.*";
WIN32_FIND_DATA fd;
HANDLE hFind = ::FindFirstFile(search_path.c_str(), &fd);
if (hFind != INVALID_HANDLE_VALUE) {
do {
// read all (real) files in current folder
// , delete '!' read other 2 default folder . and ..
if (!(fd.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY)) {
std::string temp(sInputPath + "\\" + fd.cFileName);
filelist.push_back(temp);
}
} while (::FindNextFile(hFind, &fd));
::FindClose(hFind);
} else {
std::cout << "Cannot open input directory: " << sInputPath << std::endl;
return error_code;
}
#endif
return 0;
}
int getInputDir(std::string &input_dir, const char *executable_path) {
int found = 0;
if (executable_path != 0) {
std::string executable_name = std::string(executable_path);
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
// Windows path delimiter
size_t delimiter_pos = executable_name.find_last_of('\\');
executable_name.erase(0, delimiter_pos + 1);
if (executable_name.rfind(".exe") != std::string::npos) {
// we strip .exe, only if the .exe is found
executable_name.resize(executable_name.size() - 4);
}
#else
// Linux & OSX path delimiter
size_t delimiter_pos = executable_name.find_last_of('/');
executable_name.erase(0, delimiter_pos + 1);
#endif
// Search in default paths for input images.
std::string pathname = "";
const char *searchPath[] = {
"./images", "../../../../Samples/<executable_name>/images",
"../../../Samples/<executable_name>/images",
"../../Samples/<executable_name>/images"};
for (unsigned int i = 0; i < sizeof(searchPath) / sizeof(char *); ++i) {
std::string pathname(searchPath[i]);
size_t executable_name_pos = pathname.find("<executable_name>");
// If there is executable_name variable in the searchPath
// replace it with the value
if (executable_name_pos != std::string::npos) {
pathname.replace(executable_name_pos, strlen("<executable_name>"),
executable_name);
}
if (inputDirExists(pathname.c_str())) {
input_dir = pathname + "/";
found = 1;
break;
}
}
}
return found;
}
#endif

View File

@@ -68,7 +68,37 @@ void compileFileToPTX(char *filename, int argc, char **argv, char **ptxResult,
int numCompileOptions = 0; int numCompileOptions = 0;
char *compileParams[1]; char *compileParams[2];
int major = 0, minor = 0;
char deviceName[256];
// Picks the best CUDA device available
CUdevice cuDevice = findCudaDeviceDRV(argc, (const char **)argv);
// get compute capabilities and the devicename
checkCudaErrors(cuDeviceGetAttribute(
&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, cuDevice));
checkCudaErrors(cuDeviceGetAttribute(
&minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, cuDevice));
{
// Compile for the GPU arch on which are going to run cuda kernel.
std::string compileOptions;
compileOptions = "--gpu-architecture=compute_";
compileParams[numCompileOptions] = reinterpret_cast<char *>(
malloc(sizeof(char) * (compileOptions.length() + 10)));
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
sprintf_s(compileParams[numCompileOptions], sizeof(char) * (compileOptions.length() + 10),
"%s%d%d", compileOptions.c_str(), major, minor);
#else
snprintf(compileParams[numCompileOptions], compileOptions.size() + 10, "%s%d%d",
compileOptions.c_str(), major, minor);
#endif
}
numCompileOptions++;
if (requiresCGheaders) { if (requiresCGheaders) {
std::string compileOptions; std::string compileOptions;
@@ -92,13 +122,13 @@ void compileFileToPTX(char *filename, int argc, char **argv, char **ptxResult,
argv[0]); argv[0]);
} }
compileOptions += path.c_str(); compileOptions += path.c_str();
compileParams[0] = reinterpret_cast<char *>( compileParams[numCompileOptions] = reinterpret_cast<char *>(
malloc(sizeof(char) * (compileOptions.length() + 1))); malloc(sizeof(char) * (compileOptions.length() + 1)));
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64) #if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
sprintf_s(compileParams[0], sizeof(char) * (compileOptions.length() + 1), sprintf_s(compileParams[numCompileOptions], sizeof(char) * (compileOptions.length() + 1),
"%s", compileOptions.c_str()); "%s", compileOptions.c_str());
#else #else
snprintf(compileParams[0], compileOptions.size(), "%s", snprintf(compileParams[numCompileOptions], compileOptions.size(), "%s",
compileOptions.c_str()); compileOptions.c_str());
#endif #endif
numCompileOptions++; numCompileOptions++;
@@ -137,7 +167,9 @@ void compileFileToPTX(char *filename, int argc, char **argv, char **ptxResult,
*ptxResult = ptx; *ptxResult = ptx;
*ptxResultSize = ptxSize; *ptxResultSize = ptxSize;
if (requiresCGheaders) free(compileParams[0]); for (int i = 0; i < numCompileOptions; i++) {
free(compileParams[i]);
}
} }
CUmodule loadPTX(char *ptx, int argc, char **argv) { CUmodule loadPTX(char *ptx, int argc, char **argv) {

View File

@@ -0,0 +1,124 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
////////////////////////////////////////////////////////////////////////////////
//
// Utility funcs to wrap up saving a surface or the back buffer as a PPM file
// In addition, wraps up a threshold comparision of two PPMs.
//
// These functions are designed to be used to implement an automated QA testing for SDK samples.
//
// Author: Bryan Dudash
// Email: sdkfeedback@nvidia.com
//
// Copyright (c) NVIDIA Corporation. All rights reserved.
////////////////////////////////////////////////////////////////////////////////
#include <helper_functions.h>
#include <rendercheck_d3d11.h>
HRESULT CheckRenderD3D11::ActiveRenderTargetToPPM(ID3D11Device *pDevice, const char *zFileName)
{
ID3D11DeviceContext *pDeviceCtxt;
pDevice->GetImmediateContext(&pDeviceCtxt);
ID3D11RenderTargetView *pRTV = NULL;
pDeviceCtxt->OMGetRenderTargets(1,&pRTV,NULL);
ID3D11Resource *pSourceResource = NULL;
pRTV->GetResource(&pSourceResource);
return ResourceToPPM(pDevice,pSourceResource,zFileName);
}
HRESULT CheckRenderD3D11::ResourceToPPM(ID3D11Device *pDevice, ID3D11Resource *pResource, const char *zFileName)
{
ID3D11DeviceContext *pDeviceCtxt;
pDevice->GetImmediateContext(&pDeviceCtxt);
D3D11_RESOURCE_DIMENSION rType;
pResource->GetType(&rType);
if (rType != D3D11_RESOURCE_DIMENSION_TEXTURE2D)
{
printf("SurfaceToPPM: pResource is not a 2D texture! Aborting...\n");
return E_FAIL;
}
ID3D11Texture2D *pSourceTexture = (ID3D11Texture2D *)pResource;
ID3D11Texture2D *pTargetTexture = NULL;
D3D11_TEXTURE2D_DESC desc;
pSourceTexture->GetDesc(&desc);
desc.BindFlags = 0;
desc.CPUAccessFlags = D3D11_CPU_ACCESS_READ;
desc.Usage = D3D11_USAGE_STAGING;
if (FAILED(pDevice->CreateTexture2D(&desc,NULL,&pTargetTexture)))
{
printf("SurfaceToPPM: Unable to create target Texture resoruce! Aborting... \n");
return E_FAIL;
}
pDeviceCtxt->CopyResource(pTargetTexture,pSourceTexture);
D3D11_MAPPED_SUBRESOURCE mappedTex2D;
pDeviceCtxt->Map(pTargetTexture, 0, D3D11_MAP_READ,0,&mappedTex2D);
// Need to convert from dx pitch to pitch=width
unsigned char *pPPMData = new unsigned char[desc.Width*desc.Height*4];
for (unsigned int iHeight = 0; iHeight<desc.Height; iHeight++)
{
memcpy(&(pPPMData[iHeight*desc.Width*4]),(unsigned char *)(mappedTex2D.pData)+iHeight*mappedTex2D.RowPitch,desc.Width*4);
}
pDeviceCtxt->Unmap(pTargetTexture, 0);
// Prepends the PPM header info and bumps byte data afterwards
sdkSavePPM4ub(zFileName, pPPMData, desc.Width, desc.Height);
delete [] pPPMData;
pTargetTexture->Release();
return S_OK;
}
bool CheckRenderD3D11::PPMvsPPM(const char *src_file, const char *ref_file, const char *exec_path,
const float epsilon, const float threshold)
{
char *ref_file_path = sdkFindFilePath(ref_file, exec_path);
if (ref_file_path == NULL)
{
printf("CheckRenderD3D11::PPMvsPPM unable to find <%s> in <%s> Aborting comparison!\n", ref_file, exec_path);
printf(">>> Check info.xml and [project//data] folder <%s> <<<\n", ref_file);
printf("Aborting comparison!\n");
printf(" FAILURE!\n");
return false;
}
return sdkComparePPM(src_file,ref_file_path,epsilon,threshold,true) == true;
}

View File

@@ -0,0 +1,52 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#pragma once
#ifndef _RENDERCHECK_D3D11_H_
#define _RENDERCHECK_D3D11_H_
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <assert.h>
#include <d3d11.h>
class CheckRenderD3D11
{
public:
CheckRenderD3D11() {}
static HRESULT ActiveRenderTargetToPPM(ID3D11Device *pDevice, const char *zFileName);
static HRESULT ResourceToPPM(ID3D11Device *pDevice, ID3D11Resource *pResource, const char *zFileName);
static bool PPMvsPPM(const char *src_file, const char *ref_file, const char *exec_path,
const float epsilon, const float threshold = 0.0f);
};
#endif

