rs_free_function

CMakeLists.txt

@@ -161,6 +161,7 @@ include (pythonutils)
 include (externalpackages)
 include (flexbison)
 include (cuda_macros)
+include (llvm_macros)
 
 # Include all our testing apparatus and utils, but not if it's a subproject
 if (${PROJECT_NAME}_BUILD_TESTS AND PROJECT_IS_TOP_LEVEL)

src/cmake/compiler.cmake

@@ -468,13 +468,16 @@ if (PROJECT_IS_TOP_LEVEL)
                                 "src/liboslcomp/*.cpp"
                                 "src/liboslcomp/*.h"
                                 "src/liboslexec/batched_rendservices.cpp"
+                                "src/liboslexec/rs_fallback.cpp"
                                 "src/liboslquery/*.cpp"
                                 "src/liboslquery/*.h"
                                 "src/oslinfo/*.cpp"
                                 "src/oslc/*.cpp"
                                 "src/osl.imageio/*.cpp"
                                 "src/osltoy/*.cpp"
                                 "src/osltoy/*.h"
+                                "src/testshade/render_state.h"
+                                "src/testshade/rs_simplerend.cpp"
                                 "testsuite/*.cpp"
         CACHE STRING "Glob patterns to include for clang-format")
     set (CLANG_FORMAT_EXCLUDES

src/cmake/cuda_macros.cmake

@@ -130,8 +130,8 @@ function ( LLVM_COMPILE_CUDA llvm_src headers prefix llvm_bc_cpp_generated extra
     MAKE_CUDA_BITCODE (${llvm_src} "" llvm_bc "${extra_clang_args}")
 
     add_custom_command (OUTPUT ${llvm_bc_cpp}
-        COMMAND python "${CMAKE_SOURCE_DIR}/src/liboslexec/serialize-bc.py" ${llvm_bc} ${llvm_bc_cpp} ${prefix}
+        COMMAND python "${CMAKE_SOURCE_DIR}/src/build-scripts/serialize-bc.py" ${llvm_bc} ${llvm_bc_cpp} ${prefix}
         MAIN_DEPENDENCY ${llvm_src}
-        DEPENDS "${CMAKE_SOURCE_DIR}/src/liboslexec/serialize-bc.py" ${llvm_src} ${headers} ${llvm_bc}
+        DEPENDS "${CMAKE_SOURCE_DIR}/src/build-scripts/serialize-bc.py" ${llvm_src} ${headers} ${llvm_bc}
         WORKING_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}" )
 endfunction ()

