Merge 2.10 develop into new tensile client

CMakeLists.txt

@@ -146,7 +146,7 @@ get_os_id(OS_ID)
 message (STATUS "OS detected is ${OS_ID}")
 
 # Versioning via rocm-cmake
-set ( VERSION_STRING "2.9.0" )
+set ( VERSION_STRING "2.11.0" )
 rocm_setup_version( VERSION ${VERSION_STRING} )
 
 # Append our library helper cmake path and the cmake path for hip (for convenience)
@@ -196,7 +196,7 @@ if( BUILD_WITH_TENSILE )
     message (STATUS "using GIT Tensile fork=${tensile_fork} from branch=${tensile_tag}")
   endif()
   list(APPEND CMAKE_PREFIX_PATH ${VIRTUALENV_HOME_DIR})
-  find_package(Tensile 4.12.0 EXACT REQUIRED HIP LLVM OpenMP PATHS "${INSTALLED_TENSILE_PATH}")
+  find_package(Tensile 4.13.0 EXACT REQUIRED HIP LLVM OpenMP PATHS "${INSTALLED_TENSILE_PATH}")
 endif()
 
 # Find HCC/HIP dependencies

Jenkinsfile

@@ -76,77 +76,35 @@ rocBLASCI:
     {
         platform, project->
 
-        def command
-
-        if(platform.jenkinsLabel.contains('centos') || platform.jenkinsLabel.contains('sles'))
-        {
-            if(auxiliary.isJobStartedByTimer())
-            {
-                command = """#!/usr/bin/env bash
-                        set -x
-                        cd ${project.paths.project_build_prefix}/build/release/clients/staging
-                        LD_LIBRARY_PATH=/opt/rocm/hcc/lib GTEST_LISTENER=NO_PASS_LINE_IN_LOG sudo ./rocblas-test --gtest_output=xml --gtest_color=yes --gtest_filter=*nightly*-*known_bug* #--gtest_filter=*nightly*
-                    """
+        String sudo = auxiliary.sudo(platform.jenkinsLabel)
+        def gfilter = auxiliary.isJobStartedByTimer() ? "*nightly*" : "*quick*:*pre_checkin*"
 
-                platform.runCommand(this, command)
-                junit "${project.paths.project_build_prefix}/build/release/clients/staging/*.xml"
-            }
-            else
-            {
-                command = """#!/usr/bin/env bash
+        def command = """#!/usr/bin/env bash
                         set -x
                         cd ${project.paths.project_build_prefix}/build/release/clients/staging
-                        LD_LIBRARY_PATH=/opt/rocm/hcc/lib ./example-sscal
-                        LD_LIBRARY_PATH=/opt/rocm/hcc/lib GTEST_LISTENER=NO_PASS_LINE_IN_LOG sudo ./rocblas-test --gtest_output=xml --gtest_color=yes  --gtest_filter=*quick*:*pre_checkin*-*known_bug* #--gtest_filter=*checkin*
+                        ${sudo} LD_LIBRARY_PATH=/opt/rocm/hcc/lib GTEST_LISTENER=NO_PASS_LINE_IN_LOG ./rocblas-test --gtest_output=xml --gtest_color=yes --gtest_filter=${gfilter}-*known_bug*
                     """
 
-                platform.runCommand(this, command)
-                junit "${project.paths.project_build_prefix}/build/release/clients/staging/*.xml"
-            }
-        }
-        else
-        {
-            if(auxiliary.isJobStartedByTimer())
-            {
-                command = """#!/usr/bin/env bash
-                        set -x
-                        cd ${project.paths.project_build_prefix}/build/release/clients/staging
-                        LD_LIBRARY_PATH=/opt/rocm/hcc/lib GTEST_LISTENER=NO_PASS_LINE_IN_LOG ./rocblas-test --gtest_output=xml --gtest_color=yes --gtest_filter=*nightly*-*known_bug* #--gtest_filter=*nightly*
-                    """
-
-                platform.runCommand(this, command)
-                junit "${project.paths.project_build_prefix}/build/release/clients/staging/*.xml"
-            }
-            else
-            {
-                command = """#!/usr/bin/env bash
-                        set -x
-                        cd ${project.paths.project_build_prefix}/build/release/clients/staging
-                        LD_LIBRARY_PATH=/opt/rocm/hcc/lib ./example-sscal
-                        LD_LIBRARY_PATH=/opt/rocm/hcc/lib GTEST_LISTENER=NO_PASS_LINE_IN_LOG ./rocblas-test --gtest_output=xml --gtest_color=yes  --gtest_filter=*quick*:*pre_checkin*-*known_bug* #--gtest_filter=*checkin*
-                    """
-
-                platform.runCommand(this, command)
-                junit "${project.paths.project_build_prefix}/build/release/clients/staging/*.xml"
-            }
-        }
+        platform.runCommand(this, command)
+        junit "${project.paths.project_build_prefix}/build/release/clients/staging/*.xml"
     }
 