114
README.md
View File

@@ -1,11 +1,43 @@
# CUDA Samples # CUDA Samples
Samples for CUDA Developers which demonstrates features in CUDA Toolkit. This version supports [CUDA Toolkit 10.1](https://developer.nvidia.com/cuda-downloads). Samples for CUDA Developers which demonstrates features in CUDA Toolkit. This version supports [CUDA Toolkit 11.0](https://developer.nvidia.com/cuda-downloads).
## Release Notes ## Release Notes
This section describes the release notes for the CUDA Samples on GitHub only. This section describes the release notes for the CUDA Samples on GitHub only.
### CUDA 11.0
* Added `dmmaTensorCoreGemm`. Demonstrates double precision GEMM computation using the Double precision Warp Matrix Multiply and Accumulate (WMMA) API introduced with CUDA 11 in Ampere chip family tensor cores.
* Added `bf16TensorCoreGemm`. Demonstrates __nv_bfloat16 (e8m7) GEMM computation using the __nv_bfloat16 WMMA API introduced with CUDA 11 in Ampere chip family tensor cores.
* Added `tf32TensorCoreGemm`. Demonstrates tf32 (e8m10) GEMM computation using the tf32 WMMA API introduced with CUDA 11 in Ampere chip family tensor cores.
* Added `globalToShmemAsyncCopy`. Demonstrates async copy of data from global to shared memory when on compute capability 8.0 or higher. Also demonstrates arrive-wait barrier for synchronization.
* Added `simpleAWBarrier`. Demonstrates arrive wait barriers.
* Added `simpleAttributes`. Demonstrates the stream attributes that affect L2 locality.
* Added warp aggregated atomic multi bucket increments kernel using labeled_partition cooperative groups in `warpAggregatedAtomicsCG` which can be used on compute capability 7.0 and above GPU architectures.
* Added `binaryPartitionCG`. Demonstrates binary partition cooperative groups and reduction within the thread block.
* Added two new reduction kernels in `reduction` one which demonstrates reduce_add_sync intrinstic supported on compute capability 8.0 and another which uses cooperative_groups::reduce function which does thread_block_tile level reduction introduced from CUDA 11.0.
* Added `cudaCompressibleMemory`. Demonstrates compressible memory allocation using cuMemMap API.
* Added `simpleVulkanMMAP`. Demonstrates Vulkan CUDA Interop via cuMemMap APIs.
* Added `concurrentKernels`. Demonstrates the use of CUDA streams for concurrent execution of several kernels on a GPU.
* Dropped Mac OSX support from all samples.
### CUDA 10.2
* Added `simpleD3D11`. Demonstrates CUDA-D3D11 External Resource Interoperability APIs for updating D3D11 buffers from CUDA and synchronization between D3D11 and CUDA with Keyed Mutexes.
* Added `simpleDrvRuntime`. Demonstrates CUDA Driver and Runtime APIs working together to load fatbinary of a CUDA kernel.
* Added `vectorAddMMAP`. Demonstrates how cuMemMap API allows the user to specify the physical properties of their memory while retaining the contiguous nature of their access.
* Added `memMapIPCDrv`. Demonstrates Inter Process Communication using cuMemMap APIs.
* Added `cudaNvSci`. Demonstrates CUDA-NvSciBuf/NvSciSync Interop.
* Added `jacobiCudaGraphs`. Demonstrates Instantiated CUDA Graph Update with Jacobi Iterative Method using different approaches.
* Added `cuSolverSp_LinearSolver`. Demonstrates cuSolverSP's LU, QR and Cholesky factorization.
* Added `MersenneTwisterGP11213`. Demonstrates the Mersenne Twister random number generator GP11213 in cuRAND.
### CUDA 10.1 Update 2
* Added `vulkanImageCUDA`. Demonstrates how to perform Vulkan image - CUDA Interop.
* Added `nvJPEG_encoder`. Demonstrates encoding of jpeg images using NVJPEG Library.
* Added Windows OS support to `nvJPEG` sample.
* Added `boxFilterNPP`. Demonstrates how to use NPP FilterBox function to perform a box filter.
* Added `cannyEdgeDetectorNPP`. Demonstrates the nppiFilterCannyBorder_8u_C1R Canny Edge Detection image filter function.
### CUDA 10.1 Update 1 ### CUDA 10.1 Update 1
* Added `NV12toBGRandResize`. Demonstrates how to convert and resize NV12 frames to BGR planars frames using CUDA in batch. * Added `NV12toBGRandResize`. Demonstrates how to convert and resize NV12 frames to BGR planars frames using CUDA in batch.
* Added `EGLStream_CUDA_Interop`. Demonstrates data exchange between CUDA and EGL Streams. * Added `EGLStream_CUDA_Interop`. Demonstrates data exchange between CUDA and EGL Streams.
@@ -52,8 +84,8 @@ This is the first release of CUDA Samples on GitHub:
### Prerequisites ### Prerequisites
Download and install the [CUDA Toolkit 10.1](https://developer.nvidia.com/cuda-downloads) for your corresponding platform. Download and install the [CUDA Toolkit 11.0](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
For system requirements and installation instructions of cuda toolkit, please refer to the [Linux Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-linux/), the [Windows Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html), and the [Mac Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-mac-os-x/index.html). For system requirements and installation instructions of cuda toolkit, please refer to the [Linux Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-linux/), and the [Windows Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html).
### Getting the CUDA Samples ### Getting the CUDA Samples
@@ -78,7 +110,6 @@ Each individual sample has its own set of solution files at:
`<CUDA_SAMPLES_REPO>\Samples\<sample_dir>\` `<CUDA_SAMPLES_REPO>\Samples\<sample_dir>\`
To build/examine all the samples at once, the complete solution files should be used. To build/examine a single sample, the individual sample solution files should be used. To build/examine all the samples at once, the complete solution files should be used. To build/examine a single sample, the individual sample solution files should be used.
> **Note:** Some samples require that the Microsoft DirectX SDK (June 2010 or newer) be installed and that the VC++ directory paths are properly set up (**Tools > Options...**). Check [DirectX Dependencies](#directx) section for details."
### Linux ### Linux
The Linux samples are built using makefiles. To use the makefiles, change the current directory to the sample directory you wish to build, and run make: The Linux samples are built using makefiles. To use the makefiles, change the current directory to the sample directory you wish to build, and run make:
@@ -105,61 +136,39 @@ The samples makefiles can take advantage of certain options:
$ make HOST_COMPILER=g++ $ make HOST_COMPILER=g++
``` ```
### Mac
The Mac samples are built using makefiles. To use the makefiles, change directory into the sample directory you wish to build, and run make:
```
$ cd <sample_dir>
$ make
```
The samples makefiles can take advantage of certain options:
* **dbg=1** - build with debug symbols
```
$ make dbg=1
```
* **SMS="A B ..."** - override the SM architectures for which the sample will be built, where "A B ..." is a space-delimited list of SM architectures. For example, to generate SASS for SM 50 and SM 60, use SMS="50 60".
```
$ make SMS="A B ..."
```
* **HOST_COMPILER=<host_compiler>** - override the default clang host compiler. See the [Mac Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-mac-os-x/index.html#system-requirements) for a list of supported host compilers.
```
$ make HOST_COMPILER=clang
```
## Samples list ## Samples list
### Samples by OS ### Samples by OS
#### Linux #### Linux
**[warpAggregatedAtomicsCG](./Samples/warpAggregatedAtomicsCG)** | **[EGLStream_CUDA_Interop](./Samples/EGLStream_CUDA_Interop)** | **[conjugateGradientMultiBlockCG](./Samples/conjugateGradientMultiBlockCG)** | **[simpleIPC](./Samples/simpleIPC)** | **[warpAggregatedAtomicsCG](./Samples/warpAggregatedAtomicsCG)** | **[boxFilterNPP](./Samples/boxFilterNPP)** | **[binaryPartitionCG](./Samples/binaryPartitionCG)** | **[dmmaTensorCoreGemm](./Samples/dmmaTensorCoreGemm)** |
---|---|---|---| ---|---|---|---|
**[shfl_scan](./Samples/shfl_scan)** | **[conjugateGradientCudaGraphs](./Samples/conjugateGradientCudaGraphs)** | **[nvJPEG](./Samples/nvJPEG)** | **[simpleCudaGraphs](./Samples/simpleCudaGraphs)** | **[EGLStream_CUDA_Interop](./Samples/EGLStream_CUDA_Interop)** | **[conjugateGradientMultiBlockCG](./Samples/conjugateGradientMultiBlockCG)** | **[simpleIPC](./Samples/simpleIPC)** | **[memMapIPCDrv](./Samples/memMapIPCDrv)** |
**[deviceQuery](./Samples/deviceQuery)** | **[simpleVoteIntrinsics](./Samples/simpleVoteIntrinsics)** | **[simpleCUBLASXT](./Samples/simpleCUBLASXT)** | **[UnifiedMemoryPerf](./Samples/UnifiedMemoryPerf)** | **[vectorAddMMAP](./Samples/vectorAddMMAP)** | **[shfl_scan](./Samples/shfl_scan)** | **[conjugateGradientCudaGraphs](./Samples/conjugateGradientCudaGraphs)** | **[globalToShmemAsyncCopy](./Samples/globalToShmemAsyncCopy)** |
**[cuSolverDn_LinearSolver](./Samples/cuSolverDn_LinearSolver)** | **[conjugateGradientMultiDeviceCG](./Samples/conjugateGradientMultiDeviceCG)** | **[matrixMulDrv](./Samples/matrixMulDrv)** | **[simpleCUFFT](./Samples/simpleCUFFT)** | **[nvJPEG](./Samples/nvJPEG)** | **[simpleCudaGraphs](./Samples/simpleCudaGraphs)** | **[deviceQuery](./Samples/deviceQuery)** | **[simpleVoteIntrinsics](./Samples/simpleVoteIntrinsics)** |
**[reduction](./Samples/reduction)** | **[immaTensorCoreGemm](./Samples/immaTensorCoreGemm)** | **[bandwidthTest](./Samples/bandwidthTest)** | **[simpleCUBLAS](./Samples/simpleCUBLAS)** | **[simpleCUBLASXT](./Samples/simpleCUBLASXT)** | **[simpleAttributes](./Samples/simpleAttributes)** | **[cudaNvSci](./Samples/cudaNvSci)** | **[tf32TensorCoreGemm](./Samples/tf32TensorCoreGemm)** |
**[NV12toBGRandResize](./Samples/NV12toBGRandResize)** | **[cudaTensorCoreGemm](./Samples/cudaTensorCoreGemm)** | **[simpleVulkan](./Samples/simpleVulkan)** | **[vectorAdd_nvrtc](./Samples/vectorAdd_nvrtc)** | **[UnifiedMemoryPerf](./Samples/UnifiedMemoryPerf)** | **[cudaCompressibleMemory](./Samples/cudaCompressibleMemory)** | **[bf16TensorCoreGemm](./Samples/bf16TensorCoreGemm)** | **[cuSolverDn_LinearSolver](./Samples/cuSolverDn_LinearSolver)** |
**[p2pBandwidthLatencyTest](./Samples/p2pBandwidthLatencyTest)** | **[matrixMul](./Samples/matrixMul)** | **[systemWideAtomics](./Samples/systemWideAtomics)** | **[vulkanImageCUDA](./Samples/vulkanImageCUDA)** | **[conjugateGradientMultiDeviceCG](./Samples/conjugateGradientMultiDeviceCG)** | **[matrixMulDrv](./Samples/matrixMulDrv)** | **[cuSolverSp_LinearSolver](./Samples/cuSolverSp_LinearSolver)** |
**[simpleCUFFT](./Samples/simpleCUFFT)** | **[reduction](./Samples/reduction)** | **[nvJPEG_encoder](./Samples/nvJPEG_encoder)** | **[simpleDrvRuntime](./Samples/simpleDrvRuntime)** |
**[MersenneTwisterGP11213](./Samples/MersenneTwisterGP11213)** | **[simpleAWBarrier](./Samples/simpleAWBarrier)** | **[immaTensorCoreGemm](./Samples/immaTensorCoreGemm)** | **[bandwidthTest](./Samples/bandwidthTest)** |
**[concurrentKernels](./Samples/concurrentKernels)** | **[simpleCUBLAS](./Samples/simpleCUBLAS)** | **[NV12toBGRandResize](./Samples/NV12toBGRandResize)** | **[cudaTensorCoreGemm](./Samples/cudaTensorCoreGemm)** |
**[jacobiCudaGraphs](./Samples/jacobiCudaGraphs)** | **[simpleVulkan](./Samples/simpleVulkan)** | **[vectorAdd_nvrtc](./Samples/vectorAdd_nvrtc)** | **[cannyEdgeDetectorNPP](./Samples/cannyEdgeDetectorNPP)** |
**[p2pBandwidthLatencyTest](./Samples/p2pBandwidthLatencyTest)** | **[simpleVulkanMMAP](./Samples/simpleVulkanMMAP)** | **[matrixMul](./Samples/matrixMul)** | **[systemWideAtomics](./Samples/systemWideAtomics)** |
#### Windows #### Windows
**[warpAggregatedAtomicsCG](./Samples/warpAggregatedAtomicsCG)** | **[conjugateGradientMultiBlockCG](./Samples/conjugateGradientMultiBlockCG)** | **[simpleIPC](./Samples/simpleIPC)** | **[shfl_scan](./Samples/shfl_scan)** | **[warpAggregatedAtomicsCG](./Samples/warpAggregatedAtomicsCG)** | **[boxFilterNPP](./Samples/boxFilterNPP)** | **[binaryPartitionCG](./Samples/binaryPartitionCG)** | **[dmmaTensorCoreGemm](./Samples/dmmaTensorCoreGemm)** |
---|---|---|---| ---|---|---|---|
**[conjugateGradientCudaGraphs](./Samples/conjugateGradientCudaGraphs)** | **[simpleD3D12](./Samples/simpleD3D12)** | **[simpleCudaGraphs](./Samples/simpleCudaGraphs)** | **[deviceQuery](./Samples/deviceQuery)** | **[conjugateGradientMultiBlockCG](./Samples/conjugateGradientMultiBlockCG)** | **[simpleIPC](./Samples/simpleIPC)** | **[memMapIPCDrv](./Samples/memMapIPCDrv)** | **[vectorAddMMAP](./Samples/vectorAddMMAP)** |
**[simpleVoteIntrinsics](./Samples/simpleVoteIntrinsics)** | **[simpleCUBLASXT](./Samples/simpleCUBLASXT)** | **[UnifiedMemoryPerf](./Samples/UnifiedMemoryPerf)** | **[cuSolverDn_LinearSolver](./Samples/cuSolverDn_LinearSolver)** | **[shfl_scan](./Samples/shfl_scan)** | **[conjugateGradientCudaGraphs](./Samples/conjugateGradientCudaGraphs)** | **[globalToShmemAsyncCopy](./Samples/globalToShmemAsyncCopy)** | **[nvJPEG](./Samples/nvJPEG)** |
**[conjugateGradientMultiDeviceCG](./Samples/conjugateGradientMultiDeviceCG)** | **[matrixMulDrv](./Samples/matrixMulDrv)** | **[simpleCUFFT](./Samples/simpleCUFFT)** | **[reduction](./Samples/reduction)** | **[simpleD3D12](./Samples/simpleD3D12)** | **[simpleCudaGraphs](./Samples/simpleCudaGraphs)** | **[deviceQuery](./Samples/deviceQuery)** | **[simpleVoteIntrinsics](./Samples/simpleVoteIntrinsics)** |
**[immaTensorCoreGemm](./Samples/immaTensorCoreGemm)** | **[bandwidthTest](./Samples/bandwidthTest)** | **[simpleCUBLAS](./Samples/simpleCUBLAS)** | **[NV12toBGRandResize](./Samples/NV12toBGRandResize)** | **[simpleCUBLASXT](./Samples/simpleCUBLASXT)** | **[simpleAttributes](./Samples/simpleAttributes)** | **[tf32TensorCoreGemm](./Samples/tf32TensorCoreGemm)** | **[UnifiedMemoryPerf](./Samples/UnifiedMemoryPerf)** |
**[cudaTensorCoreGemm](./Samples/cudaTensorCoreGemm)** | **[simpleVulkan](./Samples/simpleVulkan)** | **[vectorAdd_nvrtc](./Samples/vectorAdd_nvrtc)** | **[p2pBandwidthLatencyTest](./Samples/p2pBandwidthLatencyTest)** | **[cudaCompressibleMemory](./Samples/cudaCompressibleMemory)** | **[bf16TensorCoreGemm](./Samples/bf16TensorCoreGemm)** | **[cuSolverDn_LinearSolver](./Samples/cuSolverDn_LinearSolver)** | **[vulkanImageCUDA](./Samples/vulkanImageCUDA)** |
**[matrixMul](./Samples/matrixMul)** | **[conjugateGradientMultiDeviceCG](./Samples/conjugateGradientMultiDeviceCG)** | **[matrixMulDrv](./Samples/matrixMulDrv)** | **[cuSolverSp_LinearSolver](./Samples/cuSolverSp_LinearSolver)** | **[simpleCUFFT](./Samples/simpleCUFFT)** |
**[reduction](./Samples/reduction)** | **[nvJPEG_encoder](./Samples/nvJPEG_encoder)** | **[simpleDrvRuntime](./Samples/simpleDrvRuntime)** | **[simpleD3D11](./Samples/simpleD3D11)** |
#### Mac OSX **[MersenneTwisterGP11213](./Samples/MersenneTwisterGP11213)** | **[simpleAWBarrier](./Samples/simpleAWBarrier)** | **[immaTensorCoreGemm](./Samples/immaTensorCoreGemm)** | **[bandwidthTest](./Samples/bandwidthTest)** |
**[warpAggregatedAtomicsCG](./Samples/warpAggregatedAtomicsCG)** | **[shfl_scan](./Samples/shfl_scan)** | **[conjugateGradientCudaGraphs](./Samples/conjugateGradientCudaGraphs)** | **[simpleCudaGraphs](./Samples/simpleCudaGraphs)** | **[concurrentKernels](./Samples/concurrentKernels)** | **[simpleCUBLAS](./Samples/simpleCUBLAS)** | **[NV12toBGRandResize](./Samples/NV12toBGRandResize)** | **[cudaTensorCoreGemm](./Samples/cudaTensorCoreGemm)** |
---|---|---|---| **[jacobiCudaGraphs](./Samples/jacobiCudaGraphs)** | **[simpleVulkan](./Samples/simpleVulkan)** | **[vectorAdd_nvrtc](./Samples/vectorAdd_nvrtc)** | **[cannyEdgeDetectorNPP](./Samples/cannyEdgeDetectorNPP)** |
**[deviceQuery](./Samples/deviceQuery)** | **[simpleVoteIntrinsics](./Samples/simpleVoteIntrinsics)** | **[simpleCUBLASXT](./Samples/simpleCUBLASXT)** | **[UnifiedMemoryPerf](./Samples/UnifiedMemoryPerf)** | **[p2pBandwidthLatencyTest](./Samples/p2pBandwidthLatencyTest)** | **[simpleVulkanMMAP](./Samples/simpleVulkanMMAP)** | **[matrixMul](./Samples/matrixMul)** |
**[cuSolverDn_LinearSolver](./Samples/cuSolverDn_LinearSolver)** | **[matrixMulDrv](./Samples/matrixMulDrv)** | **[simpleCUFFT](./Samples/simpleCUFFT)** | **[reduction](./Samples/reduction)** |
**[bandwidthTest](./Samples/bandwidthTest)** | **[simpleCUBLAS](./Samples/simpleCUBLAS)** | **[NV12toBGRandResize](./Samples/NV12toBGRandResize)** | **[vectorAdd_nvrtc](./Samples/vectorAdd_nvrtc)** |
**[p2pBandwidthLatencyTest](./Samples/p2pBandwidthLatencyTest)** | **[matrixMul](./Samples/matrixMul)** |
## Dependencies ## Dependencies
@@ -187,7 +196,7 @@ Some samples can only be run on a 64-bit operating system.
#### DirectX #### DirectX
DirectX is a collection of APIs designed to allow development of multimedia applications on Microsoft platforms. For Microsoft platforms, NVIDIA's CUDA Driver supports DirectX. Several CUDA Samples for Windows demonstrates CUDA-DirectX Interoperability, for building such samples one needs to install [Direct X SDK (June 2010 or newer)](http://www.microsoft.com/en-us/download/details.aspx?id=6812) , this is required to be installed on Windows 7, Windows 10 and Windows Server 2008, Other Windows OSes do not need to explicitly install the DirectX SDK. DirectX is a collection of APIs designed to allow development of multimedia applications on Microsoft platforms. For Microsoft platforms, NVIDIA's CUDA Driver supports DirectX. Several CUDA Samples for Windows demonstrates CUDA-DirectX Interoperability, for building such samples one needs to install Microsoft Visual Studio 2012 or higher which provides Microsoft Windows SDK for Windows 8.
#### DirectX12 #### DirectX12
@@ -229,6 +238,9 @@ EGLOutput is a set of EGL extensions which allow EGL to render directly to the d
EGLSync is a set of EGL extensions which provides sync objects that are synchronization primitive, representing events whose completion can be tested or waited upon. EGLSync is a set of EGL extensions which provides sync objects that are synchronization primitive, representing events whose completion can be tested or waited upon.
#### NVSCI
NvSci is a set of communication interface libraries out of which CUDA interops with NvSciBuf and NvSciSync. NvSciBuf allows applications to allocate and exchange buffers in memory. NvSciSync allows applications to manage synchronization objects which coordinate when sequences of operations begin and end.
### CUDA Features ### CUDA Features
These CUDA features are needed by some CUDA samples. They are provided by either the CUDA Toolkit or CUDA Driver. Some features may not be available on your system. These CUDA features are needed by some CUDA samples. They are provided by either the CUDA Toolkit or CUDA Driver. Some features may not be available on your system.

View File

@@ -72,9 +72,9 @@ endif
# architecture # architecture
HOST_ARCH := $(shell uname -m) HOST_ARCH := $(shell uname -m)
TARGET_ARCH ?= $(HOST_ARCH) TARGET_ARCH ?= $(HOST_ARCH)
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le armv7l)) ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le armv7l))
ifneq ($(TARGET_ARCH),$(HOST_ARCH)) ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le)) ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le))
TARGET_SIZE := 64 TARGET_SIZE := 64
else ifneq (,$(filter $(TARGET_ARCH),armv7l)) else ifneq (,$(filter $(TARGET_ARCH),armv7l))
TARGET_SIZE := 32 TARGET_SIZE := 32
@@ -85,8 +85,17 @@ ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le armv7l))
else else
$(error ERROR - unsupported value $(TARGET_ARCH) for TARGET_ARCH!) $(error ERROR - unsupported value $(TARGET_ARCH) for TARGET_ARCH!)
endif endif
# sbsa and aarch64 systems look similar. Need to differentiate them at host level for now.
ifeq ($(HOST_ARCH),aarch64)
ifeq ($(CUDA_PATH)/targets/sbsa-linux,$(shell ls -1d $(CUDA_PATH)/targets/sbsa-linux))
HOST_ARCH := sbsa
TARGET_ARCH := sbsa
endif
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH)) ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-ppc64le)) ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-sbsa x86_64-ppc64le))
$(error ERROR - cross compiling from $(HOST_ARCH) to $(TARGET_ARCH) is not supported!) $(error ERROR - cross compiling from $(HOST_ARCH) to $(TARGET_ARCH) is not supported!)
endif endif
endif endif
@@ -141,6 +150,8 @@ else ifneq ($(TARGET_ARCH),$(HOST_ARCH))
else ifeq ($(TARGET_OS), android) else ifeq ($(TARGET_OS), android)
HOST_COMPILER ?= aarch64-linux-android-clang++ HOST_COMPILER ?= aarch64-linux-android-clang++
endif endif
else ifeq ($(TARGET_ARCH),sbsa)
HOST_COMPILER ?= aarch64-linux-gnu-g++
else ifeq ($(TARGET_ARCH),ppc64le) else ifeq ($(TARGET_ARCH),ppc64le)
HOST_COMPILER ?= powerpc64le-linux-gnu-g++ HOST_COMPILER ?= powerpc64le-linux-gnu-g++
endif endif
@@ -185,19 +196,27 @@ ifneq ($(TARGET_ARCH),$(HOST_ARCH))
CCFLAGS += --sysroot=$(TARGET_FS) CCFLAGS += --sysroot=$(TARGET_FS)
endif endif
LDFLAGS += --sysroot=$(TARGET_FS) LDFLAGS += --sysroot=$(TARGET_FS)
LDFLAGS += -rpath-link=$(TARGET_FS)/lib -L $(TARGET_FS)/lib LDFLAGS += -rpath-link=$(TARGET_FS)/lib -L$(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L $(TARGET_FS)/usr/lib LDFLAGS += -rpath-link=$(TARGET_FS)/lib/aarch64-linux-gnu -L$(TARGET_FS)/lib/aarch64-linux-gnu
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L $(TARGET_FS)/usr/lib/aarch64-linux-gnu LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L$(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L$(TARGET_FS)/usr/lib/aarch64-linux-gnu
LDFLAGS += --unresolved-symbols=ignore-in-shared-libs LDFLAGS += --unresolved-symbols=ignore-in-shared-libs
CCFLAGS += -isystem=$(TARGET_FS)/usr/include CCFLAGS += -isystem=$(TARGET_FS)/usr/include -I$(TARGET_FS)/usr/include
CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu -I$(TARGET_FS)/usr/include/aarch64-linux-gnu
endif
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
CCFLAGS += -DWIN_INTERFACE_CUSTOM -I/usr/include/aarch64-qnx-gnu
LDFLAGS += -lsocket
LDFLAGS += -rpath=/usr/lib/aarch64-qnx-gnu -L/usr/lib/aarch64-qnx-gnu
ifneq ($(TARGET_FS),)
LDFLAGS += -rpath=$(TARGET_FS)/usr/lib -L $(TARGET_FS)/usr/lib
LDFLAGS += -rpath=$(TARGET_FS)/usr/libnvidia -L $(TARGET_FS)/usr/libnvidia
endif
ifdef TARGET_OVERRIDE # cuda toolkit targets override
NVCCFLAGS += -target-dir $(TARGET_OVERRIDE)
endif endif
endif endif
endif
ifeq ($(TARGET_OS),qnx)
CCFLAGS += -DWIN_INTERFACE_CUSTOM
LDFLAGS += -lsocket
endif endif
# Install directory of different arch # Install directory of different arch
@@ -206,6 +225,8 @@ ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-gnueabihf/ CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-gnueabihf/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux) else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux/ CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),sbsa-linux)
CUDA_INSTALL_TARGET_DIR = targets/sbsa-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android) else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-androideabi/ CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android) else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android)
@@ -289,6 +310,10 @@ else
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/aarch64-linux/lib/stubs CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/aarch64-linux/lib/stubs
endif endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),sbsa-linux)
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/sbsa-linux/lib/stubs
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android) ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android)
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/armv7-linux-androideabi/lib/stubs CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/armv7-linux-androideabi/lib/stubs
endif endif
@@ -303,12 +328,19 @@ else
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx) ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/aarch64-qnx/lib/stubs CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/aarch64-qnx/lib/stubs
ifdef TARGET_OVERRIDE
CUDA_SEARCH_PATH := $(CUDA_PATH)/targets/$(TARGET_OVERRIDE)/lib/stubs
endif
endif endif
ifeq ($(TARGET_ARCH),ppc64le) ifeq ($(TARGET_ARCH),ppc64le)
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/ppc64le-linux/lib/stubs CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/ppc64le-linux/lib/stubs
endif endif
ifeq ($(HOST_ARCH),ppc64le)
CUDA_SEARCH_PATH += $(CUDA_PATH)/lib64/stubs
endif
CUDALIB ?= $(shell find -L $(CUDA_SEARCH_PATH) -maxdepth 1 -name libcuda.so 2> /dev/null) CUDALIB ?= $(shell find -L $(CUDA_SEARCH_PATH) -maxdepth 1 -name libcuda.so 2> /dev/null)
ifeq ("$(CUDALIB)","") ifeq ("$(CUDALIB)","")
$(info >>> WARNING - libcuda.so not found, CUDA Driver is not installed. Please re-install the driver. <<<) $(info >>> WARNING - libcuda.so not found, CUDA Driver is not installed. Please re-install the driver. <<<)

View File

@@ -49,7 +49,6 @@
<scope>1:CUDA Basic Topics</scope> <scope>1:CUDA Basic Topics</scope>
<scope>2:Graphics Interop</scope> <scope>2:Graphics Interop</scope>
</scopes> </scopes>
<sm-arch>sm30</sm-arch>
<sm-arch>sm35</sm-arch> <sm-arch>sm35</sm-arch>
<sm-arch>sm37</sm-arch> <sm-arch>sm37</sm-arch>
<sm-arch>sm50</sm-arch> <sm-arch>sm50</sm-arch>
@@ -59,6 +58,7 @@
<sm-arch>sm70</sm-arch> <sm-arch>sm70</sm-arch>
<sm-arch>sm72</sm-arch> <sm-arch>sm72</sm-arch>
<sm-arch>sm75</sm-arch> <sm-arch>sm75</sm-arch>
<sm-arch>sm80</sm-arch>
<supported_envs> <supported_envs>
<env> <env>
<arch>x86_64</arch> <arch>x86_64</arch>

View File

@@ -10,7 +10,7 @@ EGLStreams Interop
## Supported SM Architectures ## Supported SM Architectures
[SM 3.0 ](https://developer.nvidia.com/cuda-gpus) [SM 3.5 ](https://developer.nvidia.com/cuda-gpus) [SM 3.7 ](https://developer.nvidia.com/cuda-gpus) [SM 5.0 ](https://developer.nvidia.com/cuda-gpus) [SM 5.2 ](https://developer.nvidia.com/cuda-gpus) [SM 6.0 ](https://developer.nvidia.com/cuda-gpus) [SM 6.1 ](https://developer.nvidia.com/cuda-gpus) [SM 7.0 ](https://developer.nvidia.com/cuda-gpus) [SM 7.2 ](https://developer.nvidia.com/cuda-gpus) [SM 7.5 ](https://developer.nvidia.com/cuda-gpus) [SM 3.5 ](https://developer.nvidia.com/cuda-gpus) [SM 3.7 ](https://developer.nvidia.com/cuda-gpus) [SM 5.0 ](https://developer.nvidia.com/cuda-gpus) [SM 5.2 ](https://developer.nvidia.com/cuda-gpus) [SM 6.0 ](https://developer.nvidia.com/cuda-gpus) [SM 6.1 ](https://developer.nvidia.com/cuda-gpus) [SM 7.0 ](https://developer.nvidia.com/cuda-gpus) [SM 7.2 ](https://developer.nvidia.com/cuda-gpus) [SM 7.5 ](https://developer.nvidia.com/cuda-gpus) [SM 8.0 ](https://developer.nvidia.com/cuda-gpus)
## Supported OSes ## Supported OSes
@@ -30,7 +30,7 @@ cuDeviceGet, cuDeviceGetAttribute, cuDeviceComputeCapability, cuDeviceGetCount,
## Prerequisites ## Prerequisites
Download and install the [CUDA Toolkit 10.1](https://developer.nvidia.com/cuda-downloads) for your corresponding platform. Download and install the [CUDA Toolkit 11.0](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed. Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run ## Build and Run

View File

@@ -245,12 +245,30 @@ CUresult cudaProducerTest(test_cuda_producer_s *cudaProducer, char *file) {
cudaEgl.eglColorFormat = eglColorFormat; cudaEgl.eglColorFormat = eglColorFormat;
cudaEgl.cuFormat = CU_AD_FORMAT_UNSIGNED_INT8; cudaEgl.cuFormat = CU_AD_FORMAT_UNSIGNED_INT8;
static int numFramesPresented = 0;
// If there is a frame presented before we check if consumer
// is done with it using cuEGLStreamProducerReturnFrame.
while (numFramesPresented) {
CUeglFrame returnedCudaEgl;
cuStatus = cuEGLStreamProducerReturnFrame(&cudaProducer->cudaConn,
&returnedCudaEgl, NULL);
if (cuStatus == CUDA_ERROR_LAUNCH_TIMEOUT) {
continue;
} else if (cuStatus != CUDA_SUCCESS) {
printf("cuda Producer return frame FAILED with custatus= %d\n", cuStatus);
return cuStatus;
} else {
numFramesPresented--;
}
}
cuStatus = cuStatus =
cuEGLStreamProducerPresentFrame(&cudaProducer->cudaConn, cudaEgl, NULL); cuEGLStreamProducerPresentFrame(&cudaProducer->cudaConn, cudaEgl, NULL);
if (cuStatus != CUDA_SUCCESS) { if (cuStatus != CUDA_SUCCESS) {
printf("cuda Producer present frame FAILED with custatus= %d\n", cuStatus); printf("cuda Producer present frame FAILED with custatus= %d\n", cuStatus);
goto done; goto done;
} }
numFramesPresented++;
done: done:
if (file_p) { if (file_p) {
@@ -281,6 +299,13 @@ CUresult cudaDeviceCreateProducer(test_cuda_producer_s *cudaProducer,
"%d.%d\n\n", "%d.%d\n\n",
device, deviceName, major, minor); device, deviceName, major, minor);
if (major < 6) {
printf(
"EGLStreams_CUDA_Interop requires SM 6.0 or higher arch GPU. "
"Exiting...\n");
exit(2); // EXIT_WAIVED
}
if (CUDA_SUCCESS != if (CUDA_SUCCESS !=
(status = cuCtxCreate(&cudaProducer->context, 0, device))) { (status = cuCtxCreate(&cudaProducer->context, 0, device))) {
printf("failed to create CUDA context\n"); printf("failed to create CUDA context\n");

View File

@@ -133,10 +133,14 @@ ifeq ("$(TARGET_OS)","linux")
else else
endif endif
ifeq ("$(TARGET_OS)","qnx")
HOST_CCFLAGS := -V5.4.0,gcc_ntoaarch64le
endif
# Attempt to compile a minimal EGL application and run to check if EGL_SUPPORT_REUSE_NV is supported in the EGL headers available. # Attempt to compile a minimal EGL application and run to check if EGL_SUPPORT_REUSE_NV is supported in the EGL headers available.
ifneq ($(SAMPLE_ENABLED), 0) ifneq ($(SAMPLE_ENABLED), 0)
$(shell printf "#include <EGL/egl.h>\n#include <EGL/eglext.h>\nint main() {\n#ifdef EGL_SUPPORT_REUSE_NV \n #error \"Compatible EGL header found\" \n return 0;\n#endif \n return 1;\n}" > test.c; ) $(shell printf "#include <EGL/egl.h>\n#include <EGL/eglext.h>\nint main() {\n#ifdef EGL_SUPPORT_REUSE_NV \n #error \"Compatible EGL header found\" \n return 0;\n#endif \n return 1;\n}" > test.c; )
EGL_DEFINES := $(shell $(HOST_COMPILER) $(CCFLAGS) $(EXTRA_CCFLAGS) -lEGL test.c -c 2>&1 | grep -ic "Compatible EGL header found";) EGL_DEFINES := $(shell $(HOST_COMPILER) $(HOST_CCFLAGS) $(CCFLAGS) $(EXTRA_CCFLAGS) -lEGL test.c -c 2>&1 | grep -ic "Compatible EGL header found";)
SHOULD_WAIVE := 0 SHOULD_WAIVE := 0
ifeq ($(EGL_DEFINES),0) ifeq ($(EGL_DEFINES),0)
SHOULD_WAIVE := 1 SHOULD_WAIVE := 1

View File

@@ -0,0 +1,310 @@
################################################################################
# Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
# are met:
# * Redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer.
# * Redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution.
# * Neither the name of NVIDIA CORPORATION nor the names of its
# contributors may be used to endorse or promote products derived
# from this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
# EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
# PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
# PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
# OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#
################################################################################
#
# Makefile project only supported on Mac OS X and Linux Platforms)
#
################################################################################
# Location of the CUDA Toolkit
CUDA_PATH ?= /usr/local/cuda
##############################
# start deprecated interface #
##############################
ifeq ($(x86_64),1)
$(info WARNING - x86_64 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=x86_64 instead)
TARGET_ARCH ?= x86_64
endif
ifeq ($(ARMv7),1)
$(info WARNING - ARMv7 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=armv7l instead)
TARGET_ARCH ?= armv7l
endif
ifeq ($(aarch64),1)
$(info WARNING - aarch64 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=aarch64 instead)
TARGET_ARCH ?= aarch64
endif
ifeq ($(ppc64le),1)
$(info WARNING - ppc64le variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=ppc64le instead)
TARGET_ARCH ?= ppc64le
endif
ifneq ($(GCC),)
$(info WARNING - GCC variable has been deprecated)
$(info WARNING - please use HOST_COMPILER=$(GCC) instead)
HOST_COMPILER ?= $(GCC)
endif
ifneq ($(abi),)
$(error ERROR - abi variable has been removed)
endif
############################
# end deprecated interface #
############################
# architecture
HOST_ARCH := $(shell uname -m)
TARGET_ARCH ?= $(HOST_ARCH)
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le armv7l))
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le))
TARGET_SIZE := 64
else ifneq (,$(filter $(TARGET_ARCH),armv7l))
TARGET_SIZE := 32
endif
else
TARGET_SIZE := $(shell getconf LONG_BIT)
endif
else
$(error ERROR - unsupported value $(TARGET_ARCH) for TARGET_ARCH!)
endif
# sbsa and aarch64 systems look similar. Need to differentiate them at host level for now.
ifeq ($(HOST_ARCH),aarch64)
ifeq ($(CUDA_PATH)/targets/sbsa-linux,$(shell ls -1d $(CUDA_PATH)/targets/sbsa-linux))
HOST_ARCH := sbsa
TARGET_ARCH := sbsa
endif
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-sbsa x86_64-ppc64le))
$(error ERROR - cross compiling from $(HOST_ARCH) to $(TARGET_ARCH) is not supported!)
endif
endif
