Addressing PR comments

Inverting device check before adding extra compiler flags to increase compatibility. Removing useless flags in test.
codeplaysoftware · Apr 16, 2024 · 5749c0b · 5749c0b
1 parent 33bda6b
commit 5749c0b
Show file tree

Hide file tree

Showing 4 changed files with 8 additions and 16 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -144,9 +144,8 @@ if (INSTALL_HEADER_ONLY)
   set_target_properties(portblas PROPERTIES
     INTERFACE_INCLUDE_DIRECTORIES "${PORTBLAS_INCLUDE};$<INSTALL_INTERFACE:src>"
   )
-  set(non_intel_target "AMD_GPU" "NVIDIA_GPU")
-  if((${CMAKE_CXX_COMPILER_ID} STREQUAL "IntelLLVM") AND
-      (TUNING_TARGET IN_LIST non_intel_target) )
+  if((${CMAKE_CXX_COMPILER_ID} STREQUAL "IntelLLVM") AND NOT
+    (${TUNING_TARGET} STREQUAL "INTEL_GPU") )
       target_compile_options(portblas INTERFACE -fno-fast-math)
       target_compile_options(portblas INTERFACE -mllvm -loopopt=0 )
   endif()

diff --git a/cmake/Modules/FindDPCPP.cmake b/cmake/Modules/FindDPCPP.cmake
@@ -85,7 +85,6 @@ endif()
 add_definitions(-DSB_ENABLE_USM=1)
 set(SB_ENABLE_USM 1)
 list(APPEND DPCPP_FLAGS "-DSB_ENABLE_USM=1")
-set(non_intel_target "AMD_GPU" "NVIDIA_GPU")
 
 function(add_sycl_to_target)
   set(options)
@@ -97,10 +96,11 @@ function(add_sycl_to_target)
     "${multi_value_args}"
     ${ARGN}
   )
-  if((${CMAKE_CXX_COMPILER_ID} STREQUAL "IntelLLVM") AND
-      (TUNING_TARGET IN_LIST non_intel_target) )
+if((${CMAKE_CXX_COMPILER_ID} STREQUAL "IntelLLVM") AND NOT
+    (${TUNING_TARGET} STREQUAL "INTEL_GPU") )
     target_compile_options(${SB_ADD_SYCL_TARGET} PRIVATE -fno-fast-math)
     target_compile_options(${SB_ADD_SYCL_TARGET} PRIVATE -mllvm -loopopt=0 )
+    message(STATUS "Adding -fno-fast-math -mllvm -loopopt=0 to target ${SB_ADD_SYCL_TARGET}")
   endif()
   target_compile_options(${SB_ADD_SYCL_TARGET} PUBLIC ${DPCPP_FLAGS})
   get_target_property(target_type ${SB_ADD_SYCL_TARGET} TYPE)

diff --git a/src/operations/blas1_trees.hpp b/src/operations/blas1_trees.hpp
@@ -448,7 +448,7 @@ PORTBLAS_INLINE bool TupleOp<rhs_t>::valid_thread(
 template <typename rhs_t>
 PORTBLAS_INLINE typename TupleOp<rhs_t>::value_t TupleOp<rhs_t>::eval(
     typename TupleOp<rhs_t>::index_t i) {
-  return TupleOp<rhs_t>::value_t(i, cl::sycl::abs(rhs_.eval(i)));
+  return TupleOp<rhs_t>::value_t(i, cl::sycl::fabs(rhs_.eval(i)));
 }
 
 template <typename rhs_t>

diff --git a/test/unittest/CMakeLists.txt b/test/unittest/CMakeLists.txt
@@ -102,12 +102,6 @@ if(is_dpcpp)
 endif()
 
 
-# Contains tests that fail if compiled with -ffast-math
-set(SYCL_UNITTEST_NOFASTMATH
-  ${PORTBLAS_UNITTEST}/blas1/blas1_rotg_test.cpp
-  ${PORTBLAS_UNITTEST}/blas1/blas1_rotmg_test.cpp
-)
-
 if(GEMM_TALL_SKINNY_SUPPORT)
   list(APPEND SYCL_UNITTEST_SRCS ${PORTBLAS_UNITTEST}/blas3/blas3_gemm_tall_skinny_test.cpp)
 endif()
@@ -142,9 +136,8 @@ foreach(blas_test ${SYCL_UNITTEST_SRCS})
   target_link_libraries(${test_exec} PRIVATE gtest_main Clara::Clara blas::blas portblas)
   target_include_directories(${test_exec} PRIVATE ${CBLAS_INCLUDE} ${PORTBLAS_COMMON_INCLUDE_DIR})
 
-  list (FIND SYCL_UNITTEST_NOFASTMATH ${blas_test} _index)
-  if (${_index} GREATER -1 OR (${CMAKE_CXX_COMPILER_ID} STREQUAL "IntelLLVM" AND
-    TUNING_TARGET IN_LIST non_intel_target))
+  if ((${CMAKE_CXX_COMPILER_ID} STREQUAL "IntelLLVM") AND NOT
+    (${TUNING_TARGET} STREQUAL "INTEL_GPU") )
     target_compile_options(${test_exec} PRIVATE "-fno-fast-math")
   endif()