     def packageCommand =
     {
         platform, project->
+        String sudo = auxiliary.sudo(platform.jenkinsLabel)
 
         def command
 
-        if(platform.jenkinsLabel.contains('centos'))
+        if(platform.jenkinsLabel.contains('centos') || platform.jenkinsLabel.contains('sles'))
         {
             command = """
                     set -x
                     cd ${project.paths.project_build_prefix}/build/release
-                    make package
-                    mkdir -p package
-                    mv *.rpm package/
-                    rpm -qlp package/*.rpm
+                    ${sudo} make package
+                    ${sudo} mkdir -p package
+                    ${sudo} mv *.rpm package/
+                    ${sudo} rpm -qlp package/*.rpm
                 """
 
             platform.runCommand(this, command)

bump_develop_version.sh

@@ -5,12 +5,16 @@
 # - run this script in master branch 
 # - after running this script merge master into develop 
 
-OLD_ROCBLAS_VERSION="2.8.0"
-NEW_ROCBLAS_VERSION="2.9.0"
+OLD_ROCBLAS_VERSION="2.10.0"
+NEW_ROCBLAS_VERSION="2.11.0"
 
-OLD_TENSILE_VERSION="tensile_tag 015477ad8c8ea0ef6f59b8d49a65015b46b8a48e"
-NEW_TENSILE_VERSION="tensile_tag 015477ad8c8ea0ef6f59b8d49a65015b46b8a48e"
+OLD_TENSILE_VERSION="Tensile 4.12.0"
+NEW_TENSILE_VERSION="Tensile 4.13.0"
+
+OLD_TENSILE_HASH="fe4f721886d07eef6251cea4225e027181022aa5"
+NEW_TENSILE_HASH="a9379f4e42efb754c9a618047bfbf292d74dfd0f"
 
 sed -i "s/${OLD_ROCBLAS_VERSION}/${NEW_ROCBLAS_VERSION}/g" CMakeLists.txt
 sed -i "s/${OLD_TENSILE_VERSION}/${NEW_TENSILE_VERSION}/g" CMakeLists.txt
+sed -i "s/${OLD_TENSILE_HASH}/${NEW_TENSILE_HASH}/g" tensile_tag.txt
 

bump_master_version.sh

@@ -6,20 +6,24 @@
 # - after running this script and merging develop into master, run bump_develop_version.sh in master and
 #   merge master into develop
 
-OLD_ROCBLAS_VERSION="2.7.0"
-NEW_ROCBLAS_VERSION="2.8.0"
+OLD_ROCBLAS_VERSION="2.9.0"
+NEW_ROCBLAS_VERSION="2.10.0"
 
-OLD_TENSILE_VERSION="tensile_tag 9c63a0bf1c0acdb44376ddc80b867beb3386981a"
-NEW_TENSILE_VERSION="tensile_tag 015477ad8c8ea0ef6f59b8d49a65015b46b8a48e"
+OLD_TENSILE_VERSION="Tensile 4.12.0"
+NEW_TENSILE_VERSION="Tensile 4.13.0"
+
+OLD_TENSILE_HASH="3ab0890743db4cca5244d0dab30a574fa34c89b8"
+NEW_TENSILE_HASH="fe4f721886d07eef6251cea4225e027181022aa5"
 
 OLD_MINIMUM_REQUIRED_VERSION="MinimumRequiredVersion: 4.6.0"
 NEW_MINIMUM_REQUIRED_VERSION="MinimumRequiredVersion: 4.7.1"
 
 sed -i "s/${OLD_ROCBLAS_VERSION}/${NEW_ROCBLAS_VERSION}/g" CMakeLists.txt
 sed -i "s/${OLD_TENSILE_VERSION}/${NEW_TENSILE_VERSION}/g" CMakeLists.txt
+sed -i "s/${OLD_TENSILE_HASH}/${NEW_TENSILE_HASH}/g" tensile_tag.txt
 