# When on native aarch64 system with userspace of 32-bit, change TARGET_ARCH to armv7l
ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_SIZE),aarch64-aarch64-32)
TARGET_ARCH = armv7l
endif
# operating system
HOST_OS := $(shell uname -s 2>/dev/null | tr "[:upper:]" "[:lower:]")
TARGET_OS ?= $(HOST_OS)
ifeq (,$(filter $(TARGET_OS),linux darwin qnx android))
$(error ERROR - unsupported value $(TARGET_OS) for TARGET_OS!)
endif
# host compiler
ifeq ($(TARGET_OS),darwin)
ifeq ($(shell expr `xcodebuild -version | grep -i xcode | awk '{print $$2}' | cut -d'.' -f1` \>= 5),1)
HOST_COMPILER ?= clang++
endif
else ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq ($(HOST_ARCH)-$(TARGET_ARCH),x86_64-armv7l)
ifeq ($(TARGET_OS),linux)
HOST_COMPILER ?= arm-linux-gnueabihf-g++
else ifeq ($(TARGET_OS),qnx)
ifeq ($(QNX_HOST),)
$(error ERROR - QNX_HOST must be passed to the QNX host toolchain)
endif
ifeq ($(QNX_TARGET),)
$(error ERROR - QNX_TARGET must be passed to the QNX target toolchain)
endif
export QNX_HOST
export QNX_TARGET
HOST_COMPILER ?= $(QNX_HOST)/usr/bin/arm-unknown-nto-qnx6.6.0eabi-g++
else ifeq ($(TARGET_OS),android)
HOST_COMPILER ?= arm-linux-androideabi-g++
endif
else ifeq ($(TARGET_ARCH),aarch64)
ifeq ($(TARGET_OS), linux)
HOST_COMPILER ?= aarch64-linux-gnu-g++
else ifeq ($(TARGET_OS),qnx)
ifeq ($(QNX_HOST),)
$(error ERROR - QNX_HOST must be passed to the QNX host toolchain)
endif
ifeq ($(QNX_TARGET),)
$(error ERROR - QNX_TARGET must be passed to the QNX target toolchain)
endif
export QNX_HOST
export QNX_TARGET
HOST_COMPILER ?= $(QNX_HOST)/usr/bin/aarch64-unknown-nto-qnx7.0.0-g++
else ifeq ($(TARGET_OS), android)
HOST_COMPILER ?= aarch64-linux-android-clang++
endif
else ifeq ($(TARGET_ARCH),sbsa)
HOST_COMPILER ?= aarch64-linux-gnu-g++
else ifeq ($(TARGET_ARCH),ppc64le)
HOST_COMPILER ?= powerpc64le-linux-gnu-g++
endif
endif
HOST_COMPILER ?= g++
NVCC := $(CUDA_PATH)/bin/nvcc -ccbin $(HOST_COMPILER)
# internal flags
NVCCFLAGS := -m${TARGET_SIZE}
CCFLAGS :=
LDFLAGS :=
# build flags
ifeq ($(TARGET_OS),darwin)
LDFLAGS += -rpath $(CUDA_PATH)/lib
CCFLAGS += -arch $(HOST_ARCH)
else ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_OS),x86_64-armv7l-linux)
LDFLAGS += --dynamic-linker=/lib/ld-linux-armhf.so.3
CCFLAGS += -mfloat-abi=hard
else ifeq ($(TARGET_OS),android)
LDFLAGS += -pie
CCFLAGS += -fpie -fpic -fexceptions
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
ifneq ($(TARGET_FS),)
GCCVERSIONLTEQ46 := $(shell expr `$(HOST_COMPILER) -dumpversion` \<= 4.6)
ifeq ($(GCCVERSIONLTEQ46),1)
CCFLAGS += --sysroot=$(TARGET_FS)
endif
LDFLAGS += --sysroot=$(TARGET_FS)
LDFLAGS += -rpath-link=$(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/arm-linux-gnueabihf
endif
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
ifneq ($(TARGET_FS),)
GCCVERSIONLTEQ46 := $(shell expr `$(HOST_COMPILER) -dumpversion` \<= 4.6)
ifeq ($(GCCVERSIONLTEQ46),1)
CCFLAGS += --sysroot=$(TARGET_FS)
endif
LDFLAGS += --sysroot=$(TARGET_FS)
LDFLAGS += -rpath-link=$(TARGET_FS)/lib -L$(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/lib/aarch64-linux-gnu -L$(TARGET_FS)/lib/aarch64-linux-gnu
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L$(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L$(TARGET_FS)/usr/lib/aarch64-linux-gnu
LDFLAGS += --unresolved-symbols=ignore-in-shared-libs
CCFLAGS += -isystem=$(TARGET_FS)/usr/include -I$(TARGET_FS)/usr/include
CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu -I$(TARGET_FS)/usr/include/aarch64-linux-gnu
endif
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
CCFLAGS += -DWIN_INTERFACE_CUSTOM -I/usr/include/aarch64-qnx-gnu
LDFLAGS += -lsocket
LDFLAGS += -rpath=/usr/lib/aarch64-qnx-gnu -L/usr/lib/aarch64-qnx-gnu
ifneq ($(TARGET_FS),)
LDFLAGS += -rpath=$(TARGET_FS)/usr/lib -L $(TARGET_FS)/usr/lib
LDFLAGS += -rpath=$(TARGET_FS)/usr/libnvidia -L $(TARGET_FS)/usr/libnvidia
endif
ifdef TARGET_OVERRIDE # cuda toolkit targets override
NVCCFLAGS += -target-dir $(TARGET_OVERRIDE)
endif
endif
endif
# Install directory of different arch
CUDA_INSTALL_TARGET_DIR :=
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-gnueabihf/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),sbsa-linux)
CUDA_INSTALL_TARGET_DIR = targets/sbsa-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-qnx)
CUDA_INSTALL_TARGET_DIR = targets/ARMv7-linux-QNX/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-qnx/
else ifeq ($(TARGET_ARCH),ppc64le)
CUDA_INSTALL_TARGET_DIR = targets/ppc64le-linux/
endif
# Debug build flags
ifeq ($(dbg),1)
NVCCFLAGS += -g -G
BUILD_TYPE := debug
else
BUILD_TYPE := release
endif
ALL_CCFLAGS :=
ALL_CCFLAGS += $(NVCCFLAGS)
ALL_CCFLAGS += $(EXTRA_NVCCFLAGS)
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(CCFLAGS))
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(EXTRA_CCFLAGS))
ALL_LDFLAGS :=
ALL_LDFLAGS += $(ALL_CCFLAGS)
ALL_LDFLAGS += $(addprefix -Xlinker ,$(LDFLAGS))
ALL_LDFLAGS += $(addprefix -Xlinker ,$(EXTRA_LDFLAGS))
# Common includes and paths for CUDA
INCLUDES := -I../../Common
LIBRARIES :=
################################################################################
# Gencode arguments
SMS ?=
ifeq ($(GENCODE_FLAGS),)
# Generate SASS code for each SM architecture listed in $(SMS)
$(foreach sm,$(SMS),$(eval GENCODE_FLAGS += -gencode arch=compute_$(sm),code=sm_$(sm)))
ifeq ($(SMS),)
# Generate PTX code from SM 35
GENCODE_FLAGS += -gencode arch=compute_35,code=compute_35
endif
# Generate PTX code from the highest SM architecture in $(SMS) to guarantee forward-compatibility
HIGHEST_SM := $(lastword $(sort $(SMS)))
ifneq ($(HIGHEST_SM),)
GENCODE_FLAGS += -gencode arch=compute_$(HIGHEST_SM),code=compute_$(HIGHEST_SM)
endif
endif
LIBRARIES += -lcurand_static -lculibos
################################################################################
# Target rules
all: build
build: MersenneTwisterGP11213
MersenneTwister.o:MersenneTwister.cpp
$(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
MersenneTwisterGP11213: MersenneTwister.o
$(NVCC) $(ALL_LDFLAGS) $(GENCODE_FLAGS) -o $@ $+ $(LIBRARIES)
mkdir -p ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
cp $@ ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
run: build
./MersenneTwisterGP11213
clean:
rm -f MersenneTwisterGP11213 MersenneTwister.o
rm -rf ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)/MersenneTwisterGP11213
clobber: clean

View File

@@ -0,0 +1,184 @@
/* Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
/*
* This sample demonstrates the use of CURAND to generate
* random numbers on GPU and CPU.
*/
// Utilities and system includes
// includes, system
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <curand.h>
// Utilities and system includes
#include <helper_functions.h>
#include <helper_cuda.h>
#include <cuda_runtime.h>
#include <curand.h>
float compareResults(int rand_n, float *h_RandGPU, float *h_RandCPU);
const int DEFAULT_RAND_N = 2400000;
const unsigned int DEFAULT_SEED = 777;
///////////////////////////////////////////////////////////////////////////////
// Main program
///////////////////////////////////////////////////////////////////////////////
int main(int argc, char **argv)
{
// Start logs
printf("%s Starting...\n\n", argv[0]);
// initialize the GPU, either identified by --device
// or by picking the device with highest flop rate.
int devID = findCudaDevice(argc, (const char **)argv);
// parsing the number of random numbers to generate
int rand_n = DEFAULT_RAND_N;
if (checkCmdLineFlag(argc, (const char **) argv, "count"))
{
rand_n = getCmdLineArgumentInt(argc, (const char **) argv, "count");
}
printf("Allocating data for %i samples...\n", rand_n);
// parsing the seed
int seed = DEFAULT_SEED;
if (checkCmdLineFlag(argc, (const char **) argv, "seed"))
{
seed = getCmdLineArgumentInt(argc, (const char **) argv, "seed");
}
printf("Seeding with %i ...\n", seed);
cudaStream_t stream;
checkCudaErrors(cudaStreamCreateWithFlags(&stream, cudaStreamNonBlocking));
float *d_Rand;
checkCudaErrors(cudaMalloc((void **)&d_Rand, rand_n * sizeof(float)));
curandGenerator_t prngGPU;
checkCudaErrors(curandCreateGenerator(&prngGPU, CURAND_RNG_PSEUDO_MTGP32));
checkCudaErrors(curandSetStream(prngGPU, stream));
checkCudaErrors(curandSetPseudoRandomGeneratorSeed(prngGPU, seed));
curandGenerator_t prngCPU;
checkCudaErrors(curandCreateGeneratorHost(&prngCPU, CURAND_RNG_PSEUDO_MTGP32));
checkCudaErrors(curandSetPseudoRandomGeneratorSeed(prngCPU, seed));
//
// Example 1: Compare random numbers generated on GPU and CPU
float *h_RandGPU = (float *)malloc(rand_n * sizeof(float));
printf("Generating random numbers on GPU...\n\n");
checkCudaErrors(curandGenerateUniform(prngGPU, (float *) d_Rand, rand_n));
printf("\nReading back the results...\n");
checkCudaErrors(cudaMemcpyAsync(h_RandGPU, d_Rand, rand_n * sizeof(float), cudaMemcpyDeviceToHost, stream));
float *h_RandCPU = (float *)malloc(rand_n * sizeof(float));
printf("Generating random numbers on CPU...\n\n");
checkCudaErrors(curandGenerateUniform(prngCPU, (float *) h_RandCPU, rand_n));
checkCudaErrors(cudaStreamSynchronize(stream));
printf("Comparing CPU/GPU random numbers...\n\n");
float L1norm = compareResults(rand_n, h_RandGPU, h_RandCPU);
//
// Example 2: Timing of random number generation on GPU
const int numIterations = 10;
int i;
StopWatchInterface *hTimer;
sdkCreateTimer(&hTimer);
sdkResetTimer(&hTimer);
sdkStartTimer(&hTimer);
for (i = 0; i < numIterations; i++)
{
checkCudaErrors(curandGenerateUniform(prngGPU, (float *) d_Rand, rand_n));
}
checkCudaErrors(cudaStreamSynchronize(stream));
sdkStopTimer(&hTimer);
double gpuTime = 1.0e-3 * sdkGetTimerValue(&hTimer)/(double)numIterations;
printf("MersenneTwisterGP11213, Throughput = %.4f GNumbers/s, Time = %.5f s, Size = %u Numbers\n",
1.0e-9 * rand_n / gpuTime, gpuTime, rand_n);
printf("Shutting down...\n");
checkCudaErrors(curandDestroyGenerator(prngGPU));
checkCudaErrors(curandDestroyGenerator(prngCPU));
checkCudaErrors(cudaStreamDestroy(stream));
checkCudaErrors(cudaFree(d_Rand));
sdkDeleteTimer(&hTimer);
free(h_RandGPU);
free(h_RandCPU);
exit(L1norm < 1e-6 ? EXIT_SUCCESS : EXIT_FAILURE);
}
float compareResults(int rand_n, float *h_RandGPU, float *h_RandCPU)
{
int i;
float rCPU, rGPU, delta;
float max_delta = 0.;
float sum_delta = 0.;
float sum_ref = 0.;
for (i = 0; i < rand_n; i++)
{
rCPU = h_RandCPU[i];
rGPU = h_RandGPU[i];
delta = fabs(rCPU - rGPU);
sum_delta += delta;
sum_ref += fabs(rCPU);
if (delta >= max_delta)
{
max_delta = delta;
}
}
float L1norm = (float)(sum_delta / sum_ref);
printf("Max absolute error: %E\n", max_delta);
printf("L1 norm: %E\n\n", L1norm);
return L1norm;
}

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 12.00
# Visual Studio 2012
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "MersenneTwisterGP11213", "MersenneTwisterGP11213_vs2012.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,107 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>MersenneTwisterGP11213_vs2012</RootNamespace>
<ProjectName>MersenneTwisterGP11213</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v110</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;$(CudaToolkitIncludeDir);</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>curand.lib;cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/MersenneTwisterGP11213.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_35,compute_35;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<ClCompile Include="MersenneTwister.cpp" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 13.00
# Visual Studio 2013
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "MersenneTwisterGP11213", "MersenneTwisterGP11213_vs2013.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,107 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>MersenneTwisterGP11213_vs2013</RootNamespace>
<ProjectName>MersenneTwisterGP11213</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v120</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;$(CudaToolkitIncludeDir);</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>curand.lib;cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/MersenneTwisterGP11213.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_35,compute_35;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<ClCompile Include="MersenneTwister.cpp" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 14.00
# Visual Studio 2015
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "MersenneTwisterGP11213", "MersenneTwisterGP11213_vs2015.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,107 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>MersenneTwisterGP11213_vs2015</RootNamespace>
<ProjectName>MersenneTwisterGP11213</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v140</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;$(CudaToolkitIncludeDir);</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>curand.lib;cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/MersenneTwisterGP11213.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_35,compute_35;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<ClCompile Include="MersenneTwister.cpp" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 12.00
# Visual Studio 2017
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "MersenneTwisterGP11213", "MersenneTwisterGP11213_vs2017.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,112 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>MersenneTwisterGP11213_vs2017</RootNamespace>
<ProjectName>MersenneTwisterGP11213</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<PropertyGroup Condition="'$(WindowsTargetPlatformVersion)'==''">
<LatestTargetPlatformVersion>$([Microsoft.Build.Utilities.ToolLocationHelper]::GetLatestSDKTargetPlatformVersion('Windows', '10.0'))</LatestTargetPlatformVersion>
<WindowsTargetPlatformVersion Condition="'$(WindowsTargetPlatformVersion)' == ''">$(LatestTargetPlatformVersion)</WindowsTargetPlatformVersion>
<TargetPlatformVersion>$(WindowsTargetPlatformVersion)</TargetPlatformVersion>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v141</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;$(CudaToolkitIncludeDir);</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>curand.lib;cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/MersenneTwisterGP11213.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_35,compute_35;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<ClCompile Include="MersenneTwister.cpp" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 12.00
# Visual Studio 2019
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "MersenneTwisterGP11213", "MersenneTwisterGP11213_vs2019.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,108 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>MersenneTwisterGP11213_vs2019</RootNamespace>
<ProjectName>MersenneTwisterGP11213</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v142</PlatformToolset>
<WindowsTargetPlatformVersion>10.0</WindowsTargetPlatformVersion>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;$(CudaToolkitIncludeDir);</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>curand.lib;cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/MersenneTwisterGP11213.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_35,compute_35;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<ClCompile Include="MersenneTwister.cpp" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,73 @@
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE entry SYSTEM "SamplesInfo.dtd">
<entry>
<name>MersenneTwisterGP11213</name>
<description><![CDATA[This sample demonstrates the Mersenne Twister random number generator GP11213 in cuRAND.]]></description>
<devicecompilation>whole</devicecompilation>
<fallback_min_ptx>true</fallback_min_ptx>
<includepaths>
<path>./</path>
<path>../</path>
<path>../../common/inc</path>
</includepaths>
<keyconcepts>
<concept level="basic">CURAND Library</concept>
</keyconcepts>
<keywords>
<keyword>CUDA</keyword>
<keyword>CURAND</keyword>
<keyword>Monte-Carlo</keyword>
<keyword>random number generation</keyword>
<keyword>GSFR</keyword>
<keyword>Mersenne Twister</keyword>
</keywords>
<libraries>
<library>curand_static</library>
<library>culibos</library>
</libraries>
<librarypaths>
</librarypaths>
<nsight_eclipse>true</nsight_eclipse>
<primary_file>MersenneTwister.cpp</primary_file>
<required_dependencies>
<dependency>CURAND</dependency>
</required_dependencies>
<scopes>
<scope>1:CUDA Advanced Topics</scope>
</scopes>
<sm-arch>sm35</sm-arch>
<sm-arch>sm37</sm-arch>
<sm-arch>sm50</sm-arch>
<sm-arch>sm52</sm-arch>
<sm-arch>sm60</sm-arch>
<sm-arch>sm61</sm-arch>
<sm-arch>sm70</sm-arch>
<sm-arch>sm72</sm-arch>
<sm-arch>sm75</sm-arch>
<sm-arch>sm80</sm-arch>
<supported_envs>
<env>
<arch>x86_64</arch>
<platform>linux</platform>
</env>
<env>
<platform>windows7</platform>
</env>
<env>
<arch>x86_64</arch>
<platform>macosx</platform>
</env>
<env>
<arch>arm</arch>
</env>
<env>
<arch>ppc64le</arch>
<platform>linux</platform>
</env>
</supported_envs>
<supported_sm_architectures>
<include>all</include>
</supported_sm_architectures>
<title>MersenneTwisterGP11213</title>
<type>exe</type>
</entry>

View File

@@ -0,0 +1,71 @@
# MersenneTwisterGP11213 - MersenneTwisterGP11213
## Description
This sample demonstrates the Mersenne Twister random number generator GP11213 in cuRAND.
## Key Concepts
CURAND Library
## Supported SM Architectures
[SM 3.5 ](https://developer.nvidia.com/cuda-gpus) [SM 3.7 ](https://developer.nvidia.com/cuda-gpus) [SM 5.0 ](https://developer.nvidia.com/cuda-gpus) [SM 5.2 ](https://developer.nvidia.com/cuda-gpus) [SM 6.0 ](https://developer.nvidia.com/cuda-gpus) [SM 6.1 ](https://developer.nvidia.com/cuda-gpus) [SM 7.0 ](https://developer.nvidia.com/cuda-gpus) [SM 7.2 ](https://developer.nvidia.com/cuda-gpus) [SM 7.5 ](https://developer.nvidia.com/cuda-gpus) [SM 8.0 ](https://developer.nvidia.com/cuda-gpus)
## Supported OSes
Linux, Windows
## Supported CPU Architecture
x86_64, ppc64le, armv7l
## CUDA APIs involved
## Dependencies needed to build/run
[CURAND](../../README.md#curand)
## Prerequisites
Download and install the [CUDA Toolkit 11.0](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run
### Windows
The Windows samples are built using the Visual Studio IDE. Solution files (.sln) are provided for each supported version of Visual Studio, using the format:
```
*_vs<version>.sln - for Visual Studio <version>
```
Each individual sample has its own set of solution files in its directory:
To build/examine all the samples at once, the complete solution files should be used. To build/examine a single sample, the individual sample solution files should be used.
> **Note:** Some samples require that the Microsoft DirectX SDK (June 2010 or newer) be installed and that the VC++ directory paths are properly set up (**Tools > Options...**). Check DirectX Dependencies section for details."
### Linux
The Linux samples are built using makefiles. To use the makefiles, change the current directory to the sample directory you wish to build, and run make:
```
$ cd <sample_dir>
$ make
```
The samples makefiles can take advantage of certain options:
* **TARGET_ARCH=<arch>** - cross-compile targeting a specific architecture. Allowed architectures are x86_64, ppc64le, armv7l.
By default, TARGET_ARCH is set to HOST_ARCH. On a x86_64 machine, not setting TARGET_ARCH is the equivalent of setting TARGET_ARCH=x86_64.<br/>
`$ make TARGET_ARCH=x86_64` <br/> `$ make TARGET_ARCH=ppc64le` <br/> `$ make TARGET_ARCH=armv7l` <br/>
See [here](http://docs.nvidia.com/cuda/cuda-samples/index.html#cross-samples) for more details.
* **dbg=1** - build with debug symbols
```
$ make dbg=1
```
* **SMS="A B ..."** - override the SM architectures for which the sample will be built, where `"A B ..."` is a space-delimited list of SM architectures. For example, to generate SASS for SM 50 and SM 60, use `SMS="50 60"`.
```
$ make SMS="50 60"
```
* **HOST_COMPILER=<host_compiler>** - override the default g++ host compiler. See the [Linux Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#system-requirements) for a list of supported host compilers.
```
$ make HOST_COMPILER=g++
```
## References (for more details)

View File

@@ -72,9 +72,9 @@ endif
# architecture # architecture
HOST_ARCH := $(shell uname -m) HOST_ARCH := $(shell uname -m)
TARGET_ARCH ?= $(HOST_ARCH) TARGET_ARCH ?= $(HOST_ARCH)
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le armv7l)) ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le armv7l))
ifneq ($(TARGET_ARCH),$(HOST_ARCH)) ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le)) ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le))
TARGET_SIZE := 64 TARGET_SIZE := 64
else ifneq (,$(filter $(TARGET_ARCH),armv7l)) else ifneq (,$(filter $(TARGET_ARCH),armv7l))
TARGET_SIZE := 32 TARGET_SIZE := 32
@@ -85,8 +85,17 @@ ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le armv7l))
else else
$(error ERROR - unsupported value $(TARGET_ARCH) for TARGET_ARCH!) $(error ERROR - unsupported value $(TARGET_ARCH) for TARGET_ARCH!)
endif endif
# sbsa and aarch64 systems look similar. Need to differentiate them at host level for now.
ifeq ($(HOST_ARCH),aarch64)
ifeq ($(CUDA_PATH)/targets/sbsa-linux,$(shell ls -1d $(CUDA_PATH)/targets/sbsa-linux))
HOST_ARCH := sbsa
TARGET_ARCH := sbsa
endif
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH)) ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-ppc64le)) ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-sbsa x86_64-ppc64le))
$(error ERROR - cross compiling from $(HOST_ARCH) to $(TARGET_ARCH) is not supported!) $(error ERROR - cross compiling from $(HOST_ARCH) to $(TARGET_ARCH) is not supported!)
endif endif
endif endif
@@ -141,6 +150,8 @@ else ifneq ($(TARGET_ARCH),$(HOST_ARCH))
else ifeq ($(TARGET_OS), android) else ifeq ($(TARGET_OS), android)
HOST_COMPILER ?= aarch64-linux-android-clang++ HOST_COMPILER ?= aarch64-linux-android-clang++
endif endif
else ifeq ($(TARGET_ARCH),sbsa)
HOST_COMPILER ?= aarch64-linux-gnu-g++
else ifeq ($(TARGET_ARCH),ppc64le) else ifeq ($(TARGET_ARCH),ppc64le)
HOST_COMPILER ?= powerpc64le-linux-gnu-g++ HOST_COMPILER ?= powerpc64le-linux-gnu-g++
endif endif
@@ -185,19 +196,27 @@ ifneq ($(TARGET_ARCH),$(HOST_ARCH))
CCFLAGS += --sysroot=$(TARGET_FS) CCFLAGS += --sysroot=$(TARGET_FS)
endif endif
LDFLAGS += --sysroot=$(TARGET_FS) LDFLAGS += --sysroot=$(TARGET_FS)
LDFLAGS += -rpath-link=$(TARGET_FS)/lib -L $(TARGET_FS)/lib LDFLAGS += -rpath-link=$(TARGET_FS)/lib -L$(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L $(TARGET_FS)/usr/lib LDFLAGS += -rpath-link=$(TARGET_FS)/lib/aarch64-linux-gnu -L$(TARGET_FS)/lib/aarch64-linux-gnu
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L $(TARGET_FS)/usr/lib/aarch64-linux-gnu LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L$(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L$(TARGET_FS)/usr/lib/aarch64-linux-gnu
LDFLAGS += --unresolved-symbols=ignore-in-shared-libs LDFLAGS += --unresolved-symbols=ignore-in-shared-libs
CCFLAGS += -isystem=$(TARGET_FS)/usr/include CCFLAGS += -isystem=$(TARGET_FS)/usr/include -I$(TARGET_FS)/usr/include
CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu -I$(TARGET_FS)/usr/include/aarch64-linux-gnu
endif
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
CCFLAGS += -DWIN_INTERFACE_CUSTOM -I/usr/include/aarch64-qnx-gnu
LDFLAGS += -lsocket
LDFLAGS += -rpath=/usr/lib/aarch64-qnx-gnu -L/usr/lib/aarch64-qnx-gnu
ifneq ($(TARGET_FS),)
LDFLAGS += -rpath=$(TARGET_FS)/usr/lib -L $(TARGET_FS)/usr/lib
LDFLAGS += -rpath=$(TARGET_FS)/usr/libnvidia -L $(TARGET_FS)/usr/libnvidia
endif
ifdef TARGET_OVERRIDE # cuda toolkit targets override
NVCCFLAGS += -target-dir $(TARGET_OVERRIDE)
endif endif
endif endif
endif
ifeq ($(TARGET_OS),qnx)
CCFLAGS += -DWIN_INTERFACE_CUSTOM
LDFLAGS += -lsocket
endif endif
# Install directory of different arch # Install directory of different arch
@@ -206,6 +225,8 @@ ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-gnueabihf/ CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-gnueabihf/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux) else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux/ CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),sbsa-linux)
CUDA_INSTALL_TARGET_DIR = targets/sbsa-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android) else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-androideabi/ CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android) else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android)
@@ -253,9 +274,9 @@ LIBRARIES :=
# Gencode arguments # Gencode arguments
ifeq ($(TARGET_ARCH),$(filter $(TARGET_ARCH),armv7l aarch64)) ifeq ($(TARGET_ARCH),$(filter $(TARGET_ARCH),armv7l aarch64))
SMS ?= 30 35 37 50 52 60 61 70 72 75 SMS ?= 35 37 50 52 60 61 70 72 75 80
else else
SMS ?= 30 35 37 50 52 60 61 70 75 SMS ?= 35 37 50 52 60 61 70 75 80
endif endif
ifeq ($(SMS),) ifeq ($(SMS),)

View File

@@ -33,7 +33,7 @@
</PropertyGroup> </PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings"> <ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup> </ImportGroup>
<ImportGroup Label="PropertySheets"> <ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" /> <Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@@ -62,7 +62,7 @@
<OutputFile>$(OutDir)/NV12toBGRandResize.exe</OutputFile> <OutputFile>$(OutDir)/NV12toBGRandResize.exe</OutputFile>
</Link> </Link>
<CudaCompile> <CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration> <CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions> <AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include> <Include>./;../../Common</Include>
<Defines>WIN32</Defines> <Defines>WIN32</Defines>
@@ -107,6 +107,6 @@
</ItemGroup> </ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets"> <ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup> </ImportGroup>
</Project> </Project>

View File

@@ -33,7 +33,7 @@
</PropertyGroup> </PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings"> <ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup> </ImportGroup>
<ImportGroup Label="PropertySheets"> <ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" /> <Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@@ -62,7 +62,7 @@
<OutputFile>$(OutDir)/NV12toBGRandResize.exe</OutputFile> <OutputFile>$(OutDir)/NV12toBGRandResize.exe</OutputFile>
</Link> </Link>
<CudaCompile> <CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration> <CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions> <AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include> <Include>./;../../Common</Include>
<Defines>WIN32</Defines> <Defines>WIN32</Defines>
@@ -107,6 +107,6 @@
</ItemGroup> </ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets"> <ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup> </ImportGroup>
</Project> </Project>

View File

@@ -33,7 +33,7 @@
</PropertyGroup> </PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings"> <ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup> </ImportGroup>
<ImportGroup Label="PropertySheets"> <ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" /> <Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@@ -62,7 +62,7 @@
<OutputFile>$(OutDir)/NV12toBGRandResize.exe</OutputFile> <OutputFile>$(OutDir)/NV12toBGRandResize.exe</OutputFile>
</Link> </Link>
<CudaCompile> <CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration> <CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions> <AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include> <Include>./;../../Common</Include>
<Defines>WIN32</Defines> <Defines>WIN32</Defines>
@@ -107,6 +107,6 @@
</ItemGroup> </ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets"> <ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup> </ImportGroup>
</Project> </Project>

View File

@@ -38,7 +38,7 @@
</PropertyGroup> </PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings"> <ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup> </ImportGroup>
<ImportGroup Label="PropertySheets"> <ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" /> <Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@@ -67,7 +67,7 @@
<OutputFile>$(OutDir)/NV12toBGRandResize.exe</OutputFile> <OutputFile>$(OutDir)/NV12toBGRandResize.exe</OutputFile>
</Link> </Link>
<CudaCompile> <CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration> <CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions> <AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include> <Include>./;../../Common</Include>
<Defines>WIN32</Defines> <Defines>WIN32</Defines>
@@ -112,6 +112,6 @@
</ItemGroup> </ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets"> <ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup> </ImportGroup>
</Project> </Project>

View File

@@ -34,7 +34,7 @@
</PropertyGroup> </PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings"> <ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup> </ImportGroup>
<ImportGroup Label="PropertySheets"> <ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" /> <Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@@ -63,7 +63,7 @@
<OutputFile>$(OutDir)/NV12toBGRandResize.exe</OutputFile> <OutputFile>$(OutDir)/NV12toBGRandResize.exe</OutputFile>
</Link> </Link>
<CudaCompile> <CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration> <CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions> <AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include> <Include>./;../../Common</Include>
<Defines>WIN32</Defines> <Defines>WIN32</Defines>
@@ -108,6 +108,6 @@
</ItemGroup> </ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets"> <ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup> </ImportGroup>
</Project> </Project>

View File

@@ -32,7 +32,6 @@
<scope>2:Image Processing</scope> <scope>2:Image Processing</scope>
<scope>2:Computer Vision</scope> <scope>2:Computer Vision</scope>
</scopes> </scopes>
<sm-arch>sm30</sm-arch>
<sm-arch>sm35</sm-arch> <sm-arch>sm35</sm-arch>
<sm-arch>sm37</sm-arch> <sm-arch>sm37</sm-arch>
<sm-arch>sm50</sm-arch> <sm-arch>sm50</sm-arch>
@@ -42,6 +41,7 @@
<sm-arch>sm70</sm-arch> <sm-arch>sm70</sm-arch>
<sm-arch>sm72</sm-arch> <sm-arch>sm72</sm-arch>
<sm-arch>sm75</sm-arch> <sm-arch>sm75</sm-arch>
<sm-arch>sm80</sm-arch>
<supported_envs> <supported_envs>
<env> <env>
<arch>x86_64</arch> <arch>x86_64</arch>

View File

@@ -10,11 +10,11 @@ Graphics Interop, Image Processing, Video Processing
## Supported SM Architectures ## Supported SM Architectures
[SM 3.0 ](https://developer.nvidia.com/cuda-gpus) [SM 3.5 ](https://developer.nvidia.com/cuda-gpus) [SM 3.7 ](https://developer.nvidia.com/cuda-gpus) [SM 5.0 ](https://developer.nvidia.com/cuda-gpus) [SM 5.2 ](https://developer.nvidia.com/cuda-gpus) [SM 6.0 ](https://developer.nvidia.com/cuda-gpus) [SM 6.1 ](https://developer.nvidia.com/cuda-gpus) [SM 7.0 ](https://developer.nvidia.com/cuda-gpus) [SM 7.2 ](https://developer.nvidia.com/cuda-gpus) [SM 7.5 ](https://developer.nvidia.com/cuda-gpus) [SM 3.5 ](https://developer.nvidia.com/cuda-gpus) [SM 3.7 ](https://developer.nvidia.com/cuda-gpus) [SM 5.0 ](https://developer.nvidia.com/cuda-gpus) [SM 5.2 ](https://developer.nvidia.com/cuda-gpus) [SM 6.0 ](https://developer.nvidia.com/cuda-gpus) [SM 6.1 ](https://developer.nvidia.com/cuda-gpus) [SM 7.0 ](https://developer.nvidia.com/cuda-gpus) [SM 7.2 ](https://developer.nvidia.com/cuda-gpus) [SM 7.5 ](https://developer.nvidia.com/cuda-gpus) [SM 8.0 ](https://developer.nvidia.com/cuda-gpus)
## Supported OSes ## Supported OSes
Linux, Windows, MacOSX Linux, Windows
## Supported CPU Architecture ## Supported CPU Architecture
@@ -27,7 +27,7 @@ cudaMemcpy2D, cudaMallocManaged
## Prerequisites ## Prerequisites
Download and install the [CUDA Toolkit 10.1](https://developer.nvidia.com/cuda-downloads) for your corresponding platform. Download and install the [CUDA Toolkit 11.0](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
## Build and Run ## Build and Run
@@ -66,29 +66,5 @@ The samples makefiles can take advantage of certain options:
$ make HOST_COMPILER=g++ $ make HOST_COMPILER=g++
``` ```
### Mac
The Mac samples are built using makefiles. To use the makefiles, change directory into the sample directory you wish to build, and run make:
```
$ cd <sample_dir>
$ make
```
The samples makefiles can take advantage of certain options:
* **dbg=1** - build with debug symbols
```
$ make dbg=1
```
* **SMS="A B ..."** - override the SM architectures for which the sample will be built, where "A B ..." is a space-delimited list of SM architectures. For example, to generate SASS for SM 50 and SM 60, use SMS="50 60".
```
$ make SMS="A B ..."
```
* **HOST_COMPILER=<host_compiler>** - override the default clang host compiler. See the [Mac Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-mac-os-x/index.html#system-requirements) for a list of supported host compilers.
```
$ make HOST_COMPILER=clang
```
## References (for more details) ## References (for more details)

