Merge branch 'develop'

CMakeLists.txt

@@ -43,9 +43,9 @@ option( BUILD_WITH_TENSILE "Building rocBLAS with Tensile or not" ON)
 # which benchmark solution schedule
 set( Tensile_ROOT "" CACHE STRING "Local path of Tensile.")
 if ( BUILD_WITH_TENSILE )
-  set( Tensile_LOGIC "FijiROCm14" CACHE STRING "Tensile to use which logic?")
+  set( Tensile_LOGIC "FijiROCm15" CACHE STRING "Tensile to use which logic?")
   set_property( CACHE Tensile_LOGIC PROPERTY STRINGS
-    FijiROCm14
+    FijiROCm15
     )
   option( Tensile_MERGE_FILES "Tensile to merge kernels and solutions files?" OFF)
   option( Tensile_SHORT_FILENAMES "Tensile to use short file names? Use if compiler complains they're too long." OFF)
@@ -149,6 +149,7 @@ if( BUILD_LIBRARY )
     -DCMAKE_MODULE_PATH=${LIBRARY_MODULE_PATH}
     -DBUILD_SHARED_LIBS=${BUILD_SHARED_LIBS}
     -DCMAKE_CXX_COMPILER=${DEVICE_CXX_COMPILER}
+    -DCMAKE_CXX_FLAGS=${CMAKE_CXX_FLAGS}
     -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
     -DBUILD_WITH_TENSILE=${BUILD_WITH_TENSILE}
     -DTensile_LOGIC=${Tensile_LOGIC}
@@ -157,6 +158,11 @@ if( BUILD_LIBRARY )
     -DTensile_PRINT_DEBUG=${Tensile_PRINT_DEBUG}
     -DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE}
   )
+
+  if( DEFINED CPACK_PACKAGING_INSTALL_PREFIX )
+    list( APPEND LIBRARY_CMAKE_ARGS -DCPACK_PACKAGING_INSTALL_PREFIX=${CPACK_PACKAGING_INSTALL_PREFIX} )
+  endif()
+
   if (Tensile_FOUND)
     set( LIBRARY_CMAKE_ARGS ${LIBRARY_CMAKE_ARGS} -DTensile_FOUND=${Tensile_FOUND})
   else()
@@ -195,6 +201,7 @@ if( BUILD_CLIENTS )
     -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
     -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
     -DDEVICE_CXX_COMPILER=${DEVICE_CXX_COMPILER}
+    -DCMAKE_CXX_FLAGS=${CMAKE_CXX_FLAGS}
     -DCMAKE_PREFIX_PATH=${CLIENT_PREFIX_PATH}
     -DCMAKE_MODULE_PATH=${CLIENT_MODULE_PATH}
     -DBUILD_CLIENTS_SAMPLES=${BUILD_CLIENTS_SAMPLES}

Jenkinsfile

@@ -6,15 +6,19 @@ properties([buildDiscarder(logRotator(
     artifactNumToKeepStr: '',
     daysToKeepStr: '',
     numToKeepStr: '10')),
-  disableConcurrentBuilds()])
+    disableConcurrentBuilds(),
+    [$class: 'CopyArtifactPermissionProperty', projectNames: '*']
+   ])
 
-def build_type="Debug"
-def build_type_postfix="-d"
+// def build_type="Debug"
+// def build_type_postfix="-d"
+def build_type="Release"
+def build_type_postfix=""
 
