Merge remote-tracking branch 'upstream/develop' into develop

jlgreathouse · jlgreathouse · commit 3c9593036ae3 · 2015-10-16T13:39:06.000-05:00
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -38,7 +38,7 @@ endif()
 if( POLICY CMP0048 )
   cmake_policy( SET CMP0048 NEW )
 
-  project( SuperBuild.clSPARSE VERSION 0.7.2.0 )
+  project( SuperBuild.clSPARSE VERSION 0.8.0.0 )
 else( )
   project( SuperBuild.clSPARSE )
 
@@ -48,11 +48,11 @@ else( )
   endif( )
 
   if( NOT DEFINED SuperBuild.clSPARSE_VERSION_MINOR )
-    set( SuperBuild.clSPARSE_VERSION_MINOR 7 )
+    set( SuperBuild.clSPARSE_VERSION_MINOR 8 )
   endif( )
 
   if( NOT DEFINED SuperBuild.clSPARSE_VERSION_PATCH )
-    set( SuperBuild.clSPARSE_VERSION_PATCH 2 )
+    set( SuperBuild.clSPARSE_VERSION_PATCH 0 )
   endif( )
 
   if( NOT DEFINED SuperBuild.clSPARSE_VERSION_TWEAK )
diff --git a/README.md b/README.md
@@ -4,17 +4,25 @@ Pre-built binaries are available on our [releases page](https://github.com/clMat
 | Build branch | master | develop |
 |-----|-----|-----|
 | GCC/Clang x64 | [![Build Status](https://travis-ci.org/clMathLibraries/clSPARSE.svg?branch=master)](https://travis-ci.org/clMathLibraries/clSPARSE/branches) | [![Build Status](https://travis-ci.org/clMathLibraries/clSPARSE.svg?branch=develop)](https://travis-ci.org/clMathLibraries/clSPARSE/branches) |
-| Visual Studio x64 |  |[![Build status](https://ci.appveyor.com/api/projects/status/93518qe0efy6n7fy/branch/develop?svg=true)](https://ci.appveyor.com/project/kknox/clsparse-otonj/branch/develop) |
+| Visual Studio x64 |[![Build status](https://ci.appveyor.com/api/projects/status/93518qe0efy6n7fy/branch/master?svg=true)](https://ci.appveyor.com/project/kknox/clsparse-otonj/branch/master) |[![Build status](https://ci.appveyor.com/api/projects/status/93518qe0efy6n7fy/branch/develop?svg=true)](https://ci.appveyor.com/project/kknox/clsparse-otonj/branch/develop) |
 
 # clSPARSE
-an OpenCL&copy; library implementing Sparse linear algebra.  This project is a result of
+an OpenCL&trade; library implementing Sparse linear algebra routines.  This project is a result of
 a collaboration between [AMD Inc.](http://www.amd.com/) and
 [Vratis Ltd.](http://www.vratis.com/).
 
-## Introduction to clSPARSE
-At this time, clSPARSE provides these fundamental sparse operations for OpenCL:
+### What's new in clSPARSE **v0.8**
+- New single precision SpM-SpM (SpGEMM) function
+- Optimizations to the sparse matrix conversion routines
+- [API documentation](http://clmathlibraries.github.io/clSPARSE/) available
+- SpM-dV routines now provide [higher precision accuracy] (https://github.com/clMathLibraries/clSPARSE/wiki/Precision)
+- Various bug fixes integrated
+
+
+## clSPARSE features
 -  Sparse Matrix - dense Vector multiply (SpM-dV)
 -  Sparse Matrix - dense Matrix multiply (SpM-dM)
+-  Sparse Matrix - Sparse Matrix multiply Sparse Matrix Multiply(SpGEMM) - Single Precision
 -  Iterative conjugate gradient solver (CG)
 -  Iterative biconjugate gradient stabilized solver (BiCGStab)
 -  Dense to CSR conversions (& converse)
@@ -28,10 +36,6 @@ compared to the older clMath libraries.  OpenCL state is not explicitly passed
 through the API, which enables the library to be forward compatible when users are
 ready to switch from OpenCL 1.2 to OpenCL 2.0 <sup>[1](#opencl-2)</sup>
 
-The API’s are designed such that users are in control of where input and output
-buffers live, and they maintain control of when data transfers to/from device
-memory happen, so that there are no performance surprises.
-
 ### Google Groups
 Two mailing lists have been created for the clMath projects:
 
@@ -58,7 +62,7 @@ script for clSPARSE also builds the samples as an external project, to demonstra
 how an application would find and link to clSPARSE with cmake.
 
