Changing TUNING_TARGET from DEFAULT_CPU to DEFAULT

This tuning target is capable of compiling for every device if DPCPP_SYCL_TARGET and DPCPP_SYCL_ARCH are specified correctly. So this comment is a first step in changing this target behaviour and increasing general understanding.
codeplaysoftware · Apr 29, 2024 · 192e5e3 · 192e5e3
1 parent 896c9a7
commit 192e5e3
Show file tree

Hide file tree

Showing 9 changed files with 18 additions and 18 deletions.
diff --git a/README.md b/README.md
@@ -487,7 +487,7 @@ Some of the supported options are:
 | `BLAS_ENABLE_TESTING` | `ON`/`OFF` | Set it to `OFF` to avoid building the tests (`ON` is the default value) |
 | `BLAS_ENABLE_BENCHMARK` | `ON`/`OFF` | Set it to `OFF` to avoid building the benchmarks (`ON` is the default value) |
 | `SYCL_COMPILER` | name | Used to determine which SYCL implementation to use. By default, the first implementation found is used. Supported values are: `dpcpp`, `adaptivecpp` and `computecpp`*(deprecated)*. |
-| `TUNING_TARGET` | name | By default, this flag is set to `DEFAULT_CPU` to restrict any device specific compiler optimizations. Use this flag to tune the code for a target (**highly recommended** for performance). The supported targets are: `INTEL_GPU`, `NVIDIA_GPU`, `AMD_GPU` |
+| `TUNING_TARGET` | name | By default, this flag is set to `DEFAULT` to restrict any device specific compiler optimizations. Use this flag to tune the code for a target (**highly recommended** for performance). The supported targets are: `INTEL_GPU`, `NVIDIA_GPU`, `AMD_GPU` |
 | `CMAKE_PREFIX_PATH` | path | List of paths to check when searching for dependencies |
 | `CMAKE_INSTALL_PREFIX` | path | Specify the install location, used when invoking `ninja install` |
 | `BUILD_SHARED_LIBS` | `ON`/`OFF` | Build as shared library (`ON` by default) |

diff --git a/cmake/CmakeFunctionHelper.cmake b/cmake/CmakeFunctionHelper.cmake
@@ -98,11 +98,11 @@ function(set_target_compile_def in_target)
   elseif(${TUNING_TARGET} STREQUAL "NVIDIA_GPU")
     target_compile_definitions(${in_target} PUBLIC NVIDIA_GPU=1)
   else()
-    if(NOT ${TUNING_TARGET} STREQUAL "DEFAULT_CPU")
-      message(STATUS "${TUNING_TARGET} not supported. Switching to DEFAULT_CPU instead.")
-      set(TUNING_TARGET "DEFAULT_CPU")
+    if(NOT ${TUNING_TARGET} STREQUAL "DEFAULT")
+      message(STATUS "${TUNING_TARGET} not supported. Switching to DEFAULT instead.")
+      set(TUNING_TARGET "DEFAULT")
     endif()
-    target_compile_definitions(${in_target} PUBLIC DEFAULT_CPU=1)
+    target_compile_definitions(${in_target} PUBLIC DEFAULT=1)
   endif()
   message(STATUS "Adding ${TUNING_TARGET} backend to target ${in_target}")
   #setting tall skinny support

diff --git a/cmake/Modules/ConfigurePORTBLAS.cmake b/cmake/Modules/ConfigurePORTBLAS.cmake
@@ -56,7 +56,7 @@ if(NAIVE_GEMM)
 endif()
 
 # the TUNING_TARGET variable defines the platform for which the sycl library is tuned
-SET(TUNING_TARGET "DEFAULT_CPU" CACHE STRING "Default Platform 'DEFAULT_CPU'")
+SET(TUNING_TARGET "DEFAULT" CACHE STRING "Default Platform 'DEFAULT'")
 message(STATUS "${TUNING_TARGET} is chosen as a tuning target")
 
 if(DEFINED TARGET)

diff --git a/cmake/Modules/SYCL.cmake b/cmake/Modules/SYCL.cmake
@@ -97,7 +97,7 @@ elseif(is_adaptivecpp)
   set(CMAKE_CXX_STANDARD 17)
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O3")
   get_target_property(SYCL_INCLUDE_DIRS AdaptiveCpp::acpp-rt INTERFACE_INCLUDE_DIRECTORIES)
-  set(HIP_BENCH_UNSUPPORTED_TARGETS "INTEL_GPU" "DEFAULT_CPU")
+  set(HIP_BENCH_UNSUPPORTED_TARGETS "INTEL_GPU" "DEFAULT")
   if((${BLAS_ENABLE_BENCHMARK}) AND (${TUNING_TARGET} IN_LIST HIP_BENCH_UNSUPPORTED_TARGETS))
     message(STATUS "Benchmarks are not supported when targetting OpenCL/LevelZero backend 
             devices. portBLAS Benchmarks are disabled.")