src/cmake/llvm_macros.cmake

@@ -0,0 +1,128 @@
+# Copyright Contributors to the Open Shading Language project.
+# SPDX-License-Identifier: BSD-3-Clause
+# https://github.com/AcademySoftwareFoundation/OpenShadingLanguage
+
+
+function ( EMBED_LLVM_BITCODE_IN_CPP src_list suffix output_name list_to_append_cpp extra_clang_args)
+
+    if (VERBOSE)
+        message (STATUS "EMBED_LLVM_BITCODE_IN_CPP src_list=${src_list}")
+    endif ()
+
+    foreach ( src ${src_list} )
+        get_filename_component ( src_we ${src} NAME_WE )
+        set ( src_asm "${CMAKE_CURRENT_BINARY_DIR}/${src_we}${suffix}.s" )
+        set ( src_bc "${CMAKE_CURRENT_BINARY_DIR}/${src_we}${suffix}.bc" )
+        if (VERBOSE)
+            message (STATUS "EMBED_LLVM_BITCODE_IN_CPP in=${src}")
+            message (STATUS "EMBED_LLVM_BITCODE_IN_CPP asm=${src_asm}")
+            message (STATUS "EMBED_LLVM_BITCODE_IN_CPP bc=${src_bc}")
+        endif ()
+        list ( APPEND src_bc_list ${src_bc} )
+
+        get_property (CURRENT_DEFINITIONS DIRECTORY PROPERTY COMPILE_DEFINITIONS)
+        if (VERBOSE)
+            message (STATUS "Current #defines are ${CURRENT_DEFINITIONS}")
+        endif ()
+        foreach (def ${CURRENT_DEFINITIONS})
+            set (LLVM_COMPILE_FLAGS ${LLVM_COMPILE_FLAGS} "-D${def}")
+        endforeach()
+        set (LLVM_COMPILE_FLAGS ${LLVM_COMPILE_FLAGS} ${SIMD_COMPILE_FLAGS} ${CSTD_FLAGS} ${TOOLCHAIN_FLAGS})
+        # Avoid generating __dso_handle external global
+        set (LLVM_COMPILE_FLAGS ${LLVM_COMPILE_FLAGS} "-fno-use-cxa-atexit")
+
+        # Figure out what program we will use to make the bitcode.
+        if (NOT LLVM_BC_GENERATOR)
+            find_program (LLVM_BC_GENERATOR NAMES "clang++"
+                        PATHS "${LLVM_DIRECTORY}/bin"
+                        NO_CMAKE_PATH NO_DEFAULT_PATH NO_CMAKE_SYSTEM_PATH
+                        NO_SYSTEM_ENVIRONMENT_PATH NO_CMAKE_ENVIRONMENT_PATH)
+        endif ()
+        # If that didn't work, look anywhere
+        if (NOT LLVM_BC_GENERATOR)
+            # Wasn't in their build, look anywhere
+            find_program (LLVM_BC_GENERATOR NAMES clang++ llvm-g++)
+        endif ()
+
+        if (NOT LLVM_BC_GENERATOR)
+            message (FATAL_ERROR "You must have a valid llvm bitcode generator (clang++) somewhere.")
+        endif ()
+        if (VERBOSE)
+            message (STATUS "Using LLVM_BC_GENERATOR ${LLVM_BC_GENERATOR} to generate bitcode.")
+        endif()
+
+        # Fix specific problem I had on new Apple systems (e.g. Mavericks) with
+        # LLVM/libc++ installed -- for some reason, LLVM 3.4 wasn't finding it,
+        # so in that specific case, append another -I to point it in the right
+        # direction.
+        #if (APPLE AND ${LLVM_BC_GENERATOR} MATCHES ".*clang.*")
+        #    exec_program ( "${LLVM_BC_GENERATOR}" ARGS --version OUTPUT_VARIABLE MY_CLANG_VERSION )
+        #    string (REGEX REPLACE "clang version ([0-9][.][0-9]+).*" "\\1" MY_CLANG_VERSION "${MY_CLANG_VERSION}")
+        #    if ((${MY_CLANG_VERSION} VERSION_GREATER "3.3")
+        #          AND (EXISTS "/usr/lib/libc++.dylib")
+        #          AND (EXISTS "/Library/Developer/CommandLineTools/usr/lib/c++/v1"))
+        #        set (LLVM_COMPILE_FLAGS ${LLVM_COMPILE_FLAGS} "-I/Library/Developer/CommandLineTools/usr/lib/c++/v1")
+        #    endif ()
+        #endif ()
+
+        list (TRANSFORM IMATH_INCLUDES PREPEND -I
+            OUTPUT_VARIABLE ALL_IMATH_INCLUDES)
+        list (TRANSFORM OPENEXR_INCLUDES PREPEND -I
+            OUTPUT_VARIABLE ALL_OPENEXR_INCLUDES)
+        list (TRANSFORM OpenImageIO_INCLUDES PREPEND -I
+            OUTPUT_VARIABLE ALL_OpenImageIO_INCLUDES)
+
+
+        # Command to turn the .cpp file into LLVM assembly language .s, into
+        # LLVM bitcode .bc, then back into a C++ file with the bc embedded!
+        add_custom_command ( OUTPUT ${src_bc}
+        COMMAND ${LLVM_BC_GENERATOR}
+            ${LLVM_COMPILE_FLAGS}
+            "-I${CMAKE_CURRENT_SOURCE_DIR}"
+            "-I${CMAKE_SOURCE_DIR}/src/include"
+            "-I${CMAKE_BINARY_DIR}/include"
+            ${ALL_OpenImageIO_INCLUDES}
+            ${ALL_IMATH_INCLUDES}
+            #"-isystem ${Boost_INCLUDE_DIRS}" #Does not pick up usr installed boost/thread/tss.hpp for oslexec_pvt.h
+            "-I${Boost_INCLUDE_DIRS}"
+            -DOSL_COMPILING_TO_BITCODE=1
+            -Wno-deprecated-register
+            # the following 2 warnings can be restored when all 3rd parties have fixed their export macros
+            # (dllimport attribute is not supported when compiling for Cuda and triggers a ton of warnings)
+            -Wno-ignored-attributes -Wno-unknown-attributes
+            -O3 -fno-math-errno -S -emit-llvm ${extra_clang_args}
+            -o ${src_asm} ${src}
+        COMMAND "${LLVM_DIRECTORY}/bin/llvm-as" -f -o ${src_bc} ${src_asm}
+        # Do NOT setup a MAIN_DEPENDENCY because only 1 may exist 
+        # and we may have the several outputs dependant on the same source 
+        DEPENDS ${src} ${exec_headers} ${PROJECT_PUBLIC_HEADERS}
+        WORKING_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}" )
+    endforeach ()
+
+
+    if (VERBOSE)
+        message ( STATUS "^^^^^^^^^^^^^^^^^^^^^^^^^^" )
+        message ( STATUS "src_bc_list: ${src_bc_list} ")
+        message ( STATUS "^^^^^^^^^^^^^^^^^^^^^^^^^^" )
+    endif()
+
+    # Link all of the individual LLVM bitcode files
+    set ( linked_src_bc "${CMAKE_CURRENT_BINARY_DIR}/${output_name}.bc" )
+    add_custom_command ( OUTPUT ${linked_src_bc}
+        COMMAND "${LLVM_DIRECTORY}/bin/llvm-link" -internalize ${src_bc_list} -o ${linked_src_bc}
+        DEPENDS ${src_bc_list} ${exec_headers} ${PROJECT_PUBLIC_HEADERS} ${src_list}
+        WORKING_DIRECTORY "${CMAKE_CURRENT_BINARY_DIR}" )
+
+     # Serialize the linked bitcode into a CPP file 
+    set ( src_bc_cpp "${CMAKE_CURRENT_BINARY_DIR}/${output_name}.bc.cpp" )
+    add_custom_command ( OUTPUT ${src_bc_cpp}
+        COMMAND python "${CMAKE_SOURCE_DIR}/src/build-scripts/serialize-bc.py"
+            ${linked_src_bc} ${src_bc_cpp} ${output_name}
+        DEPENDS "${CMAKE_SOURCE_DIR}/src/build-scripts/serialize-bc.py" ${linked_src_bc}
+        ${exec_headers} ${PROJECT_PUBLIC_HEADERS}
+        WORKING_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}" )
+
+    # add generated .cpp with embedded bitcode to the list of soures
+    set ( ${list_to_append_cpp} ${${list_to_append_cpp}} ${src_bc_cpp} PARENT_SCOPE )
+
+endfunction ( )