View File

@@ -72,9 +72,9 @@ endif
# architecture # architecture
HOST_ARCH := $(shell uname -m) HOST_ARCH := $(shell uname -m)
TARGET_ARCH ?= $(HOST_ARCH) TARGET_ARCH ?= $(HOST_ARCH)
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le armv7l)) ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le armv7l))
ifneq ($(TARGET_ARCH),$(HOST_ARCH)) ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le)) ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le))
TARGET_SIZE := 64 TARGET_SIZE := 64
else ifneq (,$(filter $(TARGET_ARCH),armv7l)) else ifneq (,$(filter $(TARGET_ARCH),armv7l))
TARGET_SIZE := 32 TARGET_SIZE := 32
@@ -85,8 +85,17 @@ ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le armv7l))
else else
$(error ERROR - unsupported value $(TARGET_ARCH) for TARGET_ARCH!) $(error ERROR - unsupported value $(TARGET_ARCH) for TARGET_ARCH!)
endif endif
# sbsa and aarch64 systems look similar. Need to differentiate them at host level for now.
ifeq ($(HOST_ARCH),aarch64)
ifeq ($(CUDA_PATH)/targets/sbsa-linux,$(shell ls -1d $(CUDA_PATH)/targets/sbsa-linux))
HOST_ARCH := sbsa
TARGET_ARCH := sbsa
endif
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH)) ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-ppc64le)) ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-sbsa x86_64-ppc64le))
$(error ERROR - cross compiling from $(HOST_ARCH) to $(TARGET_ARCH) is not supported!) $(error ERROR - cross compiling from $(HOST_ARCH) to $(TARGET_ARCH) is not supported!)
endif endif
endif endif
@@ -141,6 +150,8 @@ else ifneq ($(TARGET_ARCH),$(HOST_ARCH))
else ifeq ($(TARGET_OS), android) else ifeq ($(TARGET_OS), android)
HOST_COMPILER ?= aarch64-linux-android-clang++ HOST_COMPILER ?= aarch64-linux-android-clang++
endif endif
else ifeq ($(TARGET_ARCH),sbsa)
HOST_COMPILER ?= aarch64-linux-gnu-g++
else ifeq ($(TARGET_ARCH),ppc64le) else ifeq ($(TARGET_ARCH),ppc64le)
HOST_COMPILER ?= powerpc64le-linux-gnu-g++ HOST_COMPILER ?= powerpc64le-linux-gnu-g++
endif endif
@@ -185,19 +196,27 @@ ifneq ($(TARGET_ARCH),$(HOST_ARCH))
CCFLAGS += --sysroot=$(TARGET_FS) CCFLAGS += --sysroot=$(TARGET_FS)
endif endif
LDFLAGS += --sysroot=$(TARGET_FS) LDFLAGS += --sysroot=$(TARGET_FS)
LDFLAGS += -rpath-link=$(TARGET_FS)/lib -L $(TARGET_FS)/lib LDFLAGS += -rpath-link=$(TARGET_FS)/lib -L$(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L $(TARGET_FS)/usr/lib LDFLAGS += -rpath-link=$(TARGET_FS)/lib/aarch64-linux-gnu -L$(TARGET_FS)/lib/aarch64-linux-gnu
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L $(TARGET_FS)/usr/lib/aarch64-linux-gnu LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L$(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L$(TARGET_FS)/usr/lib/aarch64-linux-gnu
LDFLAGS += --unresolved-symbols=ignore-in-shared-libs LDFLAGS += --unresolved-symbols=ignore-in-shared-libs
CCFLAGS += -isystem=$(TARGET_FS)/usr/include CCFLAGS += -isystem=$(TARGET_FS)/usr/include -I$(TARGET_FS)/usr/include
CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu -I$(TARGET_FS)/usr/include/aarch64-linux-gnu
endif
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
CCFLAGS += -DWIN_INTERFACE_CUSTOM -I/usr/include/aarch64-qnx-gnu
LDFLAGS += -lsocket
LDFLAGS += -rpath=/usr/lib/aarch64-qnx-gnu -L/usr/lib/aarch64-qnx-gnu
ifneq ($(TARGET_FS),)
LDFLAGS += -rpath=$(TARGET_FS)/usr/lib -L $(TARGET_FS)/usr/lib
LDFLAGS += -rpath=$(TARGET_FS)/usr/libnvidia -L $(TARGET_FS)/usr/libnvidia
endif
ifdef TARGET_OVERRIDE # cuda toolkit targets override
NVCCFLAGS += -target-dir $(TARGET_OVERRIDE)
endif endif
endif endif
endif
ifeq ($(TARGET_OS),qnx)
CCFLAGS += -DWIN_INTERFACE_CUSTOM
LDFLAGS += -lsocket
endif endif
# Install directory of different arch # Install directory of different arch
@@ -206,6 +225,8 @@ ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-gnueabihf/ CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-gnueabihf/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux) else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux/ CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),sbsa-linux)
CUDA_INSTALL_TARGET_DIR = targets/sbsa-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android) else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-androideabi/ CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android) else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android)
@@ -253,9 +274,9 @@ LIBRARIES :=
# Gencode arguments # Gencode arguments
ifeq ($(TARGET_ARCH),$(filter $(TARGET_ARCH),armv7l aarch64)) ifeq ($(TARGET_ARCH),$(filter $(TARGET_ARCH),armv7l aarch64))
SMS ?= 30 35 37 50 52 60 61 70 72 75 SMS ?= 35 37 50 52 60 61 70 72 75 80
else else
SMS ?= 30 35 37 50 52 60 61 70 75 SMS ?= 35 37 50 52 60 61 70 75 80
endif endif
ifeq ($(SMS),) ifeq ($(SMS),)

View File

@@ -44,16 +44,6 @@
<scope>1:CUDA Systems Integration</scope> <scope>1:CUDA Systems Integration</scope>
<scope>1:Unified Memory</scope> <scope>1:Unified Memory</scope>
</scopes> </scopes>
<sm-arch>sm30</sm-arch>
<sm-arch>sm35</sm-arch>
<sm-arch>sm37</sm-arch>
<sm-arch>sm50</sm-arch>
<sm-arch>sm52</sm-arch>
<sm-arch>sm60</sm-arch>
<sm-arch>sm61</sm-arch>
<sm-arch>sm70</sm-arch>
<sm-arch>sm72</sm-arch>
<sm-arch>sm75</sm-arch>
<supported_envs> <supported_envs>
<env> <env>
<arch>x86_64</arch> <arch>x86_64</arch>