 #only update yaml files for a Tensile major version change
 #for FILE in library/src/blas3/Tensile/Logic/*/*yaml
 #do
 #  sed -i "s/${OLD_MINIMUM_REQUIRED_VERSION}/${NEW_MINIMUM_REQUIRED_VERSION}/" $FILE
-#done
+#a9379f4e42efb754c9a618047bfbf292d74dfd0fdone

clients/benchmarks/client.cpp

@@ -21,9 +21,17 @@
 #include "testing_nrm2_batched.hpp"
 #include "testing_nrm2_strided_batched.hpp"
 #include "testing_rot.hpp"
+#include "testing_rot_batched.hpp"
+#include "testing_rot_strided_batched.hpp"
 #include "testing_rotg.hpp"
+#include "testing_rotg_batched.hpp"
+#include "testing_rotg_strided_batched.hpp"
 #include "testing_rotm.hpp"
+#include "testing_rotm_batched.hpp"
+#include "testing_rotm_strided_batched.hpp"
 #include "testing_rotmg.hpp"
+#include "testing_rotmg_batched.hpp"
+#include "testing_rotmg_strided_batched.hpp"
 #include "testing_scal.hpp"
 #include "testing_scal_batched.hpp"
 #include "testing_scal_strided_batched.hpp"
@@ -174,14 +182,18 @@ struct perf_blas<
             testing_set_get_vector<T>(arg);
         else if(!strcmp(arg.function, "set_get_matrix"))
             testing_set_get_matrix<T>(arg);
-        else if(!strcmp(arg.function, "rot"))
-            testing_rot<T>(arg);
-        else if(!strcmp(arg.function, "rotg"))
-            testing_rotg<T>(arg);
         else if(!strcmp(arg.function, "rotm"))
             testing_rotm<T>(arg);
+        else if(!strcmp(arg.function, "rotm_batched"))
+            testing_rotm_batched<T>(arg);
+        else if(!strcmp(arg.function, "rotm_strided_batched"))
+            testing_rotm_strided_batched<T>(arg);
         else if(!strcmp(arg.function, "rotmg"))
             testing_rotmg<T>(arg);
+        else if(!strcmp(arg.function, "rotmg_batched"))
+            testing_rotmg_batched<T>(arg);
+        else if(!strcmp(arg.function, "rotmg_strided_batched"))
+            testing_rotmg_strided_batched<T>(arg);
         else if(!strcmp(arg.function, "gemv"))
             testing_gemv<T>(arg);
         else if(!strcmp(arg.function, "gemv_batched"))
@@ -326,6 +338,47 @@ struct perf_blas<T,
     }
 };
 
+template <typename Ti, typename To = Ti, typename Tc = To, typename = void>
+struct perf_blas_rot : rocblas_test_invalid
+{
+};
+
+template <typename Ti, typename To, typename Tc>
+struct perf_blas_rot<
+    Ti,
+    To,
+    Tc,
+    typename std::enable_if<(
+        (std::is_same<Ti, float>{} && std::is_same<Ti, To>{} && std::is_same<To, Tc>{})
+        || (std::is_same<Ti, double>{} && std::is_same<Ti, To>{} && std::is_same<To, Tc>{})
+        || (std::is_same<Ti, rocblas_float_complex>{} && std::is_same<To, float>{}
+            && std::is_same<Tc, rocblas_float_complex>{})
+        || (std::is_same<Ti, rocblas_float_complex>{} && std::is_same<To, float>{}
+            && std::is_same<Tc, float>{})
+        || (std::is_same<Ti, rocblas_double_complex>{} && std::is_same<To, double>{}
+            && std::is_same<Tc, rocblas_double_complex>{})
+        || (std::is_same<Ti, rocblas_double_complex>{} && std::is_same<To, double>{}
+            && std::is_same<Tc, double>{}))>::type>
+{
+    explicit operator bool()
+    {
+        return true;
+    }
+
+    void operator()(const Arguments& arg)
+    {
+        if(!strcmp(arg.function, "rot"))
+            testing_rot<Ti, To, Tc>(arg);
+        else if(!strcmp(arg.function, "rot_batched"))
+            testing_rot_batched<Ti, To, Tc>(arg);
+        else if(!strcmp(arg.function, "rot_strided_batched"))
+            testing_rot_strided_batched<Ti, To, Tc>(arg);
+        else
+            throw std::invalid_argument("Invalid combination --function "s + arg.function
+                                        + " --a_type "s + rocblas_datatype2string(arg.a_type));
+    }
+};
+
 template <typename Ta, typename Tb = Ta, typename = void>
 struct perf_blas_scal : rocblas_test_invalid
 {
@@ -361,6 +414,40 @@ struct perf_blas_scal<
     }
 };
 