src/cmake/testing.cmake

@@ -90,6 +90,7 @@ macro ( TESTSUITE )
     endif ()
     set (test_all_optix $ENV{TESTSUITE_OPTIX})
     set (test_all_batched $ENV{TESTSUITE_BATCHED})
+    set (test_all_rs_bitcode $ENV{TESTSUITE_RS_BITCODE})
     # Add the tests if all is well.
     set (ALL_TEST_LIST "")
     set (_testsuite "${CMAKE_SOURCE_DIR}/testsuite")
@@ -180,6 +181,25 @@ macro ( TESTSUITE )
                                    ENV TESTSHADE_OPT=0 OSL_REGRESSION_TEST=BATCHED )
             endif ()
         endif ()
+
+        # if there is an RS_BITCODE marker file in the directory.
+        if ((EXISTS "${_testsrcdir}/RS_BITCODE" OR test_all_rs_bitcode)
+            AND NOT EXISTS "${_testsrcdir}/BATCHED_REGRESSION"
+            AND NOT EXISTS "${_testsrcdir}/RS_BITCODE_REGRESSION"
+            AND NOT EXISTS "${_testsrcdir}/NOOPTIMIZE")
+            add_one_testsuite ("${_testname}.rsbitcode.opt" "${_testsrcdir}"
+                                ENV TESTSHADE_OPT=2 TESTSHADE_RS_BITCODE=1 )
+        endif ()
+
+        # if there is an RS_BITCODE_REGRESSION marker file in the directory.
+        if (EXISTS "${_testsrcdir}/RS_BITCODE_REGRESSION" OR (
+                test_all_rs_bitcode AND EXISTS "${_testsrcdir}/BATCHED_REGRESSION")
+            AND NOT EXISTS "${_testsrcdir}/NOOPTIMIZE")
+            # optimized for right now
+            add_one_testsuite ("${_testname}.regress.rsbitcode.opt" "${_testsrcdir}"
+                                ENV TESTSHADE_OPT=2 OSL_REGRESSION_TEST=RS_BITCODE )
+        endif ()
+
     endforeach ()
     if (VERBOSE)
         message (STATUS "Added tests: ${ALL_TEST_LIST}")
@@ -348,5 +368,6 @@ macro (osl_add_all_tests)
         set_tests_properties (arithmetic-reg.regress.batched.opt PROPERTIES TIMEOUT 800)
         set_tests_properties (transform-reg.regress.batched.opt PROPERTIES TIMEOUT 800)
     endif ()
+    set_tests_properties (matrix-reg.regress.rsbitcode.opt PROPERTIES TIMEOUT 800)
 
 endmacro()

src/include/OSL/device_string.h

@@ -132,7 +132,11 @@ namespace DeviceStrings {
 #undef STRDECL
 }
 #else
-#  define STRING_PARAMS(x)  StringParams::x
+#  ifdef OSL_HOST_RS_BITCODE
+#    define STRING_PARAMS(x)  RS_##x
+#  else
+#    define STRING_PARAMS(x)  StringParams::x
+#  endif
 #endif
 