View File

@@ -10,11 +10,9 @@ CUDA Systems Integration, Unified Memory, CUDA Streams and Events, Pinned System
## Supported SM Architectures ## Supported SM Architectures
[SM 3.0 ](https://developer.nvidia.com/cuda-gpus) [SM 3.5 ](https://developer.nvidia.com/cuda-gpus) [SM 3.7 ](https://developer.nvidia.com/cuda-gpus) [SM 5.0 ](https://developer.nvidia.com/cuda-gpus) [SM 5.2 ](https://developer.nvidia.com/cuda-gpus) [SM 6.0 ](https://developer.nvidia.com/cuda-gpus) [SM 6.1 ](https://developer.nvidia.com/cuda-gpus) [SM 7.0 ](https://developer.nvidia.com/cuda-gpus) [SM 7.2 ](https://developer.nvidia.com/cuda-gpus) [SM 7.5 ](https://developer.nvidia.com/cuda-gpus)
## Supported OSes ## Supported OSes
Linux, Windows, MacOSX Linux, Windows
## Supported CPU Architecture ## Supported CPU Architecture
@@ -30,7 +28,7 @@ cudaMallocManaged, cudaStreamAttachMemAsync, cudaMemcpyAsync, cudaMallocHost, cu
## Prerequisites ## Prerequisites
Download and install the [CUDA Toolkit 10.1](https://developer.nvidia.com/cuda-downloads) for your corresponding platform. Download and install the [CUDA Toolkit 11.0](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed. Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run ## Build and Run
@@ -70,29 +68,5 @@ The samples makefiles can take advantage of certain options:
$ make HOST_COMPILER=g++ $ make HOST_COMPILER=g++
``` ```
### Mac
The Mac samples are built using makefiles. To use the makefiles, change directory into the sample directory you wish to build, and run make:
```
$ cd <sample_dir>
$ make
```
The samples makefiles can take advantage of certain options:
* **dbg=1** - build with debug symbols
```
$ make dbg=1
```
* **SMS="A B ..."** - override the SM architectures for which the sample will be built, where "A B ..." is a space-delimited list of SM architectures. For example, to generate SASS for SM 50 and SM 60, use SMS="50 60".
```
$ make SMS="A B ..."
```
* **HOST_COMPILER=<host_compiler>** - override the default clang host compiler. See the [Mac Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-mac-os-x/index.html#system-requirements) for a list of supported host compilers.
```
$ make HOST_COMPILER=clang
```
## References (for more details) ## References (for more details)

View File

@@ -33,7 +33,7 @@
</PropertyGroup> </PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings"> <ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup> </ImportGroup>
<ImportGroup Label="PropertySheets"> <ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" /> <Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@@ -62,7 +62,7 @@
<OutputFile>$(OutDir)/UnifiedMemoryPerf.exe</OutputFile> <OutputFile>$(OutDir)/UnifiedMemoryPerf.exe</OutputFile>
</Link> </Link>
<CudaCompile> <CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration> <CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions> <AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include> <Include>./;../../Common</Include>
<Defines>WIN32</Defines> <Defines>WIN32</Defines>
@@ -105,6 +105,6 @@
</ItemGroup> </ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets"> <ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup> </ImportGroup>
</Project> </Project>

View File

@@ -33,7 +33,7 @@
</PropertyGroup> </PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings"> <ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup> </ImportGroup>
<ImportGroup Label="PropertySheets"> <ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" /> <Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@@ -62,7 +62,7 @@
<OutputFile>$(OutDir)/UnifiedMemoryPerf.exe</OutputFile> <OutputFile>$(OutDir)/UnifiedMemoryPerf.exe</OutputFile>
</Link> </Link>
<CudaCompile> <CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration> <CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions> <AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include> <Include>./;../../Common</Include>
<Defines>WIN32</Defines> <Defines>WIN32</Defines>
@@ -105,6 +105,6 @@
</ItemGroup> </ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets"> <ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup> </ImportGroup>
</Project> </Project>

View File

@@ -33,7 +33,7 @@
</PropertyGroup> </PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings"> <ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup> </ImportGroup>
<ImportGroup Label="PropertySheets"> <ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" /> <Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@@ -62,7 +62,7 @@
<OutputFile>$(OutDir)/UnifiedMemoryPerf.exe</OutputFile> <OutputFile>$(OutDir)/UnifiedMemoryPerf.exe</OutputFile>
</Link> </Link>
<CudaCompile> <CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration> <CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions> <AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include> <Include>./;../../Common</Include>
<Defines>WIN32</Defines> <Defines>WIN32</Defines>
@@ -105,6 +105,6 @@
</ItemGroup> </ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets"> <ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup> </ImportGroup>
</Project> </Project>

View File

@@ -38,7 +38,7 @@
</PropertyGroup> </PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings"> <ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup> </ImportGroup>
<ImportGroup Label="PropertySheets"> <ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" /> <Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@@ -67,7 +67,7 @@
<OutputFile>$(OutDir)/UnifiedMemoryPerf.exe</OutputFile> <OutputFile>$(OutDir)/UnifiedMemoryPerf.exe</OutputFile>
</Link> </Link>
<CudaCompile> <CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration> <CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions> <AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include> <Include>./;../../Common</Include>
<Defines>WIN32</Defines> <Defines>WIN32</Defines>
@@ -110,6 +110,6 @@
</ItemGroup> </ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets"> <ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup> </ImportGroup>
</Project> </Project>

View File

@@ -34,7 +34,7 @@
</PropertyGroup> </PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings"> <ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup> </ImportGroup>
<ImportGroup Label="PropertySheets"> <ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" /> <Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@@ -63,7 +63,7 @@
<OutputFile>$(OutDir)/UnifiedMemoryPerf.exe</OutputFile> <OutputFile>$(OutDir)/UnifiedMemoryPerf.exe</OutputFile>
</Link> </Link>
<CudaCompile> <CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration> <CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions> <AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include> <Include>./;../../Common</Include>
<Defines>WIN32</Defines> <Defines>WIN32</Defines>
@@ -106,6 +106,6 @@
</ItemGroup> </ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets"> <ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup> </ImportGroup>
</Project> </Project>

View File

@@ -72,9 +72,9 @@ endif
# architecture # architecture
HOST_ARCH := $(shell uname -m) HOST_ARCH := $(shell uname -m)
TARGET_ARCH ?= $(HOST_ARCH) TARGET_ARCH ?= $(HOST_ARCH)
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le armv7l)) ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le armv7l))
ifneq ($(TARGET_ARCH),$(HOST_ARCH)) ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le)) ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le))
TARGET_SIZE := 64 TARGET_SIZE := 64
else ifneq (,$(filter $(TARGET_ARCH),armv7l)) else ifneq (,$(filter $(TARGET_ARCH),armv7l))
TARGET_SIZE := 32 TARGET_SIZE := 32
@@ -85,8 +85,17 @@ ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le armv7l))
else else
$(error ERROR - unsupported value $(TARGET_ARCH) for TARGET_ARCH!) $(error ERROR - unsupported value $(TARGET_ARCH) for TARGET_ARCH!)
endif endif
# sbsa and aarch64 systems look similar. Need to differentiate them at host level for now.
ifeq ($(HOST_ARCH),aarch64)
ifeq ($(CUDA_PATH)/targets/sbsa-linux,$(shell ls -1d $(CUDA_PATH)/targets/sbsa-linux))
HOST_ARCH := sbsa
TARGET_ARCH := sbsa
endif
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH)) ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-ppc64le)) ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-sbsa x86_64-ppc64le))
$(error ERROR - cross compiling from $(HOST_ARCH) to $(TARGET_ARCH) is not supported!) $(error ERROR - cross compiling from $(HOST_ARCH) to $(TARGET_ARCH) is not supported!)
endif endif
endif endif
@@ -141,6 +150,8 @@ else ifneq ($(TARGET_ARCH),$(HOST_ARCH))
else ifeq ($(TARGET_OS), android) else ifeq ($(TARGET_OS), android)
HOST_COMPILER ?= aarch64-linux-android-clang++ HOST_COMPILER ?= aarch64-linux-android-clang++
endif endif
else ifeq ($(TARGET_ARCH),sbsa)
HOST_COMPILER ?= aarch64-linux-gnu-g++
else ifeq ($(TARGET_ARCH),ppc64le) else ifeq ($(TARGET_ARCH),ppc64le)
HOST_COMPILER ?= powerpc64le-linux-gnu-g++ HOST_COMPILER ?= powerpc64le-linux-gnu-g++
endif endif
@@ -185,19 +196,27 @@ ifneq ($(TARGET_ARCH),$(HOST_ARCH))
CCFLAGS += --sysroot=$(TARGET_FS) CCFLAGS += --sysroot=$(TARGET_FS)
endif endif
LDFLAGS += --sysroot=$(TARGET_FS) LDFLAGS += --sysroot=$(TARGET_FS)
LDFLAGS += -rpath-link=$(TARGET_FS)/lib -L $(TARGET_FS)/lib LDFLAGS += -rpath-link=$(TARGET_FS)/lib -L$(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L $(TARGET_FS)/usr/lib LDFLAGS += -rpath-link=$(TARGET_FS)/lib/aarch64-linux-gnu -L$(TARGET_FS)/lib/aarch64-linux-gnu
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L $(TARGET_FS)/usr/lib/aarch64-linux-gnu LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L$(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L$(TARGET_FS)/usr/lib/aarch64-linux-gnu
LDFLAGS += --unresolved-symbols=ignore-in-shared-libs LDFLAGS += --unresolved-symbols=ignore-in-shared-libs
CCFLAGS += -isystem=$(TARGET_FS)/usr/include CCFLAGS += -isystem=$(TARGET_FS)/usr/include -I$(TARGET_FS)/usr/include
CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu -I$(TARGET_FS)/usr/include/aarch64-linux-gnu
endif
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
CCFLAGS += -DWIN_INTERFACE_CUSTOM -I/usr/include/aarch64-qnx-gnu
LDFLAGS += -lsocket
LDFLAGS += -rpath=/usr/lib/aarch64-qnx-gnu -L/usr/lib/aarch64-qnx-gnu
ifneq ($(TARGET_FS),)
LDFLAGS += -rpath=$(TARGET_FS)/usr/lib -L $(TARGET_FS)/usr/lib
LDFLAGS += -rpath=$(TARGET_FS)/usr/libnvidia -L $(TARGET_FS)/usr/libnvidia
endif
ifdef TARGET_OVERRIDE # cuda toolkit targets override
NVCCFLAGS += -target-dir $(TARGET_OVERRIDE)
endif endif
endif endif
endif
ifeq ($(TARGET_OS),qnx)
CCFLAGS += -DWIN_INTERFACE_CUSTOM
LDFLAGS += -lsocket
endif endif
# Install directory of different arch # Install directory of different arch
@@ -206,6 +225,8 @@ ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-gnueabihf/ CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-gnueabihf/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux) else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux/ CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),sbsa-linux)
CUDA_INSTALL_TARGET_DIR = targets/sbsa-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android) else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-androideabi/ CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android) else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android)
@@ -247,9 +268,9 @@ LIBRARIES :=
# Gencode arguments # Gencode arguments
ifeq ($(TARGET_ARCH),$(filter $(TARGET_ARCH),armv7l aarch64)) ifeq ($(TARGET_ARCH),$(filter $(TARGET_ARCH),armv7l aarch64))
SMS ?= 30 35 37 50 52 60 61 70 72 75 SMS ?= 35 37 50 52 60 61 70 72 75 80
else else
SMS ?= 30 35 37 50 52 60 61 70 75 SMS ?= 35 37 50 52 60 61 70 75 80
endif endif
ifeq ($(SMS),) ifeq ($(SMS),)

View File

@@ -41,7 +41,6 @@
<scope>1:CUDA Basic Topics</scope> <scope>1:CUDA Basic Topics</scope>
<scope>1:Performance Strategies</scope> <scope>1:Performance Strategies</scope>
</scopes> </scopes>
<sm-arch>sm30</sm-arch>
<sm-arch>sm35</sm-arch> <sm-arch>sm35</sm-arch>
<sm-arch>sm37</sm-arch> <sm-arch>sm37</sm-arch>
<sm-arch>sm50</sm-arch> <sm-arch>sm50</sm-arch>
@@ -51,6 +50,7 @@
<sm-arch>sm70</sm-arch> <sm-arch>sm70</sm-arch>
<sm-arch>sm72</sm-arch> <sm-arch>sm72</sm-arch>
<sm-arch>sm75</sm-arch> <sm-arch>sm75</sm-arch>
<sm-arch>sm80</sm-arch>
<supported_envs> <supported_envs>
<env> <env>
<arch>x86_64</arch> <arch>x86_64</arch>

View File

@@ -10,11 +10,11 @@ CUDA Streams and Events, Performance Strategies
## Supported SM Architectures ## Supported SM Architectures
[SM 3.0 ](https://developer.nvidia.com/cuda-gpus) [SM 3.5 ](https://developer.nvidia.com/cuda-gpus) [SM 3.7 ](https://developer.nvidia.com/cuda-gpus) [SM 5.0 ](https://developer.nvidia.com/cuda-gpus) [SM 5.2 ](https://developer.nvidia.com/cuda-gpus) [SM 6.0 ](https://developer.nvidia.com/cuda-gpus) [SM 6.1 ](https://developer.nvidia.com/cuda-gpus) [SM 7.0 ](https://developer.nvidia.com/cuda-gpus) [SM 7.2 ](https://developer.nvidia.com/cuda-gpus) [SM 7.5 ](https://developer.nvidia.com/cuda-gpus) [SM 3.5 ](https://developer.nvidia.com/cuda-gpus) [SM 3.7 ](https://developer.nvidia.com/cuda-gpus) [SM 5.0 ](https://developer.nvidia.com/cuda-gpus) [SM 5.2 ](https://developer.nvidia.com/cuda-gpus) [SM 6.0 ](https://developer.nvidia.com/cuda-gpus) [SM 6.1 ](https://developer.nvidia.com/cuda-gpus) [SM 7.0 ](https://developer.nvidia.com/cuda-gpus) [SM 7.2 ](https://developer.nvidia.com/cuda-gpus) [SM 7.5 ](https://developer.nvidia.com/cuda-gpus) [SM 8.0 ](https://developer.nvidia.com/cuda-gpus)
## Supported OSes ## Supported OSes
Linux, Windows, MacOSX Linux, Windows
## Supported CPU Architecture ## Supported CPU Architecture
@@ -27,7 +27,7 @@ cudaSetDevice, cudaHostAlloc, cudaFree, cudaMallocHost, cudaFreeHost, cudaMemcpy
## Prerequisites ## Prerequisites
Download and install the [CUDA Toolkit 10.1](https://developer.nvidia.com/cuda-downloads) for your corresponding platform. Download and install the [CUDA Toolkit 11.0](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
## Build and Run ## Build and Run
@@ -66,29 +66,5 @@ The samples makefiles can take advantage of certain options:
$ make HOST_COMPILER=g++ $ make HOST_COMPILER=g++
``` ```
### Mac
The Mac samples are built using makefiles. To use the makefiles, change directory into the sample directory you wish to build, and run make:
```
$ cd <sample_dir>
$ make
```
The samples makefiles can take advantage of certain options:
* **dbg=1** - build with debug symbols
```
$ make dbg=1
```
* **SMS="A B ..."** - override the SM architectures for which the sample will be built, where "A B ..." is a space-delimited list of SM architectures. For example, to generate SASS for SM 50 and SM 60, use SMS="50 60".
```
$ make SMS="A B ..."
```
* **HOST_COMPILER=<host_compiler>** - override the default clang host compiler. See the [Mac Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-mac-os-x/index.html#system-requirements) for a list of supported host compilers.
```
$ make HOST_COMPILER=clang
```
## References (for more details) ## References (for more details)

View File

@@ -915,7 +915,7 @@ void printResultsCSV(unsigned int *memSizes, double *bandwidths,
double dSeconds = 0.0; double dSeconds = 0.0;
for (i = 0; i < count; i++) { for (i = 0; i < count; i++) {
dSeconds = (double)memSizes[i] / (bandwidths[i] * (double)(1 << 20)); dSeconds = (double)memSizes[i] / (bandwidths[i] * (double)(1e9));
printf( printf(
"bandwidthTest-%s, Bandwidth = %.1f GB/s, Time = %.5f s, Size = %u " "bandwidthTest-%s, Bandwidth = %.1f GB/s, Time = %.5f s, Size = %u "
"bytes, NumDevsUsed = %d\n", "bytes, NumDevsUsed = %d\n",

View File

@@ -33,7 +33,7 @@
</PropertyGroup> </PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings"> <ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup> </ImportGroup>
<ImportGroup Label="PropertySheets"> <ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" /> <Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@@ -62,7 +62,7 @@
<OutputFile>$(OutDir)/bandwidthTest.exe</OutputFile> <OutputFile>$(OutDir)/bandwidthTest.exe</OutputFile>
</Link> </Link>
<CudaCompile> <CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration> <CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions> <AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include> <Include>./;../../Common</Include>
<Defines>WIN32</Defines> <Defines>WIN32</Defines>
@@ -102,6 +102,6 @@
</ItemGroup> </ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets"> <ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup> </ImportGroup>
</Project> </Project>

View File

@@ -33,7 +33,7 @@
</PropertyGroup> </PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings"> <ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup> </ImportGroup>
<ImportGroup Label="PropertySheets"> <ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" /> <Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@@ -62,7 +62,7 @@
<OutputFile>$(OutDir)/bandwidthTest.exe</OutputFile> <OutputFile>$(OutDir)/bandwidthTest.exe</OutputFile>
</Link> </Link>
<CudaCompile> <CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration> <CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions> <AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include> <Include>./;../../Common</Include>
<Defines>WIN32</Defines> <Defines>WIN32</Defines>
@@ -102,6 +102,6 @@
</ItemGroup> </ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets"> <ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup> </ImportGroup>
</Project> </Project>

View File

@@ -33,7 +33,7 @@
</PropertyGroup> </PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings"> <ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup> </ImportGroup>
<ImportGroup Label="PropertySheets"> <ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" /> <Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@@ -62,7 +62,7 @@
<OutputFile>$(OutDir)/bandwidthTest.exe</OutputFile> <OutputFile>$(OutDir)/bandwidthTest.exe</OutputFile>
</Link> </Link>
<CudaCompile> <CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration> <CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions> <AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include> <Include>./;../../Common</Include>
<Defines>WIN32</Defines> <Defines>WIN32</Defines>
@@ -102,6 +102,6 @@
</ItemGroup> </ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets"> <ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup> </ImportGroup>
</Project> </Project>

View File

@@ -38,7 +38,7 @@
</PropertyGroup> </PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings"> <ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup> </ImportGroup>
<ImportGroup Label="PropertySheets"> <ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" /> <Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@@ -67,7 +67,7 @@
<OutputFile>$(OutDir)/bandwidthTest.exe</OutputFile> <OutputFile>$(OutDir)/bandwidthTest.exe</OutputFile>
</Link> </Link>
<CudaCompile> <CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration> <CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions> <AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include> <Include>./;../../Common</Include>
<Defines>WIN32</Defines> <Defines>WIN32</Defines>
@@ -107,6 +107,6 @@
</ItemGroup> </ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets"> <ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup> </ImportGroup>
</Project> </Project>

View File

@@ -34,7 +34,7 @@
</PropertyGroup> </PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings"> <ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.props" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup> </ImportGroup>
<ImportGroup Label="PropertySheets"> <ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" /> <Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@@ -63,7 +63,7 @@
<OutputFile>$(OutDir)/bandwidthTest.exe</OutputFile> <OutputFile>$(OutDir)/bandwidthTest.exe</OutputFile>
</Link> </Link>
<CudaCompile> <CudaCompile>
<CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;</CodeGeneration> <CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions> <AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include> <Include>./;../../Common</Include>
<Defines>WIN32</Defines> <Defines>WIN32</Defines>
@@ -103,6 +103,6 @@
</ItemGroup> </ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets"> <ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 10.1.targets" /> <Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup> </ImportGroup>
</Project> </Project>

View File

@@ -0,0 +1,362 @@
################################################################################
# Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
# are met:
# * Redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer.
# * Redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution.
# * Neither the name of NVIDIA CORPORATION nor the names of its
# contributors may be used to endorse or promote products derived
# from this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
# EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
# PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
# PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
# OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#
################################################################################
#
# Makefile project only supported on Mac OS X and Linux Platforms)
#
################################################################################
# Location of the CUDA Toolkit
CUDA_PATH ?= /usr/local/cuda
##############################
# start deprecated interface #
##############################
ifeq ($(x86_64),1)
$(info WARNING - x86_64 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=x86_64 instead)
TARGET_ARCH ?= x86_64
endif
ifeq ($(ARMv7),1)
$(info WARNING - ARMv7 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=armv7l instead)
TARGET_ARCH ?= armv7l
endif
ifeq ($(aarch64),1)
$(info WARNING - aarch64 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=aarch64 instead)
TARGET_ARCH ?= aarch64
endif
ifeq ($(ppc64le),1)
$(info WARNING - ppc64le variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=ppc64le instead)
TARGET_ARCH ?= ppc64le
endif
ifneq ($(GCC),)
$(info WARNING - GCC variable has been deprecated)
$(info WARNING - please use HOST_COMPILER=$(GCC) instead)
HOST_COMPILER ?= $(GCC)
endif
ifneq ($(abi),)
$(error ERROR - abi variable has been removed)
endif
############################
# end deprecated interface #
############################
# architecture
HOST_ARCH := $(shell uname -m)
TARGET_ARCH ?= $(HOST_ARCH)
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le armv7l))
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le))
TARGET_SIZE := 64
else ifneq (,$(filter $(TARGET_ARCH),armv7l))
TARGET_SIZE := 32
endif
else
TARGET_SIZE := $(shell getconf LONG_BIT)
endif
else
$(error ERROR - unsupported value $(TARGET_ARCH) for TARGET_ARCH!)
endif
# sbsa and aarch64 systems look similar. Need to differentiate them at host level for now.
ifeq ($(HOST_ARCH),aarch64)
ifeq ($(CUDA_PATH)/targets/sbsa-linux,$(shell ls -1d $(CUDA_PATH)/targets/sbsa-linux))
HOST_ARCH := sbsa
TARGET_ARCH := sbsa
endif
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-sbsa x86_64-ppc64le))
$(error ERROR - cross compiling from $(HOST_ARCH) to $(TARGET_ARCH) is not supported!)
endif
endif
# When on native aarch64 system with userspace of 32-bit, change TARGET_ARCH to armv7l
ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_SIZE),aarch64-aarch64-32)
TARGET_ARCH = armv7l
endif
# operating system
HOST_OS := $(shell uname -s 2>/dev/null | tr "[:upper:]" "[:lower:]")
TARGET_OS ?= $(HOST_OS)
ifeq (,$(filter $(TARGET_OS),linux darwin qnx android))
$(error ERROR - unsupported value $(TARGET_OS) for TARGET_OS!)
endif
# host compiler
ifeq ($(TARGET_OS),darwin)
ifeq ($(shell expr `xcodebuild -version | grep -i xcode | awk '{print $$2}' | cut -d'.' -f1` \>= 5),1)
HOST_COMPILER ?= clang++
endif
else ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq ($(HOST_ARCH)-$(TARGET_ARCH),x86_64-armv7l)
ifeq ($(TARGET_OS),linux)
HOST_COMPILER ?= arm-linux-gnueabihf-g++
else ifeq ($(TARGET_OS),qnx)
ifeq ($(QNX_HOST),)
$(error ERROR - QNX_HOST must be passed to the QNX host toolchain)
endif
ifeq ($(QNX_TARGET),)
$(error ERROR - QNX_TARGET must be passed to the QNX target toolchain)
endif
export QNX_HOST
export QNX_TARGET
HOST_COMPILER ?= $(QNX_HOST)/usr/bin/arm-unknown-nto-qnx6.6.0eabi-g++
else ifeq ($(TARGET_OS),android)
HOST_COMPILER ?= arm-linux-androideabi-g++
endif
else ifeq ($(TARGET_ARCH),aarch64)
ifeq ($(TARGET_OS), linux)
HOST_COMPILER ?= aarch64-linux-gnu-g++
else ifeq ($(TARGET_OS),qnx)
ifeq ($(QNX_HOST),)
$(error ERROR - QNX_HOST must be passed to the QNX host toolchain)
endif
ifeq ($(QNX_TARGET),)
$(error ERROR - QNX_TARGET must be passed to the QNX target toolchain)
endif
export QNX_HOST
export QNX_TARGET
HOST_COMPILER ?= $(QNX_HOST)/usr/bin/aarch64-unknown-nto-qnx7.0.0-g++
else ifeq ($(TARGET_OS), android)
HOST_COMPILER ?= aarch64-linux-android-clang++
endif
else ifeq ($(TARGET_ARCH),sbsa)
HOST_COMPILER ?= aarch64-linux-gnu-g++
else ifeq ($(TARGET_ARCH),ppc64le)
HOST_COMPILER ?= powerpc64le-linux-gnu-g++
endif
endif
HOST_COMPILER ?= g++
NVCC := $(CUDA_PATH)/bin/nvcc -ccbin $(HOST_COMPILER)
# internal flags
NVCCFLAGS := -m${TARGET_SIZE}
CCFLAGS :=
LDFLAGS :=
# build flags
ifeq ($(TARGET_OS),darwin)
LDFLAGS += -rpath $(CUDA_PATH)/lib
CCFLAGS += -arch $(HOST_ARCH)
else ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_OS),x86_64-armv7l-linux)
LDFLAGS += --dynamic-linker=/lib/ld-linux-armhf.so.3
CCFLAGS += -mfloat-abi=hard
else ifeq ($(TARGET_OS),android)
LDFLAGS += -pie
CCFLAGS += -fpie -fpic -fexceptions
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
ifneq ($(TARGET_FS),)
GCCVERSIONLTEQ46 := $(shell expr `$(HOST_COMPILER) -dumpversion` \<= 4.6)
ifeq ($(GCCVERSIONLTEQ46),1)
CCFLAGS += --sysroot=$(TARGET_FS)
endif
LDFLAGS += --sysroot=$(TARGET_FS)
LDFLAGS += -rpath-link=$(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/arm-linux-gnueabihf
endif
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
ifneq ($(TARGET_FS),)
GCCVERSIONLTEQ46 := $(shell expr `$(HOST_COMPILER) -dumpversion` \<= 4.6)
ifeq ($(GCCVERSIONLTEQ46),1)
CCFLAGS += --sysroot=$(TARGET_FS)
endif
LDFLAGS += --sysroot=$(TARGET_FS)
LDFLAGS += -rpath-link=$(TARGET_FS)/lib -L$(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/lib/aarch64-linux-gnu -L$(TARGET_FS)/lib/aarch64-linux-gnu
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L$(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L$(TARGET_FS)/usr/lib/aarch64-linux-gnu
LDFLAGS += --unresolved-symbols=ignore-in-shared-libs
CCFLAGS += -isystem=$(TARGET_FS)/usr/include -I$(TARGET_FS)/usr/include
CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu -I$(TARGET_FS)/usr/include/aarch64-linux-gnu
endif
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
CCFLAGS += -DWIN_INTERFACE_CUSTOM -I/usr/include/aarch64-qnx-gnu
LDFLAGS += -lsocket
LDFLAGS += -rpath=/usr/lib/aarch64-qnx-gnu -L/usr/lib/aarch64-qnx-gnu
ifneq ($(TARGET_FS),)
LDFLAGS += -rpath=$(TARGET_FS)/usr/lib -L $(TARGET_FS)/usr/lib
LDFLAGS += -rpath=$(TARGET_FS)/usr/libnvidia -L $(TARGET_FS)/usr/libnvidia
endif
ifdef TARGET_OVERRIDE # cuda toolkit targets override
NVCCFLAGS += -target-dir $(TARGET_OVERRIDE)
endif
endif
endif
# Install directory of different arch
CUDA_INSTALL_TARGET_DIR :=
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-gnueabihf/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),sbsa-linux)
CUDA_INSTALL_TARGET_DIR = targets/sbsa-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-qnx)
CUDA_INSTALL_TARGET_DIR = targets/ARMv7-linux-QNX/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-qnx/
else ifeq ($(TARGET_ARCH),ppc64le)
CUDA_INSTALL_TARGET_DIR = targets/ppc64le-linux/
endif
# Debug build flags
ifeq ($(dbg),1)
NVCCFLAGS += -g -G
BUILD_TYPE := debug
else
BUILD_TYPE := release
endif
ALL_CCFLAGS :=
ALL_CCFLAGS += $(NVCCFLAGS)
ALL_CCFLAGS += $(EXTRA_NVCCFLAGS)
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(CCFLAGS))
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(EXTRA_CCFLAGS))
SAMPLE_ENABLED := 1
# This sample is not supported on Mac OSX
ifeq ($(TARGET_OS),darwin)
$(info >>> WARNING - bf16TensorCoreGemm is not supported on Mac OSX - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
# This sample is not supported on ARMv7
ifeq ($(TARGET_ARCH),armv7l)
$(info >>> WARNING - bf16TensorCoreGemm is not supported on ARMv7 - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
ALL_LDFLAGS :=
ALL_LDFLAGS += $(ALL_CCFLAGS)
ALL_LDFLAGS += $(addprefix -Xlinker ,$(LDFLAGS))
ALL_LDFLAGS += $(addprefix -Xlinker ,$(EXTRA_LDFLAGS))
# Common includes and paths for CUDA
INCLUDES := -I../../Common
LIBRARIES :=
################################################################################
#Detect if installed version of GCC supports required C++11
ifeq ($(TARGET_OS),linux)
empty :=
space := $(empty) $(empty)
GCCVERSIONSTRING := $(shell expr `$(HOST_COMPILER) -dumpversion`)
#Create version number without "."
GCCVERSION := $(shell expr `echo $(GCCVERSIONSTRING)` | cut -f1 -d.)
GCCVERSION += $(shell expr `echo $(GCCVERSIONSTRING)` | cut -f2 -d.)
GCCVERSION += $(shell expr `echo $(GCCVERSIONSTRING)` | cut -f3 -d.)
# Make sure the version number has at least 3 decimals
GCCVERSION += 00
# Remove spaces from the version number
GCCVERSION := $(subst $(space),$(empty),$(GCCVERSION))
# Crop the version number to 3 decimals.
GCCVERSION := $(shell expr `echo $(GCCVERSION)` | cut -b1-3)
#$(warning $(GCCVERSION))
IS_MIN_VERSION := $(shell expr `echo $(GCCVERSION)` \>= 500)
ifeq ($(IS_MIN_VERSION), 1)
$(info >>> GCC Version is greater or equal to 5.0.0 <<<)
else
$(info >>> Waiving build. Minimum GCC version required is 5.0.0<<<)
SAMPLE_ENABLED := 0
endif
endif
# Gencode arguments
SMS ?= 80
ifeq ($(SMS),)
$(info >>> WARNING - no SM architectures have been specified - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
ifeq ($(GENCODE_FLAGS),)
# Generate SASS code for each SM architecture listed in $(SMS)
$(foreach sm,$(SMS),$(eval GENCODE_FLAGS += -gencode arch=compute_$(sm),code=sm_$(sm)))
# Generate PTX code from the highest SM architecture in $(SMS) to guarantee forward-compatibility
HIGHEST_SM := $(lastword $(sort $(SMS)))
ifneq ($(HIGHEST_SM),)
GENCODE_FLAGS += -gencode arch=compute_$(HIGHEST_SM),code=compute_$(HIGHEST_SM)
endif
endif
ALL_CCFLAGS += --std=c++11
ifeq ($(SAMPLE_ENABLED),0)
EXEC ?= @echo "[@]"
endif
################################################################################
# Target rules
all: build
build: bf16TensorCoreGemm
check.deps:
ifeq ($(SAMPLE_ENABLED),0)
@echo "Sample will be waived due to the above missing dependencies"
else
@echo "Sample is ready - all dependencies have been met"
endif
bf16TensorCoreGemm.o:bf16TensorCoreGemm.cu
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
bf16TensorCoreGemm: bf16TensorCoreGemm.o
$(EXEC) $(NVCC) $(ALL_LDFLAGS) $(GENCODE_FLAGS) -o $@ $+ $(LIBRARIES)
$(EXEC) mkdir -p ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
$(EXEC) cp $@ ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
run: build
$(EXEC) ./bf16TensorCoreGemm
clean:
rm -f bf16TensorCoreGemm bf16TensorCoreGemm.o
rm -rf ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)/bf16TensorCoreGemm
clobber: clean