 ### clSPARSE library documentation
-**API documentation** is not yet available, but the samples above give an excellent
+**API documentation** is now available http://clmathlibraries.github.io/clSPARSE/ . The included samples will give an excellent
 starting point to basic library operations.
 
 ### Contributing code
diff --git a/docs/Doxyfile b/docs/Doxyfile
@@ -38,7 +38,7 @@ PROJECT_NAME           = clSPARSE
 # could be handy for archiving the generated documentation or if some version
 # control system is used.
 
-PROJECT_NUMBER         = v0.6.2.0
+PROJECT_NUMBER         = v0.8.0.0
 
 # Using the PROJECT_BRIEF tag one can provide an optional one line description
 # for a project that appears at the top of each page and should give viewer a
diff --git a/docs/performance/clSPARSE-0.8.0.0/W9100/Coo2Csr.csv b/docs/performance/clSPARSE-0.8.0.0/W9100/Coo2Csr.csv
@@ -0,0 +1,18 @@
+Coo2Csr, 
+,
+OpenCL runtime,1642.5 (VM)
+OpenCL Device,w9100
+,
+Matrix Name,Gi-Elements/s
+cant,0.825084
+consph,0.934099
+cop20k_A,0.731746
+mac_econ_fwd500,0.502334
+mc2depi,0.586327
+pdb1HYS,0.849179
+pwtk,1.14935
+rail4284,1.08582
+rma10,0.696027
+scircuit,0.441488
+shipsec1,2.08336
+webbase_1M,0.773179
diff --git a/docs/performance/clSPARSE-0.8.0.0/W9100/Csr2Coo.csv b/docs/performance/clSPARSE-0.8.0.0/W9100/Csr2Coo.csv
@@ -0,0 +1,18 @@
+Csr2Coo, 
+,
+OpenCL runtime,1642.5 (VM)
+OpenCL Device,w9100
+,
+Matrix Name,Gi-Elements/s
+cant,4.22181
+consph,4.92386
+cop20k_A,3.71303
+mac_econ_fwd500,2.16156
+mc2depi,3.24346
+pdb1HYS,4.52208
+pwtk,5.42111
+rail4284,5.30094
+rma10,3.55844
+scircuit,1.93272
+shipsec1,4.31598
+webbase_1M,2.96583
diff --git a/docs/performance/clSPARSE-0.8.0.0/W9100/Csr2Dense.csv b/docs/performance/clSPARSE-0.8.0.0/W9100/Csr2Dense.csv
@@ -0,0 +1,16 @@
+Csr2Dense, 
+,
+OpenCL runtime,1642.5 (VM)
+OpenCL Device,w9100
+,
+Matrix Name,Gi-Elements/s
+Dubcova1,0.025644
+hydr1c_A_11,0.0169621
+hydr1c_A_72,0.0170693
+hydr1c_A_76,0.0179919
+Maragal_6,0.0623863
+Na5,0.165182
+psse1,0.00876455
+Reuters911,0.0398191
+Si10H16,0.0777071
+tomography,0.101941
diff --git a/docs/performance/clSPARSE-0.8.0.0/W9100/Dense2Csr.csv b/docs/performance/clSPARSE-0.8.0.0/W9100/Dense2Csr.csv
@@ -0,0 +1,16 @@
+Dense2Csr, 
+,
+OpenCL runtime,1642.5 (VM)
+OpenCL Device,w9100
+,
+Matrix Name,Gi-Elements/s
+Dubcova1,3.97644
+hydr1c_A_11,2.69697
+hydr1c_A_72,2.70712
+hydr1c_A_76,2.68411
+Maragal_6,3.93096
+Na5,2.70179
+psse1,3.80035
+Reuters911,3.81953
+Si10H16,4.00379
+tomography,0.0410165
diff --git a/docs/performance/clSPARSE-0.8.0.0/W9100/README.md b/docs/performance/clSPARSE-0.8.0.0/W9100/README.md
@@ -0,0 +1,31 @@
+# Benchmarking
+## Hardware
+w9100
+
+## Environment
+Ubuntu 14.04
+
+clSPARSE v0.8.0.0
+
+[Catalyst FirePro](http://support.amd.com/en-us/download/workstation?os=Linux%20x86_64#catalyst-pro) 14.502.1040
+
+## Tool
