Merge pull request #1961 from IntelPython/use-offload-compression

Use --offload-compress linker option to compress offload sections

Author: oleksandr-pavlyk

CMakeLists.txt

@@ -34,7 +34,16 @@ set(DPCTL_TARGET_HIP
     CACHE STRING
     "Build DPCTL to target a HIP device architecture"
 )
-option(DPCTL_WITH_REDIST "Build DPCTL assuming DPC++ redistributable is installed into Python prefix" OFF)
+option(
+     DPCTL_WITH_REDIST
+     "Build DPCTL assuming DPC++ redistributable is installed into Python prefix"
+     OFF)
+option(
+     DPCTL_OFFLOAD_COMPRESS
+     "Build using offload section compression feature of DPC++ to reduce \
+size of shared object with offloading sections"
+     OFF
+)
 
 find_package(IntelSYCL REQUIRED PATHS ${CMAKE_SOURCE_DIR}/cmake NO_DEFAULT_PATH)
 

dpctl/CMakeLists.txt

@@ -111,6 +111,11 @@ function(build_dpctl_ext _trgt _src _dest)
     Python_add_library(${_trgt} MODULE WITH_SOABI ${_generated_src})
     if (BUILD_DPCTL_EXT_SYCL)
         add_sycl_to_target(TARGET ${_trgt} SOURCES ${_generated_src})
+        target_compile_options(${_trgt} PRIVATE -fno-sycl-id-queries-fit-in-int)
+	target_link_options(${_trgt} PRIVATE -fsycl-device-code-split=per_kernel)
+	if (DPCTL_OFFLOAD_COMPRESS)
+            target_link_options(${_trgt} PRIVATE --offload-compress)
+	endif()
         if(_dpctl_sycl_targets)
             # make fat binary
             target_compile_options(

dpctl/tensor/CMakeLists.txt

@@ -268,6 +268,10 @@ set(_linker_options "LINKER:${DPCTL_LDFLAGS}")
 foreach(python_module_name ${_py_trgts})
     target_compile_options(${python_module_name} PRIVATE -fno-sycl-id-queries-fit-in-int)
     target_link_options(${python_module_name} PRIVATE -fsycl-device-code-split=per_kernel)
+    if (DPCTL_OFFLOAD_COMPRESS)
+        target_link_options(${python_module_name} PRIVATE --offload-compress)
+    endif()
+
     target_include_directories(${python_module_name}
         PRIVATE
         ${CMAKE_CURRENT_SOURCE_DIR}/libtensor/include
@@ -279,7 +283,7 @@ foreach(python_module_name ${_py_trgts})
             target_compile_options(${python_module_name}
                 PRIVATE -fprofile-instr-generate -fcoverage-mapping
             )
-	endif()
+        endif()
         target_link_options(${python_module_name}
             PRIVATE -fprofile-instr-generate -fcoverage-mapping
         )

dpctl/utils/CMakeLists.txt

@@ -29,6 +29,10 @@ set(_linker_options "LINKER:${DPCTL_LDFLAGS}")
 foreach(python_module_name ${_pybind11_targets})
     target_compile_options(${python_module_name} PRIVATE -fno-sycl-id-queries-fit-in-int)
     target_link_options(${python_module_name} PRIVATE -fsycl-device-code-split=per_kernel)
+    if (DPCTL_OFFLOAD_COMPRESS)
+        target_link_options(${python_module_name} PRIVATE --offload-compress)
+    endif()
+
     target_include_directories(${python_module_name}
         PRIVATE
         ${CMAKE_CURRENT_SOURCE_DIR}/libtensor/include
@@ -40,7 +44,7 @@ foreach(python_module_name ${_pybind11_targets})
             target_compile_options(${python_module_name}
                 PRIVATE -fprofile-instr-generate -fcoverage-mapping
             )
-	endif()
+        endif()
         target_link_options(${python_module_name}
             PRIVATE -fprofile-instr-generate -fcoverage-mapping
         )