View File

@@ -0,0 +1,67 @@
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE entry SYSTEM "SamplesInfo.dtd">
<entry>
<name>bf16TensorCoreGemm</name>
<cflags>
<flag>--std=c++11</flag>
</cflags>
<cuda_api_list>
<toolkit>cudaMallocManaged</toolkit>
<toolkit>cudaDeviceSynchronize</toolkit>
<toolkit>cudaFuncSetAttribute</toolkit>
<toolkit>cudaEventCreate</toolkit>
<toolkit>cudaEventRecord</toolkit>
<toolkit>cudaEventSynchronize</toolkit>
<toolkit>cudaEventElapsedTime</toolkit>
<toolkit>cudaFree</toolkit>
</cuda_api_list>
<description><![CDATA[A CUDA sample demonstrating __nv_bfloat16 (e8m7) GEMM computation using the Warp Matrix Multiply and Accumulate (WMMA) API introduced with CUDA 11 in Ampere chip family tensor cores for faster matrix operations. This sample also uses async copy provided by cuda pipeline interface for gmem to shmem async loads which improves kernel performance and reduces register presssure.]]></description>
<devicecompilation>whole</devicecompilation>
<includepaths>
<path>./</path>
<path>../</path>
<path>../../common/inc</path>
</includepaths>
<keyconcepts>
<concept level="basic">Matrix Multiply</concept>
<concept level="advanced">WMMA</concept>
<concept level="advanced">Tensor Cores</concept>
</keyconcepts>
<keywords>
<keyword>matrix multiply</keyword>
<keyword>Async copy</keyword>
<keyword>CPP11</keyword>
<keyword>GCC 5.0.0</keyword>
</keywords>
<libraries>
</libraries>
<librarypaths>
</librarypaths>
<nsight_eclipse>true</nsight_eclipse>
<primary_file>bf16TensorCoreGemm.cu</primary_file>
<scopes>
<scope>1:CUDA Basic Topics</scope>
</scopes>
<sm-arch>sm80</sm-arch>
<supported_envs>
<env>
<arch>x86_64</arch>
<platform>linux</platform>
</env>
<env>
<arch>aarch64</arch>
</env>
<env>
<platform>windows7</platform>
</env>
<env>
<arch>ppc64le</arch>
<platform>linux</platform>
</env>
</supported_envs>
<supported_sm_architectures>
<from>8.0</from>
</supported_sm_architectures>
<title>bfloat16 Tensor Core GEMM</title>
<type>exe</type>
</entry>

View File

@@ -0,0 +1,70 @@
# bf16TensorCoreGemm - bfloat16 Tensor Core GEMM
## Description
A CUDA sample demonstrating __nv_bfloat16 (e8m7) GEMM computation using the Warp Matrix Multiply and Accumulate (WMMA) API introduced with CUDA 11 in Ampere chip family tensor cores for faster matrix operations. This sample also uses async copy provided by cuda pipeline interface for gmem to shmem async loads which improves kernel performance and reduces register presssure.
## Key Concepts
Matrix Multiply, WMMA, Tensor Cores
## Supported SM Architectures
[SM 8.0 ](https://developer.nvidia.com/cuda-gpus)
## Supported OSes
Linux, Windows
## Supported CPU Architecture
x86_64, ppc64le, aarch64
## CUDA APIs involved
### [CUDA Runtime API](http://docs.nvidia.com/cuda/cuda-runtime-api/index.html)
cudaMallocManaged, cudaDeviceSynchronize, cudaFuncSetAttribute, cudaEventCreate, cudaEventRecord, cudaEventSynchronize, cudaEventElapsedTime, cudaFree
## Prerequisites
Download and install the [CUDA Toolkit 11.0](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
## Build and Run
### Windows
The Windows samples are built using the Visual Studio IDE. Solution files (.sln) are provided for each supported version of Visual Studio, using the format:
```
*_vs<version>.sln - for Visual Studio <version>
```
Each individual sample has its own set of solution files in its directory:
To build/examine all the samples at once, the complete solution files should be used. To build/examine a single sample, the individual sample solution files should be used.
> **Note:** Some samples require that the Microsoft DirectX SDK (June 2010 or newer) be installed and that the VC++ directory paths are properly set up (**Tools > Options...**). Check DirectX Dependencies section for details."
### Linux
The Linux samples are built using makefiles. To use the makefiles, change the current directory to the sample directory you wish to build, and run make:
```
$ cd <sample_dir>
$ make
```
The samples makefiles can take advantage of certain options:
* **TARGET_ARCH=<arch>** - cross-compile targeting a specific architecture. Allowed architectures are x86_64, ppc64le, aarch64.
By default, TARGET_ARCH is set to HOST_ARCH. On a x86_64 machine, not setting TARGET_ARCH is the equivalent of setting TARGET_ARCH=x86_64.<br/>
`$ make TARGET_ARCH=x86_64` <br/> `$ make TARGET_ARCH=ppc64le` <br/> `$ make TARGET_ARCH=aarch64` <br/>
See [here](http://docs.nvidia.com/cuda/cuda-samples/index.html#cross-samples) for more details.
* **dbg=1** - build with debug symbols
```
$ make dbg=1
```
* **SMS="A B ..."** - override the SM architectures for which the sample will be built, where `"A B ..."` is a space-delimited list of SM architectures. For example, to generate SASS for SM 50 and SM 60, use `SMS="50 60"`.
```
$ make SMS="50 60"
```
* **HOST_COMPILER=<host_compiler>** - override the default g++ host compiler. See the [Linux Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#system-requirements) for a list of supported host compilers.
```
$ make HOST_COMPILER=g++
```
## References (for more details)