+template <typename Ta, typename Tb = Ta, typename = void>
+struct perf_blas_rotg : rocblas_test_invalid
+{
+};
+
+template <typename Ta, typename Tb>
+struct perf_blas_rotg<
+    Ta,
+    Tb,
+    typename std::enable_if<
+        (std::is_same<Ta, rocblas_double_complex>{} && std::is_same<Tb, double>{})
+        || (std::is_same<Ta, rocblas_float_complex>{} && std::is_same<Tb, float>{})
+        || (std::is_same<Ta, Tb>{} && std::is_same<Ta, float>{})
+        || (std::is_same<Ta, Tb>{} && std::is_same<Ta, double>{})>::type>
+{
+    explicit operator bool()
+    {
+        return true;
+    }
+    void operator()(const Arguments& arg)
+    {
+        if(!strcmp(arg.function, "rotg"))
+            testing_rotg<Ta, Tb>(arg);
+        else if(!strcmp(arg.function, "rotg_batched"))
+            testing_rotg_batched<Ta, Tb>(arg);
+        else if(!strcmp(arg.function, "rotg_strided_batched"))
+            testing_rotg_strided_batched<Ta, Tb>(arg);
+        else
+            throw std::invalid_argument("Invalid combination --function "s + arg.function
+                                        + " --a_type " + rocblas_datatype2string(arg.a_type)
+                                        + " --b_type " + rocblas_datatype2string(arg.b_type));
+    }
+};
+
 int run_bench_test(Arguments& arg)
 {
     // disable unit_check in client benchmark, it is only used in gtest unit test
@@ -523,6 +610,12 @@ int run_bench_test(Arguments& arg)
         if(!strcmp(function, "scal") || !strcmp(function, "scal_batched")
            || !strcmp(function, "scal_strided_batched"))
             rocblas_blas1_dispatch<perf_blas_scal>(arg);
+        else if(!strcmp(function, "rotg") || !strcmp(function, "rotg_batched")
+                || !strcmp(function, "rotg_strided_batched"))
+            rocblas_blas1_dispatch<perf_blas_rotg>(arg);
+        else if(!strcmp(function, "rot") || !strcmp(function, "rot_batched")
+                || !strcmp(function, "rot_strided_batched"))
+            rocblas_blas1_dispatch<perf_blas_rot>(arg);
         else
             rocblas_simple_dispatch<perf_blas>(arg);
     }