 
@@ -146,3 +150,5 @@ namespace StringParams = OSL_NAMESPACE::DeviceStrings;
 #else
 namespace StringParams = OSL_NAMESPACE::Strings;
 #endif
+
+

src/include/OSL/llvm_util.h

@@ -228,6 +228,15 @@ class OSLEXECPUBLIC LLVM_Util {
     // Name for this TargetISA enum.
     static const char* target_isa_name(TargetISA isa);
 
+    /// Add a global mapping of a variable to its address
+    /// explicitly instead of relying on dlsym.
+    static void add_global_mapping (const char *global_var_name, void *global_var_addr);
+
+    /// Check all external global variables to ensure they have 
+    /// been mapped to an address, a vector is populated with
+    /// the names that are unmapped.
+    void validate_global_mappings (std::vector<std::string> &names_of_unmapped_globals);
+
     // For CPU compilation, inventory the host CPU capabilities. You can
     // optionally request a specific ISA by name. If `no_fma` is true,
     // specifically pretend there is no FMA capability, even if the hardware
@@ -908,6 +917,11 @@ class OSLEXECPUBLIC LLVM_Util {
     /// file.  If err is not NULL, errors will be deposited there.
     void write_bitcode_file (const char *filename, std::string *err=NULL);
 
+    /// Insert bitcode from another module into the current Module
+    /// returns true if successful, false otherwise. 
+    /// Ownership of the other_module is transferred.
+    bool absorb_module (std::unique_ptr<llvm::Module> other_module );
+
     /// Generate PTX for the current Module and return it as a string
     bool ptx_compile_group (llvm::Module* lib_module, const std::string& name,
                             std::string& out);

src/include/OSL/oslexec.h

@@ -29,6 +29,11 @@ template<typename DataT, int WidthT> struct Wide;
 #endif
 
 
+/// RendererServices free function bitcode may reference global variables that
+/// only exist in the renderer's program space and won't automatically be 
+/// found when referenced in JIT'd code inside OSL.  A renderer may register
+/// the addresses of these global variables to allow valid JIT to occur.
+void register_JIT_Global(const char* global_var_name, void* global_var_addr);
 
 /// Opaque pointer to whatever the renderer uses to represent a
 /// (potentially motion-blurred) coordinate transformation.
@@ -45,9 +50,7 @@ class ShadingSystemImpl;
 }
 
 #if defined(__CUDA_ARCH__) && OPTIX_VERSION >= 70000
-#  define STRINGIFY(x) XSTR(x)
-#  define XSTR(x) #x
-#  define STRING_PARAMS(x)  UStringHash::Hash(STRINGIFY(x))
+#  define STRING_PARAMS(x)  UStringHash::Hash(__OSL_STRINGIFY(x))
 #else
 #  define STRING_PARAMS(x)  StringParams::x
 #endif

src/include/OSL/oslversion.h.in

@@ -103,4 +103,7 @@ namespace @PROJ_NAME@ = @PROJ_NAMESPACE_V@;
     #define __OSL_WIDE_PVT __OSL_CONCAT5(b,__OSL_WIDTH,_,__OSL_TARGET_ISA,_pvt)
 #endif
 
+#define __OSL_INDIRECT_STRINGIFY(x) #x
+#define __OSL_STRINGIFY(x) __OSL_INDIRECT_STRINGIFY(x)
+
 #endif /* OSLVERSION_H */

src/include/OSL/rendererservices.h

@@ -42,6 +42,7 @@ OSL_GCC_PRAGMA(GCC diagnostic ignored "-Wunused-parameter")
 /// RendererServices defines an abstract interface through which a
 /// renderer may provide callback to the ShadingSystem.
 class OSLEXECPUBLIC RendererServices {
+    // Keep interface in sync with rs_free_function.h
 public:
     typedef TextureSystem::TextureHandle TextureHandle;
     typedef TextureSystem::Perthread TexturePerthread;