View File

@@ -0,0 +1,838 @@
/* Copyright (c) 2020, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
// CUDA sample demonstrating a __nv_bfloat16 (E8M7) GEMM computation using the Warp Matrix Multiply
// and Accumulate API introduced in CUDA 11.0.
// In this program, the compute_gemm kernel computes the result of a matrix multiplication
// and addition: D = alpha * A * B + beta * C. The dimensions of both C and D matrices
// are M_GLOBAL x N_GLOBAL. The A matrix is M_GLOBAL x K_GLOBAL (row-major), the B matrix
// is K_GLOBAL x N_GLOBAL (column-major).
// In that kernel, each CTA computes one 128 x 128 tile of the resulting matrix
// per iteration. When the tile is computed, the CTA stores it to the global memory
// and begins a new iteration, selecting a new 128 x 128 tile to compute.
// Each CTA consists of eight warps. For the 128 x 128 tile, each warp computes eight
// 16 x 16 subtiles, organized in a 2 x 4 two-dimensional array.
// Warps compute the 16 x 16 subtiles using nvcuda::wmma::mma_sync operations by
// moving through the K_GLOBAL dimension of the A and B matrices and accumulating
// the intermediate result in the local thread state.
// There are a number of simple optimizations used in the algorithm:
// - The CTA copies the 128 x 128 tile of the C matrix from the global memory to
// shared memory. After that is done, each warp loads the C matrix fragments from
// shared memory, thus avoiding a random global memory access.
// - On each internal iteration, the CTA copies a portion of the A and B matrices from
// global memory to shared memory. After that, all warps in the CTA reuse the A and B
// data from shared memory, thus reducing the number of data copies from global memory.
// - The portions of the A and B matrices are stored in shared memory with an additional
// padding (skew) to reduce the number of shared memory access bank conflicts.
// (See a detailed explanation near the SKEW_BF16 macro definition.)
// - When the CTA finishes computing the tiles of the resulting matrix, each warp stores
// its subtiles to shared memory. The CTA then copies the shared memory contents to
// global memory, again avoiding redundant random global memory accesses.
// - Note that the CTA tile size is chosen to maximize the GPU register utilization,
// but carefully enough to avoid local memory use.
#include <assert.h>
#include <stdio.h>
#include <cuda.h>
#include <cuda_bf16.h>
#include <mma.h>
#include <cuda_pipeline.h>
// helper functions and utilities to work with CUDA
#include <helper_functions.h>
#include <helper_cuda.h>
// Externally configurable parameters.
// Switch for choosing cpp interface for cuda pipeline
// vs primitives interface.
#define USE_CPP_API 0
#ifndef CPU_DEBUG
// Set this to 1 to verify the correctness of the GPU-computed matrix.
#define CPU_DEBUG 0
#endif
#ifndef SHARED_MEMORY_LIMIT_64K
// Set this to 0 to use more than 64 Kb of shared memory to cache data, to
// improve the performance of the computations on GPU.
// Note that you need a GPU that can have more than 64 Kb of shared memory
// per multiprocessor.
#define SHARED_MEMORY_LIMIT_64K 0
#endif
// GPU configuration.
#define WARP_SIZE 32
// MMA matrix tile dimensions.
#define M 16
#define N 16
#define K 16
// GEMM configuration.
#define M_TILES 512
#define N_TILES 512
#define K_TILES 512
#define M_GLOBAL (M * M_TILES)
#define N_GLOBAL (N * N_TILES)
#define K_GLOBAL (K * K_TILES)
#define C_LAYOUT wmma::mem_row_major
// Implementation constants.
#define WARPS_PER_BLOCK 8
#define THREADS_PER_BLOCK (WARP_SIZE * WARPS_PER_BLOCK)
#if SHARED_MEMORY_LIMIT_64K
// With only 64 Kb shared memory available, we can fit two 8-tile chunks of
// the A and B matrix data, that is (M = 16) * (K = 16) * 8 * (CHUNK_K = 8)
// * sizeof(__nv_bfloat16) = 32 Kb each.
// (i.e. two 8x8 arrays of tiles of 16x16 __nv_bfloat16-typed elements per CTA).
// But we cannot account the 8 Kb total skew overhead, without which the performance
// would be severely impacted. So we choose to reduce the chunk size in half,
// i.e. the amount of A and B matrix data we cache in shared memory.
// Accordingly, this doubles the number of outer iterations across the global K
// dimension, which only slightly impacts the performance.
#define CHUNK_K 4
#else
#define CHUNK_K 8
#endif
#define CHUNK_LINE_BYTES (CHUNK_K * K * sizeof(__nv_bfloat16))
#define WARP_COPY_BYTES (WARP_SIZE * sizeof(int4))
#define CHUNK_COPY_LINES_PER_WARP (WARP_COPY_BYTES / CHUNK_LINE_BYTES)
#define CHUNK_COPY_LINE_LANES (WARP_SIZE / CHUNK_COPY_LINES_PER_WARP)
#define BLOCK_ROW_WARPS 2
#define BLOCK_COL_WARPS 4
#define WARP_ROW_TILES 4
#define WARP_COL_TILES 2
#define BLOCK_ROW_TILES (WARP_ROW_TILES * BLOCK_ROW_WARPS)
#define BLOCK_COL_TILES (WARP_COL_TILES * BLOCK_COL_WARPS)
#define GLOBAL_MEM_STRIDE N_GLOBAL
#define SHMEM_STRIDE (N * BLOCK_ROW_TILES)
#define SHMEM_OFFSET (N * WARP_ROW_TILES)
// The macro below is used to shift rows of the A matrix and columns of the B matrix
// in shared memory to minimize possible bank conflicts.
// Before performing the nvcuda::wmma::mma_sync operation, the warp must load the matrix
// data using the nvcuda::wmma::load_matrix_sync operation. Although the memory access pattern
// is not specified for that function, each lane in the warp can read one or multiple matrix
// elements from different matrix rows or columns.
// For shared memory, such access can result in bank conflicts if different rows / columns
// of the matrix map to the same bank. By shifting each row and column by a few bytes, we
// make sure that they map to different banks, thus reducing the number of possible bank
// conflicts.
// The number of 16 two-byte "__nv_bfloat16" elements is chosen as the minimum possible shift because
// we must keep each row and column 256-bit aligned, as required by nvcuda::wmma::load_matrix_sync.
#define SKEW_BF16 16
#define checkKernelErrors(expr) do { \
expr; \
\
cudaError_t __err = cudaGetLastError(); \
if (__err != cudaSuccess) { \
printf("Line %d: '%s' failed: %s\n", __LINE__, # expr, cudaGetErrorString(__err)); \
abort(); \
} \
} while(0)
enum kernels
{
bf16mma_shmem_gemm_async_copy = 0, // __nv_bfloat16 MMA shmem using kernel with async_copy
bf16mma_shmem_gemm = 1, // __nv_bfloat16 MMA shmem using kernel normal copy (without async_copy).
simple_bf16mma_gemm = 2 // __nv_bfloat16 MMA non-shmem using simple kernel.
};
const char* kernelNames[] = {"compute_bf16gemm_async_copy", "compute_bf16gemm",
"simple_wmma_bf16gemm"};
using namespace nvcuda;
namespace nvcuda_namespace = nvcuda::experimental;
__host__ void init_host_matrices(__nv_bfloat16 *a, __nv_bfloat16 *b, float *c)
{
for (int i = 0; i < M_GLOBAL; i++) {
for (int j = 0; j < K_GLOBAL; j++) {
a[i*K_GLOBAL+j] = (__nv_bfloat16)(rand() % 3);
}
}
for (int i = 0; i < N_GLOBAL; i++) {
for (int j = 0; j < K_GLOBAL; j++) {
b[i*K_GLOBAL+j] = (__nv_bfloat16)(rand() % 3);
}
}
for (int t = 0; t < M_GLOBAL * N_GLOBAL; t++) {
c[t] = (float)(rand() % 3);
}
}
__global__ void compute_bf16gemm(const __nv_bfloat16 *A, const __nv_bfloat16 *B, const float *C, float *D, float alpha, float beta)
{
#if __CUDA_ARCH__ >= 800
extern __shared__ __nv_bfloat16 shmem[][CHUNK_K * K + SKEW_BF16];
// Warp and lane identification.
const unsigned int warpId = threadIdx.x / WARP_SIZE;
const unsigned int laneId = threadIdx.x % WARP_SIZE;
// Offset in shared memory from which the B matrix is stored.
const size_t shmem_idx_b_off = BLOCK_COL_TILES * M;
// This pointer is used to access the C and D matrix tiles this warp computes.
float *shmem_warp_tile_ptr = (float*)&shmem[0][0] + (warpId / BLOCK_ROW_WARPS) * SHMEM_STRIDE * N * BLOCK_ROW_WARPS + (warpId % BLOCK_ROW_WARPS) * SHMEM_OFFSET;
// This pointer is used to stream the C and D matrices block-wide tile to and from shared memory.
float *shmem_warp_stream_ptr = (float*)&shmem[0][0] + warpId * SHMEM_STRIDE * N;
// Adjust the beta scaler, as it'll be multiplied by alpha at the end of
// each tile computation. Technically this is not generally correct (may result
// in a loss of precision). Zero still needs to be specially handled though.
beta /= alpha;
// Each CTA slides along the 128 x 128 tiles from the top left corner of the matrix to the
// right and down, and selects the next tile to compute. Once there's no such tile,
// all warps in this CTA exit.
for(unsigned int block_pos = blockIdx.x;; block_pos += gridDim.x) {
const unsigned int block_tile_i = ((block_pos * BLOCK_ROW_TILES) / N_TILES) * (BLOCK_COL_TILES);
const unsigned int block_tile_j = (block_pos * BLOCK_COL_TILES) % N_TILES;
// Stop when there are no more D matrix tiles to compute in this CTA.
if (block_tile_i >= M_TILES) {
break;
}
// This warp's pointer to the C matrix data to copy memory from to shared memory.
const size_t gmem_idx = (block_tile_i + warpId) * M * GLOBAL_MEM_STRIDE + block_tile_j * N;
const float *src_gmem_warp_stream_ptr = &C[gmem_idx];
// Stream multiple C tiles to shared memory.
#pragma unroll
for (int i = 0; i < N; i++) {
*((int4*)(shmem_warp_stream_ptr + SHMEM_STRIDE * i) + laneId) =
*((int4*)(src_gmem_warp_stream_ptr + GLOBAL_MEM_STRIDE * i) + laneId);
}
__syncthreads();
// These fragments will accumulate the result of A and B matrix fragment multiplications
// along the K_GLOBAL dimension.
wmma::fragment<wmma::accumulator, M, N, K, float> c[WARP_COL_TILES][WARP_ROW_TILES];
// Load the C matrix tiles into fragments from shared memory.
#pragma unroll
for (int i = 0; i < WARP_COL_TILES; i++) {
#pragma unroll
for (int j = 0; j < WARP_ROW_TILES; j++) {
const float *tile_ptr = shmem_warp_tile_ptr + i * SHMEM_STRIDE * N + j * N;
wmma::load_matrix_sync(c[i][j], tile_ptr, SHMEM_STRIDE, C_LAYOUT);
}
}
__syncthreads();
// Scale the C matrix.
#pragma unroll
for (int i = 0; i < WARP_COL_TILES; i++) {
#pragma unroll
for (int j = 0; j < WARP_ROW_TILES; j++) {
#pragma unroll
for (int t = 0; t < c[i][j].num_elements; t++) {
c[i][j].x[t] *= beta;
}
}
}
// Select what warp copies what matrix to shared memory.
// Warps 0-3 copy the A matrix, warps 4-7 copy the B matrix.
const __nv_bfloat16 *warp_ptr = (warpId < (WARPS_PER_BLOCK/2)) ? (&A[block_tile_i * M * K_GLOBAL] + M * K_GLOBAL * (warpId % (WARPS_PER_BLOCK/2)) * 2) :
(&B[block_tile_j * N * K_GLOBAL] + N * K_GLOBAL * (warpId % (WARPS_PER_BLOCK/2)) * 2);
// Go through the global K dimension by a fixed step at a time.
#pragma unroll
for (int tile_k = 0; tile_k < K_TILES; tile_k += CHUNK_K) {
// Copy slices of the A and B matrices to shared memory.
// The first half of the warps in the CTA copy the A matrix, the rest copy the B matrix.
size_t shmem_idx = warpId < (WARPS_PER_BLOCK/2) ? (M * (warpId % (WARPS_PER_BLOCK/2)) * 2) :
(N * (warpId % (WARPS_PER_BLOCK/2)) * 2 + shmem_idx_b_off);
// First half of the warp copies the first row / column of the matrix,
// the second half of the warp copies the next.
const __nv_bfloat16 *lane_ptr = (warp_ptr + tile_k * K + (laneId / CHUNK_COPY_LINE_LANES) * K_GLOBAL);
// Shift the second half of the warp to the next row / column in the shared memory.
shmem_idx += laneId / CHUNK_COPY_LINE_LANES;
#pragma unroll
for(int i = 0; i < ((WARP_SIZE/2) / CHUNK_COPY_LINES_PER_WARP) * 2; i++) {
// Copy 16 bytes at once in each lane.
*((int4*)&shmem[shmem_idx][0] + (laneId % CHUNK_COPY_LINE_LANES)) = *((int4*)lane_ptr + (laneId % CHUNK_COPY_LINE_LANES));
// Advance the global memory pointer and the shared memory index.
lane_ptr = lane_ptr + K_GLOBAL * CHUNK_COPY_LINES_PER_WARP;
shmem_idx += CHUNK_COPY_LINES_PER_WARP;
}
__syncthreads();
// Compute a grid of C matrix tiles in each warp.
#pragma unroll
for (int k_step = 0; k_step < CHUNK_K; k_step++) {
wmma::fragment<wmma::matrix_a, M, N, K, __nv_bfloat16, wmma::row_major> a[WARP_COL_TILES];
wmma::fragment<wmma::matrix_b, M, N, K, __nv_bfloat16, wmma::col_major> b[WARP_ROW_TILES];
#pragma unroll
for (int i = 0; i < WARP_COL_TILES; i++) {
size_t shmem_idx_a = (warpId/BLOCK_ROW_WARPS) * M * BLOCK_ROW_WARPS + (i * M);
const __nv_bfloat16 *tile_ptr = &shmem[shmem_idx_a][k_step * K];
wmma::load_matrix_sync(a[i], tile_ptr, K * CHUNK_K + SKEW_BF16);
#pragma unroll
for (int j = 0; j < WARP_ROW_TILES; j++) {
if (i == 0) {
// Load the B matrix fragment once, because it is going to be reused
// against the other A matrix fragments.
size_t shmem_idx_b = shmem_idx_b_off + (WARP_ROW_TILES * N) * (warpId%2) + (j * N);
const __nv_bfloat16 *tile_ptr = &shmem[shmem_idx_b][k_step * K];
wmma::load_matrix_sync(b[j], tile_ptr, K * CHUNK_K + SKEW_BF16);
}
wmma::mma_sync(c[i][j], a[i], b[j], c[i][j]);
}
}
}
__syncthreads();
}
// Store the D fragments to shared memory.
#pragma unroll
for (int i = 0; i < WARP_COL_TILES; i++) {
#pragma unroll
for (int j = 0; j < WARP_ROW_TILES; j++) {
#pragma unroll
// Uniform, point-wise transformations of ALL fragment elements by ALL threads in the
// warp are well-defined even though element indices within fragment storage are not defined.
for (int t = 0; t < c[i][j].num_elements; t++)
c[i][j].x[t] *= alpha;
float *tile_ptr = shmem_warp_tile_ptr + i * SHMEM_STRIDE * K + j * N;
wmma::store_matrix_sync(tile_ptr, c[i][j], SHMEM_STRIDE, C_LAYOUT);
}
}
__syncthreads();
// Now that shared memory contains all the D tiles, stream them to global memory.
float *dst_gmem_warp_stream_ptr = &D[gmem_idx];
#pragma unroll
for (int i = 0; i < N; i++) {
*((int4*)(dst_gmem_warp_stream_ptr + GLOBAL_MEM_STRIDE * i) + laneId) =
*((int4*)(shmem_warp_stream_ptr + SHMEM_STRIDE * i) + laneId);
}
__syncthreads();
}
#endif
}
__global__ void compute_bf16gemm_async_copy(const __nv_bfloat16 *A, const __nv_bfloat16 *B, const float *C, float *D, float alpha, float beta)
{
#if __CUDA_ARCH__ >= 800
extern __shared__ __nv_bfloat16 shmem[][CHUNK_K * K + SKEW_BF16];
// Warp and lane identification.
const unsigned int warpId = threadIdx.x / WARP_SIZE;
const unsigned int laneId = threadIdx.x % WARP_SIZE;
// Offset in shared memory from which the B matrix is stored.
const size_t shmem_idx_b_off = BLOCK_COL_TILES * M;
// This pointer is used to access the C and D matrix tiles this warp computes.
float *shmem_warp_tile_ptr = (float*)&shmem[0][0] + (warpId / BLOCK_ROW_WARPS) * SHMEM_STRIDE * N * BLOCK_ROW_WARPS + (warpId % BLOCK_ROW_WARPS) * SHMEM_OFFSET;
// This pointer is used to stream the C and D matrices block-wide tile to and from shared memory.
float *shmem_warp_stream_ptr = (float*)&shmem[0][0] + warpId * SHMEM_STRIDE * N;
// Adjust the beta scaler, as it'll be multiplied by alpha at the end of
// each tile computation. Technically this is not generally correct (may result
// in a loss of precision). Zero still needs to be specially handled though.
beta /= alpha;
#if USE_CPP_API
nvcuda_namespace::pipeline pipe;
#endif
// Each CTA slides along the 128 x 128 tiles from the top left corner of the matrix to the
// right and down, and selects the next tile to compute. Once there's no such tile,
// all warps in this CTA exit.
for(unsigned int block_pos = blockIdx.x;; block_pos += gridDim.x) {
const unsigned int block_tile_i = ((block_pos * BLOCK_ROW_TILES) / N_TILES) * (BLOCK_COL_TILES);
const unsigned int block_tile_j = (block_pos * BLOCK_COL_TILES) % N_TILES;
// Stop when there are no more D matrix tiles to compute in this CTA.
if (block_tile_i >= M_TILES) {
break;
}
// This warp's pointer to the C matrix data to copy memory from to shared memory.
const size_t gmem_idx = (block_tile_i + warpId) * M * GLOBAL_MEM_STRIDE + block_tile_j * N;
const float *src_gmem_warp_stream_ptr = &C[gmem_idx];
// Stream multiple C tiles to shared memory.
#pragma unroll
for (int i = 0; i < N; i++) {
#if USE_CPP_API
nvcuda_namespace::memcpy_async(*((int4*)(shmem_warp_stream_ptr + SHMEM_STRIDE * i) + laneId),
*((int4*)(src_gmem_warp_stream_ptr + GLOBAL_MEM_STRIDE * i) + laneId),
pipe);
pipe.commit();
#else
__pipeline_memcpy_async((reinterpret_cast<int4*>(&shmem_warp_stream_ptr[(SHMEM_STRIDE * i)])) + laneId,
(reinterpret_cast<const int4*>(&src_gmem_warp_stream_ptr[(GLOBAL_MEM_STRIDE * i)])) + laneId,
sizeof(int4));
__pipeline_commit();
#endif
}
#if USE_CPP_API
pipe.wait_prior<0>();
#else
__pipeline_wait_prior(0);
#endif
__syncthreads();
// These fragments will accumulate the result of A and B matrix fragment multiplications
// along the K_GLOBAL dimension.
wmma::fragment<wmma::accumulator, M, N, K, float> c[WARP_COL_TILES][WARP_ROW_TILES];
// Load the C matrix tiles into fragments from shared memory.
#pragma unroll
for (int i = 0; i < WARP_COL_TILES; i++) {
#pragma unroll
for (int j = 0; j < WARP_ROW_TILES; j++) {
const float *tile_ptr = shmem_warp_tile_ptr + i * SHMEM_STRIDE * N + j * N;
wmma::load_matrix_sync(c[i][j], tile_ptr, SHMEM_STRIDE, C_LAYOUT);
}
}
__syncthreads();
// Scale the C matrix.
#pragma unroll
for (int i = 0; i < WARP_COL_TILES; i++) {
#pragma unroll
for (int j = 0; j < WARP_ROW_TILES; j++) {
#pragma unroll
for (int t = 0; t < c[i][j].num_elements; t++) {
c[i][j].x[t] *= beta;
}
}
}
// Select what warp copies what matrix to shared memory.
// Warps 0-3 copy the A matrix, warps 4-7 copy the B matrix.
const __nv_bfloat16 *warp_ptr = (warpId < (WARPS_PER_BLOCK/2)) ? (&A[block_tile_i * M * K_GLOBAL] + M * K_GLOBAL * (warpId % (WARPS_PER_BLOCK/2)) * 2) :
(&B[block_tile_j * N * K_GLOBAL] + N * K_GLOBAL * (warpId % (WARPS_PER_BLOCK/2)) * 2);
// Go through the global K dimension by a fixed step at a time.
#pragma unroll
for (int tile_k = 0; tile_k < K_TILES; tile_k += CHUNK_K) {
// Copy slices of the A and B matrices to shared memory.
// The first half of the warps in the CTA copy the A matrix, the rest copy the B matrix.
size_t shmem_idx = warpId < (WARPS_PER_BLOCK/2) ? (M * (warpId % (WARPS_PER_BLOCK/2)) * 2) :
(N * (warpId % (WARPS_PER_BLOCK/2)) * 2 + shmem_idx_b_off);
// First half of the warp copies the first row / column of the matrix,
// the second half of the warp copies the next.
const __nv_bfloat16 *lane_ptr = (warp_ptr + tile_k * K + (laneId / CHUNK_COPY_LINE_LANES) * K_GLOBAL);
// Shift the second half of the warp to the next row / column in the shared memory.
shmem_idx += laneId / CHUNK_COPY_LINE_LANES;
#pragma unroll
for(int i = 0; i < ((WARP_SIZE/2) / CHUNK_COPY_LINES_PER_WARP) * 2; i++) {
// Copy 16 bytes at once in each lane.
#if USE_CPP_API
nvcuda_namespace::memcpy_async(*((int4*)&shmem[shmem_idx][0] + (laneId % CHUNK_COPY_LINE_LANES)),
*((int4*)lane_ptr + (laneId % CHUNK_COPY_LINE_LANES)), pipe);
pipe.commit();
#else
__pipeline_memcpy_async((int4*)&shmem[shmem_idx][0] + (laneId % CHUNK_COPY_LINE_LANES),
(int4*)lane_ptr + (laneId % CHUNK_COPY_LINE_LANES), sizeof(int4));
__pipeline_commit();
#endif
// Advance the global memory pointer and the shared memory index.
lane_ptr = lane_ptr + K_GLOBAL * CHUNK_COPY_LINES_PER_WARP;
shmem_idx += CHUNK_COPY_LINES_PER_WARP;
}
#if USE_CPP_API
pipe.wait_prior<0>();
#else
__pipeline_wait_prior(0);
#endif
__syncthreads();
// Compute a grid of C matrix tiles in each warp.
#pragma unroll
for (int k_step = 0; k_step < CHUNK_K; k_step++) {
wmma::fragment<wmma::matrix_a, M, N, K, __nv_bfloat16, wmma::row_major> a[WARP_COL_TILES];
wmma::fragment<wmma::matrix_b, M, N, K, __nv_bfloat16, wmma::col_major> b[WARP_ROW_TILES];
#pragma unroll
for (int i = 0; i < WARP_COL_TILES; i++) {
size_t shmem_idx_a = (warpId / BLOCK_ROW_WARPS) * M * BLOCK_ROW_WARPS + (i * M);
const __nv_bfloat16 *tile_ptr = &shmem[shmem_idx_a][k_step * K];
wmma::load_matrix_sync(a[i], tile_ptr, K * CHUNK_K + SKEW_BF16);
#pragma unroll
for (int j = 0; j < WARP_ROW_TILES; j++) {
if (i == 0) {
// Load the B matrix fragment once, because it is going to be reused
// against the other A matrix fragments.
size_t shmem_idx_b = shmem_idx_b_off + (WARP_ROW_TILES * N) * (warpId%2) + (j * N);
const __nv_bfloat16 *tile_ptr = &shmem[shmem_idx_b][k_step * K];
wmma::load_matrix_sync(b[j], tile_ptr, K * CHUNK_K + SKEW_BF16);
}
wmma::mma_sync(c[i][j], a[i], b[j], c[i][j]);
}
}
}
__syncthreads();
}
// Store the D fragments to shared memory.
#pragma unroll
for (int i = 0; i < WARP_COL_TILES; i++) {
#pragma unroll
for (int j = 0; j < WARP_ROW_TILES; j++) {
#pragma unroll
// Uniform, point-wise transformations of ALL fragment elements by ALL threads in the
// warp are well-defined even though element indices within fragment storage are not defined.
for (int t = 0; t < c[i][j].num_elements; t++)
c[i][j].x[t] *= alpha;
float *tile_ptr = shmem_warp_tile_ptr + i * SHMEM_STRIDE * K + j * N;
wmma::store_matrix_sync(tile_ptr, c[i][j], SHMEM_STRIDE, C_LAYOUT);
}
}
__syncthreads();
// Now that shared memory contains all the D tiles, stream them to global memory.
float *dst_gmem_warp_stream_ptr = &D[gmem_idx];
#pragma unroll
for (int i = 0; i < N; i++) {
*((int4*)(dst_gmem_warp_stream_ptr + GLOBAL_MEM_STRIDE * i) + laneId) =
*((int4*)(shmem_warp_stream_ptr + SHMEM_STRIDE * i) + laneId);
}
__syncthreads();
}
#endif
}
// Performs an MxNxK bf16 GEMM (C=alpha*A*B + beta*C) assuming:
// 1) Matrices are packed in memory.
// 2) M, N and K are multiples of 16, 16 and 16 respectively.
// 3) A is row major, B is column major matrix.
// Note: This is a less performant version of the compute_bf16gemm kernel. It is designed for
// demonstration purposes only to show the CUDA WMMA API use without relying on
// availability of the shared memory.
__global__ void simple_wmma_bf16gemm(__nv_bfloat16 *a, __nv_bfloat16 *b, float *c, float *d, int m_ld, int n_ld, int k_ld, float alpha, float beta)
{
#if __CUDA_ARCH__ >= 800
// Leading dimensions. Packed with no transpositions.
int lda = k_ld;
int ldb = k_ld;
int ldc = n_ld;
// Tile using a 2D grid
int warpM = (blockIdx.x * blockDim.x + threadIdx.x) / warpSize;
int warpN = (blockIdx.y * blockDim.y + threadIdx.y);
// Declare the fragments
wmma::fragment<wmma::matrix_a, M, N, K, __nv_bfloat16, wmma::row_major> a_frag;
wmma::fragment<wmma::matrix_b, M, N, K, __nv_bfloat16, wmma::col_major> b_frag;
wmma::fragment<wmma::accumulator, M, N, K, float> acc_frag;
wmma::fragment<wmma::accumulator, M, N, K, float> c_frag;
wmma::fill_fragment(acc_frag, 0.0f);
// Loop over k
for (int i = 0; i < k_ld; i += K) {
int aCol = i;
int aRow = warpM * M;
int bCol = i;
int bRow = warpN * N;
// Bounds checking
if (aRow < m_ld && aCol < k_ld && bRow < k_ld && bCol < n_ld) {
// Load the inputs
wmma::load_matrix_sync(a_frag, a + aCol + aRow * lda, lda);
wmma::load_matrix_sync(b_frag, b + bRow + bCol * ldb, ldb);
// Perform the matrix multiplication
wmma::mma_sync(acc_frag, a_frag, b_frag, acc_frag);
}
}
// Load in the current value of c, scale it by beta, and add this our result scaled by alpha
int cCol = warpN * N;
int cRow = warpM * M;
if (cRow < m_ld && cCol < n_ld) {
wmma::load_matrix_sync(c_frag, c + cCol + cRow * ldc, ldc, wmma::mem_row_major);
for(int i=0; i < c_frag.num_elements; i++) {
c_frag.x[i] = alpha * acc_frag.x[i] + beta * c_frag.x[i];
}
// Store the output
wmma::store_matrix_sync(d + cCol + cRow * ldc, c_frag, ldc, wmma::mem_row_major);
}
#endif
}
__host__ void matMultiplyOnHost(__nv_bfloat16 *A, __nv_bfloat16 *B, float *C,
float alpha, float beta,
int numARows, int numAColumns,
int numBRows, int numBColumns,
int numCRows, int numCColumns)
{
for (int i = 0; i < numCRows; i++) {
for (int j = 0; j < numCColumns; j++) {
float temp = 0.0;
for (int k = 0; k < numAColumns; k++) {
temp += (float)A[i * numAColumns + k] * (float)B[j * numBRows + k];
}
C[i*numCColumns + j] = temp * alpha + beta * C[i * numCColumns + j];
}
}
}
int main(int argc, char **argv)
{
printf("Initializing...\n");
int dev = findCudaDevice(argc, (const char **)argv);
cudaDeviceProp deviceProp;
checkCudaErrors(cudaGetDeviceProperties(&deviceProp, dev));
// Tensor cores require a GPU of Volta (SM8X) architecture or higher.
if (deviceProp.major < 8) {
printf("bf16TensorCoreGemm requires requires SM 8.0 or higher to use Tensor Cores. Exiting...\n");
exit(EXIT_WAIVED);
}
printf("M: %d (%d x %d)\n", M_GLOBAL, M, M_TILES);
printf("N: %d (%d x %d)\n", N_GLOBAL, N, N_TILES);
printf("K: %d (%d x %d)\n", K_GLOBAL, K, K_TILES);
__nv_bfloat16 *A_h = NULL;
__nv_bfloat16 *B_h = NULL;
float *C_h = NULL;
#if CPU_DEBUG
float *result_hD = NULL;
float *result_host = NULL;
#endif
A_h = (__nv_bfloat16*) malloc(sizeof(__nv_bfloat16) * M_GLOBAL * K_GLOBAL);
B_h = (__nv_bfloat16*) malloc(sizeof(__nv_bfloat16) * K_GLOBAL * N_GLOBAL);
C_h = (float*) malloc(sizeof(float) * M_GLOBAL * N_GLOBAL);
#if CPU_DEBUG
result_hD = (float*) malloc(sizeof(float) * M_GLOBAL * N_GLOBAL);
result_host = (float*) malloc(sizeof(float) * M_GLOBAL * N_GLOBAL);
#endif
__nv_bfloat16 *A = NULL;
__nv_bfloat16 *B = NULL;
float *C = NULL;
float *D = NULL;
checkCudaErrors(cudaMalloc((void**)&A, sizeof(__nv_bfloat16) * M_GLOBAL * K_GLOBAL));
checkCudaErrors(cudaMalloc((void**)&B, sizeof(__nv_bfloat16) * N_GLOBAL * K_GLOBAL));
checkCudaErrors(cudaMalloc((void**)&C, sizeof(float) * M_GLOBAL * N_GLOBAL));
checkCudaErrors(cudaMalloc((void**)&D, sizeof(float) * M_GLOBAL * N_GLOBAL));
assert(((unsigned long long)A) % 128 == 0);
assert(((unsigned long long)B) % 128 == 0);
assert(((unsigned long long)C) % 128 == 0);
assert(((unsigned long long)D) % 128 == 0);
init_host_matrices(A_h, B_h, C_h);
printf("Preparing data for GPU...\n");
checkCudaErrors(cudaMemcpy(A, A_h, sizeof(__nv_bfloat16) * M_GLOBAL * K_GLOBAL, cudaMemcpyHostToDevice));
checkCudaErrors(cudaMemcpy(B, B_h, sizeof(__nv_bfloat16) * N_GLOBAL * K_GLOBAL, cudaMemcpyHostToDevice));
checkCudaErrors(cudaMemcpy(C, C_h, sizeof(float) * M_GLOBAL * N_GLOBAL, cudaMemcpyHostToDevice));
checkCudaErrors(cudaMemset(D, 0, sizeof(float) * M_GLOBAL * N_GLOBAL));
enum {
// Compute the right amount of shared memory to request.
// We need shared memory to hold per-CTA C and D matrix tiles, and to cache per-CTA chunks
// of the A and B matrices. Therefore, the right amount to request is the maximum of those
// two numbers.
SHMEM_SZ = MAX(sizeof(__nv_bfloat16) * (BLOCK_COL_TILES * M) * (CHUNK_K * K + SKEW_BF16) * 2,
M * (BLOCK_ROW_WARPS * WARP_ROW_TILES) * N * (BLOCK_COL_WARPS * WARP_COL_TILES) * sizeof(float))
};
printf("Required shared memory size: %lu Kb\n", SHMEM_SZ / 1024UL);
const float alpha = 1.1f;
const float beta = 1.2f;
cudaEvent_t start, stop;
checkCudaErrors(cudaEventCreate(&start));
checkCudaErrors(cudaEventCreate(&stop));
checkCudaErrors(cudaEventRecord(start));
// kernel to run - default (b16mma_shmem_gemm_async_copy == 0)
kernels selected_kernel = bf16mma_shmem_gemm_async_copy;
if (checkCmdLineFlag(argc, (const char **)argv, "kernel")) {
int kernel_number = getCmdLineArgumentInt(argc, (const char **)argv, "kernel");
if (kernel_number < 3) {
selected_kernel = (kernels)kernel_number;
}
else {
printf("Error: kernel number should be between 0 to 2, you have entered %d\n", kernel_number);
exit(EXIT_FAILURE);
}
}
// If enough shared memory available on the GPU use high performant kernel
if ((deviceProp.sharedMemPerMultiprocessor >= SHMEM_SZ) && (selected_kernel != simple_bf16mma_gemm)) {
printf("Computing using high performance kernel = %d - %s\n", selected_kernel, kernelNames[selected_kernel]);
switch (selected_kernel)
{
case bf16mma_shmem_gemm_async_copy :
default:
checkCudaErrors(cudaFuncSetAttribute(compute_bf16gemm_async_copy, cudaFuncAttributeMaxDynamicSharedMemorySize, SHMEM_SZ));
checkKernelErrors((compute_bf16gemm_async_copy<<<deviceProp.multiProcessorCount*2, THREADS_PER_BLOCK, SHMEM_SZ>>>(A, B, C, D, alpha, beta)));
break;
case bf16mma_shmem_gemm :
checkCudaErrors(cudaFuncSetAttribute(compute_bf16gemm, cudaFuncAttributeMaxDynamicSharedMemorySize, SHMEM_SZ));
checkKernelErrors((compute_bf16gemm<<<deviceProp.multiProcessorCount*2, THREADS_PER_BLOCK, SHMEM_SZ>>>(A, B, C, D, alpha, beta)));
break;
}
#if CPU_DEBUG
checkCudaErrors(cudaMemcpy(result_hD, D, sizeof(float)*M_GLOBAL*N_GLOBAL, cudaMemcpyDeviceToHost));
#endif
}
else {
dim3 gridDim;
dim3 blockDim;
// blockDim.x must be a multple of warpSize
// 128x4 means we have 16 warps and a block computes a 64x64 output tile
blockDim.x = 128;
blockDim.y = 4;
gridDim.x = (M_GLOBAL + (M * blockDim.x / 32 - 1)) / (M * blockDim.x / 32);
gridDim.y = (N_GLOBAL + N * blockDim.y - 1) / (N * blockDim.y);
printf("Computing... using simple_wmma_gemm kernel\n");
simple_wmma_bf16gemm<<<gridDim, blockDim>>>(A, B, C, D, M_GLOBAL, N_GLOBAL, K_GLOBAL, alpha, beta);
#if CPU_DEBUG
checkCudaErrors(cudaMemcpy(result_hD, D, sizeof(float) * M_GLOBAL * N_GLOBAL, cudaMemcpyDeviceToHost));
#endif
}
checkCudaErrors(cudaEventRecord(stop));
checkCudaErrors(cudaEventSynchronize(stop));
#if CPU_DEBUG
printf("Verifying correctness of the computations...\n");
memcpy(result_host, C_h, sizeof(float) * M_GLOBAL * N_GLOBAL);
matMultiplyOnHost(A_h, B_h, result_host,
alpha, beta,
M_GLOBAL, K_GLOBAL,
K_GLOBAL, N_GLOBAL,
M_GLOBAL, N_GLOBAL);
for (int i = 0; i < N_GLOBAL * M_GLOBAL; i++) {
if (fabs(result_hD[i] - result_host[i]) > 0.1f) {
printf("mismatch i=%d result_hD=%f result_host=%f\n", i, result_hD[i], result_host[i]);
}
}
free(result_hD);
free(result_host);
#endif
float milliseconds = 0;
checkCudaErrors(cudaEventElapsedTime(&milliseconds, start, stop));
printf("Time: %f ms\n", milliseconds);
printf("TFLOPS: %.2f\n", (((double)M_GLOBAL * N_GLOBAL * K_GLOBAL * 2)/(milliseconds/1000.)) / 1e12);
free(A_h);
free(B_h);
free(C_h);
checkCudaErrors(cudaFree((void*)A));
checkCudaErrors(cudaFree((void*)B));
checkCudaErrors(cudaFree((void*)C));
checkCudaErrors(cudaFree((void*)D));
return 0;
}