 // Currently, YADP (yet-another-docker-plugin v0.1.0-rc30) does not load balance between clouds with the same label
 // They recommend to use docker swarm, but not yet work with docker 1.12 'swarm mode'
 // Manually load balance by picking a particular machine
-node('rocm-1.3 && hawaii')
+node('rocm-1.5 && fiji')
 {
   def node_list = env.NODE_LABELS.tokenize()
   // sh "echo node_list: ${node_list}"
@@ -67,40 +71,25 @@ node('rocm-1.3 && hawaii')
           stage("configure clang release") {
               sh """#!/usr/bin/env bash
                 sudo apt-get update
-                sudo apt-get install python-yaml
-                cmake -DCMAKE_BUILD_TYPE=${build_type} -DCMAKE_PREFIX_PATH=/opt/boost/clang -DBUILD_LIBRARY=ON -DBUILD_WITH_TENSILE=ON \
-                -DBUILD_CLIENTS=ON -DBUILD_CLIENTS_SAMPLES=ON -DBUILD_CLIENTS_TESTS=ON ${scm_dir}
+                sudo DEBIAN_FRONTEND=noninteractive apt-get install --no-install-recommends -y rpm
+                cmake -DCMAKE_BUILD_TYPE=${build_type} -DCMAKE_PREFIX_PATH=/opt/boost/clang-3.8 -DBUILD_SHARED_LIBS=ON -DBUILD_LIBRARY=ON -DBUILD_WITH_TENSILE=ON \
+                -DBUILD_CLIENTS=ON -DBUILD_CLIENTS_SAMPLES=ON -DBUILD_CLIENTS_TESTS=ON -DCPACK_PACKAGING_INSTALL_PREFIX=/opt/rocm/rocblas ${scm_dir}
                 """
           }
 
           stage("Build")
           {
-              if (env.NODE_LABELS ==~ /.*fiji.*/)
-              {
-              sh 'echo Target Fiji ISA'
-                withEnv(['HCC_AMDGPU_TARGET=AMD:AMDGPU:8:0:3'])
-                {
-                  sh '''#!/usr/bin/env bash
-                        make -j 8
-                    '''
-                }
-              }
-              else if (env.NODE_LABELS ==~ /.*hawaii.*/)
-              {
-                sh 'echo Target Hawaii ISA'
-                withEnv(['HCC_AMDGPU_TARGET=AMD:AMDGPU:7:0:1'])
-                {
-                    sh '''#!/usr/bin/env bash
-                          make -j 8
-                      '''
-                }
-              }
+              sh '''#!/usr/bin/env bash
+                    make -j $(nproc)
+                '''
           }
 
           stage("Package Debian") {
             sh 'cd library-build; make package'
-            archive includes: 'library-build/*.deb'
-          }
+            archiveArtifacts artifacts: 'library-build/*.deb', fingerprint: true
+            archiveArtifacts artifacts: 'library-build/*.rpm', fingerprint: true
+            sh "sudo dpkg -c library-build/*.deb"
+         }
 
           // Cap the maximum amount of testing to be a few hours; assume failure if the time limit is hit
           timeout(time: 1, unit: 'HOURS')

clients/CMakeLists.txt

@@ -18,12 +18,12 @@ set( rocblas_clients_dependencies )
 
 # If the user does not explicitely specify BOOST_ROOT, build our
 # own on the fly
-#if( BUILD_CLIENTS_BENCHMARKS )
-#  if( NOT DEFINED BOOST_ROOT )
-#    include( external-boost )
-#    list( APPEND rocblas_clients_dependencies boost )
-#  endif( )
-#endif()
+if( BUILD_CLIENTS_BENCHMARKS )
+  if( NOT DEFINED BOOST_ROOT )
+    include( external-boost )
+    list( APPEND rocblas_clients_dependencies boost )
+  endif( )
+endif()
 
 if( BUILD_CLIENTS_TESTS )
   if( NOT DEFINED GTEST_ROOT )

clients/benchmarks/perf_script/README.txt

@@ -1,13 +1,16 @@
 
-Instruction to run the script to collect performance data
+Instruction to run the python script to collect performance data
 
-1) copy client.exe (Windows, client on Linux) to this folder 
+1) copy client.exe (Windows, client on Linux) to this folder
+   For example, "cp ../client ./" on linux
 
 2) "python measurePerformance.py --help" to see how to run the command line mode
     For example, "python measurePerformance.py -s 64-5760:64 -f gemm -r s --transa none --transb transpose > sgemm.txt" 
     will collect performance of sgemm NT of square matrices [64:5760] with step 64.
     All the output will be dumped into sgemm.txt
 
-3) Open the sgemm.txt and use "grep" to fetch the performance line. You can extract the Gflop/s number from the performance line
+3)  Plot the figure by refactoring with plotPerformance.py provided. 
+    "python plotPerformance.py" by default
+    will read step (2) output sgemm.txt and plot a performance figure in sgemm.pdf
 
 4) You can ingore the other automatic generated .txt files and folders during the run.