+[clsparse-bench](clSPARSE\src\benchmarks\clsparse-bench)
+
+## Methodology
+For each data point, we took 20 samples. Each sample consists of 20 calls
+with a wait afterward. We benchmark with respect to the API, utilizing host timers
+(not pure kernel time with ).  
+Outlying samples beyond 1 standard deviation were removed.
+
+Conversion routines benchmarked as number of Gi-Elements/s converted
+
+SpM-dV routine calculated as Gi-Bytes/s
+```c
+( sizeof( cl_int )*( csrMtx.num_nonzeros + csrMtx.num_rows ) + sizeof( T ) * ( csrMtx.num_nonzeros + csrMtx.num_cols + csrMtx.num_rows ) ) / time_in_ns( );
+```
+SpGEMM routine calculated as Mega-Flops/s
+```c
+ (2 * (upper bound of number of nonzeros of result matrix))/ time_in_ms( ) ;
+ ```
+ 
diff --git a/docs/performance/clSPARSE-0.8.0.0/W9100/SpGEMM.csv b/docs/performance/clSPARSE-0.8.0.0/W9100/SpGEMM.csv
@@ -0,0 +1,27 @@
+SpGemm, 
+,
+OpenCL runtime,1642.5 (VM)
+OpenCL Device,w9100
+,
+Matrix Name,G-Flop/s
+2cubes_sphere,0.928289
+cage12,1.06262
+cant,3.41712
+consph,3.47913
+cop20k_A,1.79591
+filter3D,1.73384
+hood,4.59367
+m133-b3,0.499885
+mac_econ_fwd500,0.36654
+majorbasis,1.69818
+mario002,0.647947
+mc2depi,0.564147
+offshore,1.15078
+patents_main,0.0800872
+pdb1HYS,3.18338
+poisson3Da,1.02704
+pwtk,4.93038
+rma10,3.59276
+scircuit,0.247432
+shipsec1,1.75098
+webbase_1M,0.278071
diff --git a/docs/performance/clSPARSE-0.8.0.0/W9100/SpM-dV.csv b/docs/performance/clSPARSE-0.8.0.0/W9100/SpM-dV.csv
@@ -0,0 +1,18 @@
+SpM-dV, 
+,
+OpenCL runtime,1642.5 (VM)
+OpenCL Device,w9100
+,
+Matrix Name,Gi-Bytes/s
+cant,133.421
+consph,187.416
+cop20k_A,92.3342
+mac_econ_fwd500,80.9133
+mc2depi,112.604
+pdb1HYS,144.192
+pwtk,182.1
+rail4284,65.8575
+rma10,109.81
+scircuit,61.9386
+shipsec1,128.57
+webbase_1M,129.624
diff --git a/docs/performance/cuSPARSE-7.5/Tesla_K40/Coo2Csr.csv b/docs/performance/cuSPARSE-7.5/Tesla_K40/Coo2Csr.csv
@@ -0,0 +1,18 @@
+Coo2Csr, 
+,
+CUDA Driver,352.39
+OpenCL Device,k40c
+,
+Matrix Name,Gi-Elements/s
+cant,0.311614
+consph,0.0322091
+cop20k_A,0.0012033
+mac_econ_fwd500,0.152202
+mc2depi,0.229676
+pdb1HYS,0.0620883
+pwtk,0.0158725
+rail4284,0.0706112
+rma10,0.0198922
+scircuit,0.00166464
+shipsec1,0.0521875
+webbase_1M,0.000965286
diff --git a/docs/performance/cuSPARSE-7.5/Tesla_K40/Csr2Coo.csv b/docs/performance/cuSPARSE-7.5/Tesla_K40/Csr2Coo.csv
@@ -0,0 +1,18 @@
+Csr2Coo, 
+,
+CUDA Driver,352.39
+OpenCL Device,k40c
+,
+Matrix Name,Gi-Elements/s
+cant,12.5796
+consph,13.2247
+cop20k_A,6.11234
+mac_econ_fwd500,1.88724
+mc2depi,1.2577
+pdb1HYS,16.6086
+pwtk,12.4184
+rail4284,14.4921
+rma10,10.8246
+scircuit,1.714
+shipsec1,13.0084
+webbase_1M,0.989887
diff --git a/docs/performance/cuSPARSE-7.5/Tesla_K40/Csr2Dense.csv b/docs/performance/cuSPARSE-7.5/Tesla_K40/Csr2Dense.csv
@@ -0,0 +1,16 @@
+Csr2Dense, 
+,
+CUDA Driver,352.39
+OpenCL Device,k40c
+,