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 14.00
# Visual Studio 2015
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "bf16TensorCoreGemm", "bf16TensorCoreGemm_vs2015.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,107 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>bf16TensorCoreGemm_vs2015</RootNamespace>
<ProjectName>bf16TensorCoreGemm</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v140</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/bf16TensorCoreGemm.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_80,sm_80;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="bf16TensorCoreGemm.cu" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 12.00
# Visual Studio 2017
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "bf16TensorCoreGemm", "bf16TensorCoreGemm_vs2017.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,112 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>bf16TensorCoreGemm_vs2017</RootNamespace>
<ProjectName>bf16TensorCoreGemm</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<PropertyGroup Condition="'$(WindowsTargetPlatformVersion)'==''">
<LatestTargetPlatformVersion>$([Microsoft.Build.Utilities.ToolLocationHelper]::GetLatestSDKTargetPlatformVersion('Windows', '10.0'))</LatestTargetPlatformVersion>
<WindowsTargetPlatformVersion Condition="'$(WindowsTargetPlatformVersion)' == ''">$(LatestTargetPlatformVersion)</WindowsTargetPlatformVersion>
<TargetPlatformVersion>$(WindowsTargetPlatformVersion)</TargetPlatformVersion>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v141</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/bf16TensorCoreGemm.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_80,sm_80;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="bf16TensorCoreGemm.cu" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 12.00
# Visual Studio 2019
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "bf16TensorCoreGemm", "bf16TensorCoreGemm_vs2019.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,108 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>bf16TensorCoreGemm_vs2019</RootNamespace>
<ProjectName>bf16TensorCoreGemm</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v142</PlatformToolset>
<WindowsTargetPlatformVersion>10.0</WindowsTargetPlatformVersion>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/bf16TensorCoreGemm.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_80,sm_80;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="bf16TensorCoreGemm.cu" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,360 @@
################################################################################
# Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
# are met:
# * Redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer.
# * Redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution.
# * Neither the name of NVIDIA CORPORATION nor the names of its
# contributors may be used to endorse or promote products derived
# from this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
# EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
# PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
# PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
# OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#
################################################################################
#
# Makefile project only supported on Mac OS X and Linux Platforms)
#
################################################################################
# Location of the CUDA Toolkit
CUDA_PATH ?= /usr/local/cuda
##############################
# start deprecated interface #
##############################
ifeq ($(x86_64),1)
$(info WARNING - x86_64 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=x86_64 instead)
TARGET_ARCH ?= x86_64
endif
ifeq ($(ARMv7),1)
$(info WARNING - ARMv7 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=armv7l instead)
TARGET_ARCH ?= armv7l
endif
ifeq ($(aarch64),1)
$(info WARNING - aarch64 variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=aarch64 instead)
TARGET_ARCH ?= aarch64
endif
ifeq ($(ppc64le),1)
$(info WARNING - ppc64le variable has been deprecated)
$(info WARNING - please use TARGET_ARCH=ppc64le instead)
TARGET_ARCH ?= ppc64le
endif
ifneq ($(GCC),)
$(info WARNING - GCC variable has been deprecated)
$(info WARNING - please use HOST_COMPILER=$(GCC) instead)
HOST_COMPILER ?= $(GCC)
endif
ifneq ($(abi),)
$(error ERROR - abi variable has been removed)
endif
############################
# end deprecated interface #
############################
# architecture
HOST_ARCH := $(shell uname -m)
TARGET_ARCH ?= $(HOST_ARCH)
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le armv7l))
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le))
TARGET_SIZE := 64
else ifneq (,$(filter $(TARGET_ARCH),armv7l))
TARGET_SIZE := 32
endif
else
TARGET_SIZE := $(shell getconf LONG_BIT)
endif
else
$(error ERROR - unsupported value $(TARGET_ARCH) for TARGET_ARCH!)
endif
# sbsa and aarch64 systems look similar. Need to differentiate them at host level for now.
ifeq ($(HOST_ARCH),aarch64)
ifeq ($(CUDA_PATH)/targets/sbsa-linux,$(shell ls -1d $(CUDA_PATH)/targets/sbsa-linux))
HOST_ARCH := sbsa
TARGET_ARCH := sbsa
endif
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-sbsa x86_64-ppc64le))
$(error ERROR - cross compiling from $(HOST_ARCH) to $(TARGET_ARCH) is not supported!)
endif
endif
# When on native aarch64 system with userspace of 32-bit, change TARGET_ARCH to armv7l
ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_SIZE),aarch64-aarch64-32)
TARGET_ARCH = armv7l
endif
# operating system
HOST_OS := $(shell uname -s 2>/dev/null | tr "[:upper:]" "[:lower:]")
TARGET_OS ?= $(HOST_OS)
ifeq (,$(filter $(TARGET_OS),linux darwin qnx android))
$(error ERROR - unsupported value $(TARGET_OS) for TARGET_OS!)
endif
# host compiler
ifeq ($(TARGET_OS),darwin)
ifeq ($(shell expr `xcodebuild -version | grep -i xcode | awk '{print $$2}' | cut -d'.' -f1` \>= 5),1)
HOST_COMPILER ?= clang++
endif
else ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq ($(HOST_ARCH)-$(TARGET_ARCH),x86_64-armv7l)
ifeq ($(TARGET_OS),linux)
HOST_COMPILER ?= arm-linux-gnueabihf-g++
else ifeq ($(TARGET_OS),qnx)
ifeq ($(QNX_HOST),)
$(error ERROR - QNX_HOST must be passed to the QNX host toolchain)
endif
ifeq ($(QNX_TARGET),)
$(error ERROR - QNX_TARGET must be passed to the QNX target toolchain)
endif
export QNX_HOST
export QNX_TARGET
HOST_COMPILER ?= $(QNX_HOST)/usr/bin/arm-unknown-nto-qnx6.6.0eabi-g++
else ifeq ($(TARGET_OS),android)
HOST_COMPILER ?= arm-linux-androideabi-g++
endif
else ifeq ($(TARGET_ARCH),aarch64)
ifeq ($(TARGET_OS), linux)
HOST_COMPILER ?= aarch64-linux-gnu-g++
else ifeq ($(TARGET_OS),qnx)
ifeq ($(QNX_HOST),)
$(error ERROR - QNX_HOST must be passed to the QNX host toolchain)
endif
ifeq ($(QNX_TARGET),)
$(error ERROR - QNX_TARGET must be passed to the QNX target toolchain)
endif
export QNX_HOST
export QNX_TARGET
HOST_COMPILER ?= $(QNX_HOST)/usr/bin/aarch64-unknown-nto-qnx7.0.0-g++
else ifeq ($(TARGET_OS), android)
HOST_COMPILER ?= aarch64-linux-android-clang++
endif
else ifeq ($(TARGET_ARCH),sbsa)
HOST_COMPILER ?= aarch64-linux-gnu-g++
else ifeq ($(TARGET_ARCH),ppc64le)
HOST_COMPILER ?= powerpc64le-linux-gnu-g++
endif
endif
HOST_COMPILER ?= g++
NVCC := $(CUDA_PATH)/bin/nvcc -ccbin $(HOST_COMPILER)
# internal flags
NVCCFLAGS := -m${TARGET_SIZE}
CCFLAGS :=
LDFLAGS :=
# build flags
ifeq ($(TARGET_OS),darwin)
LDFLAGS += -rpath $(CUDA_PATH)/lib
CCFLAGS += -arch $(HOST_ARCH)
else ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_OS),x86_64-armv7l-linux)
LDFLAGS += --dynamic-linker=/lib/ld-linux-armhf.so.3
CCFLAGS += -mfloat-abi=hard
else ifeq ($(TARGET_OS),android)
LDFLAGS += -pie
CCFLAGS += -fpie -fpic -fexceptions
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
ifneq ($(TARGET_FS),)
GCCVERSIONLTEQ46 := $(shell expr `$(HOST_COMPILER) -dumpversion` \<= 4.6)
ifeq ($(GCCVERSIONLTEQ46),1)
CCFLAGS += --sysroot=$(TARGET_FS)
endif
LDFLAGS += --sysroot=$(TARGET_FS)
LDFLAGS += -rpath-link=$(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/arm-linux-gnueabihf
endif
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
ifneq ($(TARGET_FS),)
GCCVERSIONLTEQ46 := $(shell expr `$(HOST_COMPILER) -dumpversion` \<= 4.6)
ifeq ($(GCCVERSIONLTEQ46),1)
CCFLAGS += --sysroot=$(TARGET_FS)
endif
LDFLAGS += --sysroot=$(TARGET_FS)
LDFLAGS += -rpath-link=$(TARGET_FS)/lib -L$(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/lib/aarch64-linux-gnu -L$(TARGET_FS)/lib/aarch64-linux-gnu
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L$(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L$(TARGET_FS)/usr/lib/aarch64-linux-gnu
LDFLAGS += --unresolved-symbols=ignore-in-shared-libs
CCFLAGS += -isystem=$(TARGET_FS)/usr/include -I$(TARGET_FS)/usr/include
CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu -I$(TARGET_FS)/usr/include/aarch64-linux-gnu
endif
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
CCFLAGS += -DWIN_INTERFACE_CUSTOM -I/usr/include/aarch64-qnx-gnu
LDFLAGS += -lsocket
LDFLAGS += -rpath=/usr/lib/aarch64-qnx-gnu -L/usr/lib/aarch64-qnx-gnu
ifneq ($(TARGET_FS),)
LDFLAGS += -rpath=$(TARGET_FS)/usr/lib -L $(TARGET_FS)/usr/lib
LDFLAGS += -rpath=$(TARGET_FS)/usr/libnvidia -L $(TARGET_FS)/usr/libnvidia
endif
ifdef TARGET_OVERRIDE # cuda toolkit targets override
NVCCFLAGS += -target-dir $(TARGET_OVERRIDE)
endif
endif
endif
# Install directory of different arch
CUDA_INSTALL_TARGET_DIR :=
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-gnueabihf/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),sbsa-linux)
CUDA_INSTALL_TARGET_DIR = targets/sbsa-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-qnx)
CUDA_INSTALL_TARGET_DIR = targets/ARMv7-linux-QNX/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-qnx/
else ifeq ($(TARGET_ARCH),ppc64le)
CUDA_INSTALL_TARGET_DIR = targets/ppc64le-linux/
endif
# Debug build flags
ifeq ($(dbg),1)
NVCCFLAGS += -g -G
BUILD_TYPE := debug
else
BUILD_TYPE := release
endif
ALL_CCFLAGS :=
ALL_CCFLAGS += $(NVCCFLAGS)
ALL_CCFLAGS += $(EXTRA_NVCCFLAGS)
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(CCFLAGS))
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(EXTRA_CCFLAGS))
SAMPLE_ENABLED := 1
# This sample is not supported on Mac OSX
ifeq ($(TARGET_OS),darwin)
$(info >>> WARNING - binaryPartitionCG is not supported on Mac OSX - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
ALL_LDFLAGS :=
ALL_LDFLAGS += $(ALL_CCFLAGS)
ALL_LDFLAGS += $(addprefix -Xlinker ,$(LDFLAGS))
ALL_LDFLAGS += $(addprefix -Xlinker ,$(EXTRA_LDFLAGS))
# Common includes and paths for CUDA
INCLUDES := -I../../Common
LIBRARIES :=
################################################################################
#Detect if installed version of GCC supports required C++11
ifeq ($(TARGET_OS),linux)
empty :=
space := $(empty) $(empty)
GCCVERSIONSTRING := $(shell expr `$(HOST_COMPILER) -dumpversion`)
#Create version number without "."
GCCVERSION := $(shell expr `echo $(GCCVERSIONSTRING)` | cut -f1 -d.)
GCCVERSION += $(shell expr `echo $(GCCVERSIONSTRING)` | cut -f2 -d.)
GCCVERSION += $(shell expr `echo $(GCCVERSIONSTRING)` | cut -f3 -d.)
# Make sure the version number has at least 3 decimals
GCCVERSION += 00
# Remove spaces from the version number
GCCVERSION := $(subst $(space),$(empty),$(GCCVERSION))
# Crop the version number to 3 decimals.
GCCVERSION := $(shell expr `echo $(GCCVERSION)` | cut -b1-3)
#$(warning $(GCCVERSION))
IS_MIN_VERSION := $(shell expr `echo $(GCCVERSION)` \>= 470)
ifeq ($(IS_MIN_VERSION), 1)
$(info >>> GCC Version is greater or equal to 4.7.0 <<<)
else
$(info >>> Waiving build. Minimum GCC version required is 4.7.0<<<)
SAMPLE_ENABLED := 0
endif
endif
# Gencode arguments
ifeq ($(TARGET_ARCH),$(filter $(TARGET_ARCH),armv7l aarch64))
SMS ?= 35 37 50 52 60 61 70 72 75 80
else
SMS ?= 35 37 50 52 60 61 70 75 80
endif
ifeq ($(SMS),)
$(info >>> WARNING - no SM architectures have been specified - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
ifeq ($(GENCODE_FLAGS),)
# Generate SASS code for each SM architecture listed in $(SMS)
$(foreach sm,$(SMS),$(eval GENCODE_FLAGS += -gencode arch=compute_$(sm),code=sm_$(sm)))
# Generate PTX code from the highest SM architecture in $(SMS) to guarantee forward-compatibility
HIGHEST_SM := $(lastword $(sort $(SMS)))
ifneq ($(HIGHEST_SM),)
GENCODE_FLAGS += -gencode arch=compute_$(HIGHEST_SM),code=compute_$(HIGHEST_SM)
endif
endif
ALL_CCFLAGS += --std=c++11
ifeq ($(SAMPLE_ENABLED),0)
EXEC ?= @echo "[@]"
endif
################################################################################
# Target rules
all: build
build: binaryPartitionCG
check.deps:
ifeq ($(SAMPLE_ENABLED),0)
@echo "Sample will be waived due to the above missing dependencies"
else
@echo "Sample is ready - all dependencies have been met"
endif
binaryPartitionCG.o:binaryPartitionCG.cu
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
binaryPartitionCG: binaryPartitionCG.o
$(EXEC) $(NVCC) $(ALL_LDFLAGS) $(GENCODE_FLAGS) -o $@ $+ $(LIBRARIES)
$(EXEC) mkdir -p ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
$(EXEC) cp $@ ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
run: build
$(EXEC) ./binaryPartitionCG
clean:
rm -f binaryPartitionCG binaryPartitionCG.o
rm -rf ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)/binaryPartitionCG
clobber: clean

View File

@@ -0,0 +1,64 @@
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE entry SYSTEM "SamplesInfo.dtd">
<entry>
<name>binaryPartitionCG</name>
<cflags>
<flag>--std=c++11</flag>
</cflags>
<description><![CDATA[This sample is a simple code that illustrates binary partition cooperative groups and reduce within the thread block.]]></description>
<devicecompilation>whole</devicecompilation>
<includepaths>
<path>./</path>
<path>../</path>
<path>../../common/inc</path>
</includepaths>
<keyconcepts>
<concept level="basic">Cooperative Groups</concept>
</keyconcepts>
<keywords>
<keyword>CUDA</keyword>
<keyword>Parallel Reduction</keyword>
<keyword>Cooperative Groups</keyword>
<keyword>CPP11</keyword>
</keywords>
<libraries>
</libraries>
<librarypaths>
</librarypaths>
<nsight_eclipse>true</nsight_eclipse>
<primary_file>binaryPartitionCG.cu</primary_file>
<scopes>
<scope>1:CUDA Basic Topics</scope>
</scopes>
<sm-arch>sm35</sm-arch>
<sm-arch>sm37</sm-arch>
<sm-arch>sm50</sm-arch>
<sm-arch>sm52</sm-arch>
<sm-arch>sm60</sm-arch>
<sm-arch>sm61</sm-arch>
<sm-arch>sm70</sm-arch>
<sm-arch>sm72</sm-arch>
<sm-arch>sm75</sm-arch>
<sm-arch>sm80</sm-arch>
<supported_envs>
<env>
<arch>x86_64</arch>
<platform>linux</platform>
</env>
<env>
<platform>windows7</platform>
</env>
<env>
<arch>arm</arch>
</env>
<env>
<arch>ppc64le</arch>
<platform>linux</platform>
</env>
</supported_envs>
<supported_sm_architectures>
<include>all</include>
</supported_sm_architectures>
<title>Binary Partition Cooperative Groups</title>
<type>exe</type>
</entry>

View File

@@ -0,0 +1,67 @@
# binaryPartitionCG - Binary Partition Cooperative Groups
## Description
This sample is a simple code that illustrates binary partition cooperative groups and reduce within the thread block.
## Key Concepts
Cooperative Groups
## Supported SM Architectures
[SM 3.5 ](https://developer.nvidia.com/cuda-gpus) [SM 3.7 ](https://developer.nvidia.com/cuda-gpus) [SM 5.0 ](https://developer.nvidia.com/cuda-gpus) [SM 5.2 ](https://developer.nvidia.com/cuda-gpus) [SM 6.0 ](https://developer.nvidia.com/cuda-gpus) [SM 6.1 ](https://developer.nvidia.com/cuda-gpus) [SM 7.0 ](https://developer.nvidia.com/cuda-gpus) [SM 7.2 ](https://developer.nvidia.com/cuda-gpus) [SM 7.5 ](https://developer.nvidia.com/cuda-gpus) [SM 8.0 ](https://developer.nvidia.com/cuda-gpus)
## Supported OSes
Linux, Windows
## Supported CPU Architecture
x86_64, ppc64le, armv7l
## CUDA APIs involved
## Prerequisites
Download and install the [CUDA Toolkit 11.0](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
## Build and Run
### Windows
The Windows samples are built using the Visual Studio IDE. Solution files (.sln) are provided for each supported version of Visual Studio, using the format:
```
*_vs<version>.sln - for Visual Studio <version>
```
Each individual sample has its own set of solution files in its directory:
To build/examine all the samples at once, the complete solution files should be used. To build/examine a single sample, the individual sample solution files should be used.
> **Note:** Some samples require that the Microsoft DirectX SDK (June 2010 or newer) be installed and that the VC++ directory paths are properly set up (**Tools > Options...**). Check DirectX Dependencies section for details."
### Linux
The Linux samples are built using makefiles. To use the makefiles, change the current directory to the sample directory you wish to build, and run make:
```
$ cd <sample_dir>
$ make
```
The samples makefiles can take advantage of certain options:
* **TARGET_ARCH=<arch>** - cross-compile targeting a specific architecture. Allowed architectures are x86_64, ppc64le, armv7l.
By default, TARGET_ARCH is set to HOST_ARCH. On a x86_64 machine, not setting TARGET_ARCH is the equivalent of setting TARGET_ARCH=x86_64.<br/>
`$ make TARGET_ARCH=x86_64` <br/> `$ make TARGET_ARCH=ppc64le` <br/> `$ make TARGET_ARCH=armv7l` <br/>
See [here](http://docs.nvidia.com/cuda/cuda-samples/index.html#cross-samples) for more details.
* **dbg=1** - build with debug symbols
```
$ make dbg=1
```
* **SMS="A B ..."** - override the SM architectures for which the sample will be built, where `"A B ..."` is a space-delimited list of SM architectures. For example, to generate SASS for SM 50 and SM 60, use `SMS="50 60"`.
```
$ make SMS="50 60"
```
* **HOST_COMPILER=<host_compiler>** - override the default g++ host compiler. See the [Linux Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#system-requirements) for a list of supported host compilers.
```
$ make HOST_COMPILER=g++
```
## References (for more details)

View File

@@ -0,0 +1,155 @@
/* Copyright (c) 2020, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of NVIDIA CORPORATION nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
/*
* This sample illustrates basic usage of binary partition cooperative groups
* within the thread block tile when divergent path exists.
* 1.) Each thread loads a value from random array.
* 2.) then checks if it is odd or even.
* 3.) create binary partition group based on the above predicate
* 4.) we count the number of odd/even in the group based on size of the binary groups
* 5.) write it global counter of odd.
* 6.) sum the values loaded by individual threads(using reduce) and write it to global
* even & odd elements sum.
*
* **NOTE** : binary_partition results in splitting warp into divergent thread groups
this is not good from performance perspective, but in cases where warp
divergence is inevitable one can use binary_partition group.
*/
#include <stdio.h>
#include <cooperative_groups.h>
#include <cooperative_groups/reduce.h>
#include <helper_cuda.h>
namespace cg = cooperative_groups;
void initOddEvenArr(int *inputArr, unsigned int size)
{
for (int i=0; i < size; i++)
{
inputArr[i] = rand() % 50;
}
}
/**
* CUDA kernel device code
*
* Creates cooperative groups and performs odd/even counting & summation.
*/
__global__ void oddEvenCountAndSumCG(int *inputArr, int *numOfOdds, int *sumOfOddAndEvens, unsigned int size)
{
cg::thread_block cta = cg::this_thread_block();
cg::grid_group grid = cg::this_grid();
cg::thread_block_tile<32> tile32 = cg::tiled_partition<32>(cta);
for (int i = grid.thread_rank(); i < size; i += grid.size())
{
int elem = inputArr[i];
auto subTile = cg::binary_partition(tile32, elem & 1);
if (elem & 1) // Odd numbers group
{
int oddGroupSum = cg::reduce(subTile, elem, cg::plus<int>());
if (subTile.thread_rank() == 0)
{
// Add number of odds present in this group of Odds.
atomicAdd(numOfOdds, subTile.size());
// Add local reduction of odds present in this group of Odds.
atomicAdd(&sumOfOddAndEvens[0], oddGroupSum);
}
}
else // Even numbers group
{
int evenGroupSum = cg::reduce(subTile, elem, cg::plus<int>());
if (subTile.thread_rank() == 0)
{
// Add local reduction of even present in this group of evens.
atomicAdd(&sumOfOddAndEvens[1], evenGroupSum);
}
}
// reconverge warp so for next loop iteration we ensure convergence of
// above diverged threads to perform coalesced loads of inputArr.
cg::sync(tile32);
}
}
/**
* Host main routine
*/
int main(int argc, const char **argv)
{
int deviceId = findCudaDevice(argc, argv);
int *h_inputArr, *d_inputArr;
int *h_numOfOdds, *d_numOfOdds;
int *h_sumOfOddEvenElems, *d_sumOfOddEvenElems;
unsigned int arrSize = 1024 * 100;
h_inputArr = new int[arrSize];
h_numOfOdds = new int[1];
h_sumOfOddEvenElems = new int[2];
initOddEvenArr(h_inputArr, arrSize);
cudaStream_t stream;
checkCudaErrors(cudaStreamCreateWithFlags(&stream, cudaStreamNonBlocking));
checkCudaErrors(cudaMalloc(&d_inputArr, sizeof(int)*arrSize));
checkCudaErrors(cudaMalloc(&d_numOfOdds, sizeof(int)));
checkCudaErrors(cudaMalloc(&d_sumOfOddEvenElems, sizeof(int)*2));
checkCudaErrors(cudaMemcpyAsync(d_inputArr, h_inputArr, sizeof(int)*arrSize, cudaMemcpyHostToDevice, stream));
checkCudaErrors(cudaMemsetAsync(d_numOfOdds, 0, sizeof(int), stream));
checkCudaErrors(cudaMemsetAsync(d_sumOfOddEvenElems, 0, 2*sizeof(int), stream));
//Launch the kernel
int threadsPerBlock=1024;
int blocksPerGrid = arrSize / threadsPerBlock;
printf("\nLaunching %d blocks with %d threads...\n\n",blocksPerGrid, threadsPerBlock);
oddEvenCountAndSumCG<<<blocksPerGrid, threadsPerBlock, 0, stream>>>(d_inputArr, d_numOfOdds, d_sumOfOddEvenElems, arrSize);
checkCudaErrors(cudaMemcpyAsync(h_numOfOdds, d_numOfOdds, sizeof(int), cudaMemcpyDeviceToHost, stream));
checkCudaErrors(cudaMemcpyAsync(h_sumOfOddEvenElems, d_sumOfOddEvenElems, 2*sizeof(int), cudaMemcpyDeviceToHost, stream));
printf("Array size = %d Num of Odds = %d Sum of Odds = %d Sum of Evens %d\n", arrSize, h_numOfOdds[0], h_sumOfOddEvenElems[0], h_sumOfOddEvenElems[1]);
printf("\n...Done.\n\n");
delete[] h_inputArr;
delete[] h_numOfOdds;
delete[] h_sumOfOddEvenElems;
checkCudaErrors(cudaFree(d_inputArr));
checkCudaErrors(cudaFree(d_numOfOdds));
checkCudaErrors(cudaFree(d_sumOfOddEvenElems));
return EXIT_SUCCESS;
}

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 14.00
# Visual Studio 2015
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "binaryPartitionCG", "binaryPartitionCG_vs2015.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,107 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>binaryPartitionCG_vs2015</RootNamespace>
<ProjectName>binaryPartitionCG</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v140</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/binaryPartitionCG.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="binaryPartitionCG.cu" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 12.00
# Visual Studio 2017
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "binaryPartitionCG", "binaryPartitionCG_vs2017.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,112 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>binaryPartitionCG_vs2017</RootNamespace>
<ProjectName>binaryPartitionCG</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<PropertyGroup Condition="'$(WindowsTargetPlatformVersion)'==''">
<LatestTargetPlatformVersion>$([Microsoft.Build.Utilities.ToolLocationHelper]::GetLatestSDKTargetPlatformVersion('Windows', '10.0'))</LatestTargetPlatformVersion>
<WindowsTargetPlatformVersion Condition="'$(WindowsTargetPlatformVersion)' == ''">$(LatestTargetPlatformVersion)</WindowsTargetPlatformVersion>
<TargetPlatformVersion>$(WindowsTargetPlatformVersion)</TargetPlatformVersion>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v141</PlatformToolset>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/binaryPartitionCG.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="binaryPartitionCG.cu" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup>
</Project>

View File

@@ -0,0 +1,20 @@

Microsoft Visual Studio Solution File, Format Version 12.00
# Visual Studio 2019
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "binaryPartitionCG", "binaryPartitionCG_vs2019.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|x64 = Debug|x64
Release|x64 = Release|x64
EndGlobalSection
GlobalSection(ProjectConfigurationPlatforms) = postSolution
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
EndGlobalSection
EndGlobal

View File

@@ -0,0 +1,108 @@
<?xml version="1.0" encoding="utf-8"?>
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
<PropertyGroup>
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
</PropertyGroup>
<ItemGroup Label="ProjectConfigurations">
<ProjectConfiguration Include="Debug|x64">
<Configuration>Debug</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
<ProjectConfiguration Include="Release|x64">
<Configuration>Release</Configuration>
<Platform>x64</Platform>
</ProjectConfiguration>
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
<RootNamespace>binaryPartitionCG_vs2019</RootNamespace>
<ProjectName>binaryPartitionCG</ProjectName>
<CudaToolkitCustomDir />
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
<PropertyGroup>
<ConfigurationType>Application</ConfigurationType>
<CharacterSet>MultiByte</CharacterSet>
<PlatformToolset>v142</PlatformToolset>
<WindowsTargetPlatformVersion>10.0</WindowsTargetPlatformVersion>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
<UseDebugLibraries>true</UseDebugLibraries>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)'=='Release'">
<WholeProgramOptimization>true</WholeProgramOptimization>
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
</ImportGroup>
<PropertyGroup Label="UserMacros" />
<PropertyGroup>
<IntDir>$(Platform)/$(Configuration)/</IntDir>
<IncludePath>$(IncludePath)</IncludePath>
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
<CodeAnalysisRules />
<CodeAnalysisRuleAssemblies />
</PropertyGroup>
<PropertyGroup Condition="'$(Platform)'=='x64'">
<OutDir>../../bin/win64/$(Configuration)/</OutDir>
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<WarningLevel>Level3</WarningLevel>
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
</ClCompile>
<Link>
<SubSystem>Console</SubSystem>
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
<OutputFile>$(OutDir)/binaryPartitionCG.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
<ClCompile>
<Optimization>Disabled</Optimization>
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>true</GenerateDebugInformation>
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MTd</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
<ClCompile>
<Optimization>MaxSpeed</Optimization>
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
</ClCompile>
<Link>
<GenerateDebugInformation>false</GenerateDebugInformation>
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
</Link>
<CudaCompile>
<Runtime>MT</Runtime>
<TargetMachinePlatform>64</TargetMachinePlatform>
</CudaCompile>
</ItemDefinitionGroup>
<ItemGroup>
<CudaCompile Include="binaryPartitionCG.cu" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.0.targets" />
</ImportGroup>
</Project>

File diff suppressed because one or more lines are too long

Some files were not shown because too many files have changed in this diff Show More