diff --git a/src/interface/blas1/backend/default_cpu.hpp b/src/interface/blas1/backend/default_cpu.hpp
@@ -22,8 +22,8 @@
  *  @filename defaul_cpu.hpp
  *
  **************************************************************************/
-#ifndef PORTBLAS_ASUM_DEFAULT_CPU_BACKEND_HPP
-#define PORTBLAS_ASUM_DEFAULT_CPU_BACKEND_HPP
+#ifndef PORTBLAS_ASUM_DEFAULT_BACKEND_HPP
+#define PORTBLAS_ASUM_DEFAULT_BACKEND_HPP
 #include "interface/blas1_interface.h"
 
 namespace blas {

diff --git a/src/interface/blas2/backend/default_cpu.hpp b/src/interface/blas2/backend/default_cpu.hpp
@@ -22,8 +22,8 @@
  *  @filename default_cpu.hpp
  *
  **************************************************************************/
-#ifndef PORTBLAS_GEMV_DEFAULT_CPU_BACKEND_HPP
-#define PORTBLAS_GEMV_DEFAULT_CPU_BACKEND_HPP
+#ifndef PORTBLAS_GEMV_DEFAULT_BACKEND_HPP
+#define PORTBLAS_GEMV_DEFAULT_BACKEND_HPP
 #include "interface/blas2_interface.h"
 
 namespace blas {

diff --git a/src/interface/blas3/backend/default_cpu.hpp b/src/interface/blas3/backend/default_cpu.hpp
@@ -22,8 +22,8 @@
  *  @filename default_cpu.hpp
  *
  **************************************************************************/
-#ifndef PORTBLAS_GEMM_DEFAULT_CPU_BACKEND_HPP
-#define PORTBLAS_GEMM_DEFAULT_CPU_BACKEND_HPP
+#ifndef PORTBLAS_GEMM_DEFAULT_BACKEND_HPP
+#define PORTBLAS_GEMM_DEFAULT_BACKEND_HPP
 #include "interface/gemm_launcher.h"
 
 namespace blas {

diff --git a/src/interface/extension/backend/default_cpu.hpp b/src/interface/extension/backend/default_cpu.hpp
@@ -22,8 +22,8 @@
  *  @filename default_cpu.hpp
  *
  **************************************************************************/
-#ifndef PORTBLAS_TRANSPOSE_DEFAULT_CPU_BACKEND_HPP
-#define PORTBLAS_TRANSPOSE_DEFAULT_CPU_BACKEND_HPP
+#ifndef PORTBLAS_TRANSPOSE_DEFAULT_BACKEND_HPP
+#define PORTBLAS_TRANSPOSE_DEFAULT_BACKEND_HPP
 #include "interface/extension_interface.h"
 
 namespace blas {

diff --git a/test/unittest/CMakeLists.txt b/test/unittest/CMakeLists.txt
@@ -101,7 +101,7 @@ if(is_dpcpp)
   )
 endif()
 
-if(is_dpcpp AND ${TUNING_TARGET} STREQUAL "DEFAULT_CPU")
+if(is_dpcpp AND ${TUNING_TARGET} STREQUAL "DEFAULT")
   if (${DPCPP_SYCL_TARGET} STREQUAL "nvptx64-nvidia-cuda")
     set(DEFAULT_SKIP
       ${PORTBLAS_UNITTEST}/blas1/blas1_iamax_test.cpp
@@ -111,7 +111,7 @@ if(is_dpcpp AND ${TUNING_TARGET} STREQUAL "DEFAULT_CPU")
       ${PORTBLAS_UNITTEST}/blas2/blas2_tpsv_test.cpp
       ${PORTBLAS_UNITTEST}/blas3/blas3_trsm_test.cpp
       )
-    message(WARNING "Targetting NVIDIA hardware with DEFAULT_CPU TUNING_TARGET.
+    message(WARNING "Targetting NVIDIA hardware with DEFAULT TUNING_TARGET.
     Disabling tests for following operators: iamax, iamin, trsv, tbsv, tpsv, trsm.")
   elseif(${DPCPP_SYCL_TARGET} STREQUAL "amdgcn-amd-amdhsa")
     set(DEFAULT_SKIP
@@ -120,7 +120,7 @@ if(is_dpcpp AND ${TUNING_TARGET} STREQUAL "DEFAULT_CPU")
       ${PORTBLAS_UNITTEST}/blas2/blas2_tbsv_test.cpp
       ${PORTBLAS_UNITTEST}/blas2/blas2_tpsv_test.cpp
       )
-    message(WARNING "Targetting AMD hardware with DEFAULT_CPU TUNING_TARGET.
+    message(WARNING "Targetting AMD hardware with DEFAULT TUNING_TARGET.
     Disabling tests for following operators: iamax, iamin, tbsv, tpsv.")
   endif()
 endif()