+Matrix Name,Gi-Elements/s
+Dubcova1,0.010733
+hydr1c_A_11,0.0121546
+hydr1c_A_72,0.0121434
+hydr1c_A_76,0.0121483
+Maragal_6,0.0259574
+Na5,0.155875
+psse1,0.00518171
+Reuters911,0.0217409
+Si10H16,0.0308072
+tomography,0.569722
diff --git a/docs/performance/cuSPARSE-7.5/Tesla_K40/Dense2Csr.csv b/docs/performance/cuSPARSE-7.5/Tesla_K40/Dense2Csr.csv
@@ -0,0 +1,16 @@
+Dense2Csr, 
+,
+CUDA Driver,352.39
+OpenCL Device,k40c
+,
+Matrix Name,Gi-Elements/s
+Dubcova1,12.2282
+hydr1c_A_11,16.4208
+hydr1c_A_72,16.4242
+hydr1c_A_76,16.4017
+Maragal_6,12.861
+Na5,13.1853
+psse1,15.7104
+Reuters911,13.3711
+Si10H16,12.6937
+tomography,0.810857
diff --git a/docs/performance/cuSPARSE-7.5/Tesla_K40/README.md b/docs/performance/cuSPARSE-7.5/Tesla_K40/README.md
@@ -0,0 +1,31 @@
+# Benchmarking
+## Hardware
+Tesla K40c
+
+## Environment
+OpenSUSE 13.2
+
+cuSPARSE v7.5
+
+Tesla driver 352.39
+
+## Tool
+[cusparse-bench](clSPARSE\src\benchmarks\cusparse-bench)
+
+## Methodology
+For each data point, we took 20 samples. Each sample consists of 20 calls
+with a wait afterward. We benchmark with respect to the API, utilizing host timers
+(not pure kernel time with ).  
+Outlying samples beyond 1 standard deviation were removed.
+
+Conversion routines benchmarked as number of Gi-Elements/s converted
+
+SpM-dV routine calculated as Gi-Bytes/s
+```c
+( sizeof( cl_int )*( csrMtx.num_nonzeros + csrMtx.num_rows ) + sizeof( T ) * ( csrMtx.num_nonzeros + csrMtx.num_cols + csrMtx.num_rows ) ) / time_in_ns( );
+```
+SpGEMM routine calculated as Mega-Flops/s
+```c
+ (2 * (upper bound of number of nonzeros of result matrix))/ time_in_ms( ) ;
+ ```
+ 
diff --git a/docs/performance/cuSPARSE-7.5/Tesla_K40/SpGEMM.csv b/docs/performance/cuSPARSE-7.5/Tesla_K40/SpGEMM.csv
@@ -0,0 +1,27 @@
+SpGEMM, 
+,
+CUDA Driver,352.39
+OpenCL Device,k40c
+,
+Matrix Name,G-Flop/s
+2cubes_sphere,0.895677
+cage12,0.48655
+cant,5.12327
+consph,4.30423
+cop20k_A,0.67811
+filter3D,1.11123
+hood,1.9952
+m133-b3,0.15196
+mac_econ_fwd500,0.38369
+majorbasis,0.541907
+mario002,0.169338
+mc2depi,0.346675
+offshore,0.691809
+patents_main,0.0662424
+pdb1HYS,6.17243
+poisson3Da,0.911867
+pwtk,4.96245
+rma10,4.18639
+scircuit,0.296162
+shipsec1,1.64348
+webbase_1M,0.130227
diff --git a/docs/performance/cuSPARSE-7.5/Tesla_K40/SpM-dV.csv b/docs/performance/cuSPARSE-7.5/Tesla_K40/SpM-dV.csv
@@ -0,0 +1,18 @@
+SpM-dV, 
+,
+CUDA Driver,352.39
+OpenCL Device,k40c
+,
+Matrix Name,Gi-Bytes/s
+cant,132.192
+consph,136.732
+cop20k_A,85.7759
+mac_econ_fwd500,66.1886
+mc2depi,100.872
+pdb1HYS,137.62
+pwtk,94.678
+rail4284,45.4369
+rma10,87.7385
+scircuit,68.8223
+shipsec1,216.15
+webbase_1M,46.0537
diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
diff --git a/src/benchmarks/cusparse-bench/functions/cufunc_xCoo2Csr.hpp b/src/benchmarks/cusparse-bench/functions/cufunc_xCoo2Csr.hpp