clients/common/rocblas_gentest.py

@@ -199,21 +199,53 @@ def setdefaults(test):
     if test['function'] in ('asum_strided_batched', 'nrm2_strided_batched',
                             'scal_strided_batched', 'swap_strided_batched',
                             'copy_strided_batched', 'dot_strided_batched',
-                            'dotc_strided_batched'):
+                            'dotc_strided_batched', 'rot_strided_batched',
+                            'rotm_strided_batched'):
         if all([x in test for x in ('N', 'incx', 'stride_scale')]):
             test.setdefault('stride_x', int(test['N'] * abs(test['incx']) *
                                             test['stride_scale']))
         if all([x in test for x in ('N', 'incy', 'stride_scale')]):
             test.setdefault('stride_y', int(test['N'] * abs(test['incy']) *
                                             test['stride_scale']))
-
-    if test['function'] in ('ger_strided_batched'):
-        if all([x in test for x in ('M', 'incx', 'stride_scale')]):
-            test.setdefault('stride_x', int(test['M'] * abs(test['incx']) *
-                                            test['stride_scale']))
-        if all([x in test for x in ('N', 'incy', 'stride_scale')]):
-            test.setdefault('stride_y', int(test['N'] * abs(test['incy']) *
-                                            test['stride_scale']))
+        # we are using stride_c for param in rotm
+        if all([x in test for x in ('stride_scale')]):
+            test.setdefault('stride_c', int(test['stride_scale']) * 5)
+
+    if test['function'] in ('gemv_strided_batched', 'ger_strided_batched'):
+        if test['function'] in ('ger_strided_batched') or test['transA'] in ('T', 'C'):
+            if all([x in test for x in ('M', 'incx', 'stride_scale')]):
+                test.setdefault('stride_x', int(test['M'] * abs(test['incx']) *
+                                                test['stride_scale']))
+            if all([x in test for x in ('N', 'incy', 'stride_scale')]):
+                test.setdefault('stride_y', int(test['N'] * abs(test['incy']) *
+                                                test['stride_scale']))
+        else:
+            if all([x in test for x in ('N', 'incx', 'stride_scale')]):
+                test.setdefault('stride_x', int(test['N'] * abs(test['incx']) *
+                                                test['stride_scale']))
+            if all([x in test for x in ('M', 'incy', 'stride_scale')]):
+                test.setdefault('stride_y', int(test['M'] * abs(test['incy']) *
+                                                test['stride_scale']))
+
+    # we are using stride_c for arg c and stride_d for arg s in rotg
+    # these are are single values for each batch
+    if test['function'] in ('rotg_strided_batched'):
+        if 'stride_scale' in test:
+            test.setdefault('stride_a', int(test['stride_scale']))
+            test.setdefault('stride_b', int(test['stride_scale']))
+            test.setdefault('stride_c', int(test['stride_scale']))
+            test.setdefault('stride_d', int(test['stride_scale']))
+
+    # we are using stride_a for d1, stride_b for d2, and stride_c for param in
+    # rotmg. These are are single values for each batch, except param which is
+    # a 5 element array
+    if test['function'] in ('rotmg_strided_batched'):
+        if 'stride_scale' in test:
+            test.setdefault('stride_a', int(test['stride_scale']))
+            test.setdefault('stride_b', int(test['stride_scale']))
+            test.setdefault('stride_c', int(test['stride_scale']) * 5)
+            test.setdefault('stride_x', int(test['stride_scale']))
+            test.setdefault('stride_y', int(test['stride_scale']))
 
     test.setdefault('stride_x', 0)
     test.setdefault('stride_y', 0)

clients/gtest/CMakeLists.txt

@@ -179,7 +179,7 @@ endif( )
 set( ROCBLAS_TEST_DATA "${PROJECT_BINARY_DIR}/staging/rocblas_gtest.data")
 add_custom_command( OUTPUT "${ROCBLAS_TEST_DATA}"
                     COMMAND ../common/rocblas_gentest.py -I ../include rocblas_gtest.yaml -o "${ROCBLAS_TEST_DATA}"
-                    DEPENDS ../common/rocblas_gentest.py rocblas_gtest.yaml ../include/rocblas_common.yaml known_bugs.yaml blas1_gtest.yaml gemm_gtest.yaml gemm_batched_gtest.yaml gemm_strided_batched_gtest.yaml gemv_gtest.yaml gemv_batched_gtest.yaml gemv_strided_batched_gtest.yaml symv_gtest.yaml syr_gtest.yaml ger_gtest.yaml trsm_gtest.yaml trtri_gtest.yaml geam_gtest.yaml set_get_vector_gtest.yaml set_get_matrix_gtest.yaml trsv_gtest.yaml logging_mode_gtest.yaml set_get_pointer_mode_gtest.yaml
+                    DEPENDS ../common/rocblas_gentest.py rocblas_gtest.yaml ../include/rocblas_common.yaml known_bugs.yaml blas1_gtest.yaml gemm_gtest.yaml gemm_batched_gtest.yaml gemm_strided_batched_gtest.yaml gemv_gtest.yaml symv_gtest.yaml syr_gtest.yaml ger_gtest.yaml trsm_gtest.yaml trtri_gtest.yaml geam_gtest.yaml set_get_vector_gtest.yaml set_get_matrix_gtest.yaml trsv_gtest.yaml logging_mode_gtest.yaml set_get_pointer_mode_gtest.yaml
                     WORKING_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}" )
 add_custom_target( rocblas-test-data
                    DEPENDS "${ROCBLAS_TEST_DATA}" )