Solve issues with hip-vdi runtime static lib

1.Combine libamdhip64_static_base.a and libamdvdi_static.a into libamdhip64_static.a.
2.Let hipcc use -use-staticlib to link libamdhip64_static.a.
3.Add some samples for static lib.
4.Fix compiling failure of code object.

Change-Id: Ic8c95228eb139058da8b5d66ba8439486154ca6f


[ROCm/hip commit: da27fd2b09]

projects/hip/bin/hipcc

@@ -211,10 +211,7 @@ if ($HIP_PLATFORM eq "clang") {
     $HIPCXXFLAGS .= " -isystem $HIP_CLANG_INCLUDE_PATH/..";
     $HIPCFLAGS .= " -isystem $HIP_CLANG_INCLUDE_PATH/..";
     $HIPLDFLAGS .= " -L$HIP_LIB_PATH";
-    if (not $isWindows) {
-      $HIPLDFLAGS .= " -Wl,--rpath-link=$HIP_LIB_PATH";
-      $HIPLDFLAGS .= " -lhip_hcc";
-    } else {
+    if ($isWindows) {
       $HIPLDFLAGS .= " -lamdhip64";
     }
     if ($HIP_CLANG_HCC_COMPAT_MODE) {
@@ -480,6 +477,7 @@ foreach $arg (@ARGV)
     {
         $linkType = 0;
         $setLinkType = 1;
+        $swallowArg = 1;
     }
     if(($trimarg eq '-use-sharedlib') and ($setLinkType eq 0))
     {
@@ -770,6 +768,14 @@ if ($HIP_PLATFORM eq "clang") {
     if (not $isWindows) {
         $HIPLDFLAGS .= " -lgcc_s -lgcc -lpthread -lm";
     }
+
+    if (not $isWindows  and not $compileOnly) {
+      if ($linkType eq 0) {
+        $toolArgs .= " -L$HIP_LIB_PATH -lamdhip64_static -L$ROCM_PATH/lib -lhsa-runtime64 -ldl ";
+      } else {
+        $toolArgs .= " -Wl,--enable-new-dtags -Wl,--rpath=$HIP_LIB_PATH:$ROCM_PATH/lib -lhip_hcc ";
+      }
+    }
 }
 
 

projects/hip/samples/0_Intro/bit_extract/Makefile

@@ -13,10 +13,15 @@ ifeq (${HIP_PLATFORM}, nvcc)
 endif
 
 EXE=bit_extract
+EXE_STATIC=bit_extract_static
 
 $(EXE): bit_extract.cpp
 	$(HIPCC) $(HIPCC_FLAGS) $< -o $@
 
+$(EXE_STATIC): bit_extract.cpp
+	$(HIPCC) -use-staticlib  $(HIPCC_FLAGS) $< -o $@
+
+all: $(EXE) $(EXE_STATIC)
 
 clean:
-	rm -f *.o $(EXE)
+	rm -f *.o $(EXE) $(EXE_STATIC)

projects/hip/samples/0_Intro/square/Makefile

@@ -11,7 +11,7 @@ else
 	SOURCES=square.cpp
 endif
 
-all: square.out
+all: square.out square.out.static
 
 # Step
 square.cpp: square.cu
@@ -20,5 +20,8 @@ square.cpp: square.cu
 square.out: $(SOURCES)
 	$(HIPCC) $(CXXFLAGS) $(SOURCES) -o $@
 
+square.out.static: $(SOURCES)
+	$(HIPCC) -use-staticlib $(CXXFLAGS) $(SOURCES) -o $@
+
 clean:
-	rm -f *.o *.out square.cpp
+	rm -f *.o *.out *.out.static square.cpp

projects/hip/vdi/CMakeLists.txt

@@ -156,13 +156,35 @@ add_library(amdhip64_static STATIC
     $<TARGET_OBJECTS:hip64>
     )
 
+# We expect amdhip64_static to contain objects of vdi and hip. But linker
+# let amdhip64_static contain objects of hip only. So we will use a
+# a custom amdhip64_static_combiner to combine objects of vid and hip into
+# amdhip64_static. To avoid amdhip64_static contains itself,
+# amdhip64_static_temp is created internally.
+add_library(amdhip64_static_temp STATIC
+    $<TARGET_OBJECTS:hip64>
+    )
+
 add_library(host INTERFACE)
 target_link_libraries(host INTERFACE amdhip64)
 add_library(device INTERFACE)
 target_link_libraries(device INTERFACE host)
+# TODO: we may create host_static and device_static to let app
+# link amdhip64_static
 
-target_link_libraries(amdhip64_static PRIVATE amdvdi_static pthread dl)
 target_link_libraries(amdhip64 PRIVATE amdvdi_static pthread dl)
+target_link_libraries(amdhip64_static PRIVATE pthread dl)
+target_link_libraries(amdhip64_static_temp PRIVATE pthread dl)
+
+# combine objects of vid and hip into amdhip64_static
+add_custom_target(
+    amdhip64_static_combiner
+    ALL
+    COMMAND rm -f $<TARGET_FILE:amdhip64_static>  # Must remove old one, otherwise the new one will contain obsolete stuff
+    COMMAND ${CMAKE_AR} -rcsT $<TARGET_FILE:amdhip64_static> $<TARGET_FILE:amdhip64_static_temp> $<TARGET_FILE:amdvdi_static>
+    DEPENDS amdhip64_static amdhip64_static_temp amdvdi_static # To make sure this is the last step
+    COMMENT "Combining static libs into amdhip64_static"
+)
 
 
 INSTALL(PROGRAMS $<TARGET_FILE:amdhip64_static> DESTINATION lib COMPONENT MAIN)
@@ -174,6 +196,6 @@ INSTALL(FILES ${CMAKE_BINARY_DIR}/lib/libhip_hcc.so DESTINATION lib COMPONENT MA
 
 INSTALL(FILES ${CMAKE_BINARY_DIR}/lib/libhiprtc.so DESTINATION lib COMPONENT MAIN)
 
-INSTALL(TARGETS amdhip64_static amdhip64 host device EXPORT hip-targets DESTINATION ${LIB_INSTALL_DIR})
+INSTALL(TARGETS amdhip64 amdhip64_static host device EXPORT hip-targets DESTINATION ${LIB_INSTALL_DIR})
 INSTALL(EXPORT hip-targets DESTINATION ${CONFIG_PACKAGE_INSTALL_DIR} NAMESPACE hip::)
 

projects/hip/vdi/hip_internal.hpp

@@ -132,6 +132,7 @@ namespace hip {
   extern void init();
 
   extern Device* getCurrentDevice();
+
   extern void setCurrentDevice(unsigned int index);
 
   /// Get VDI queue associated with hipStream
@@ -255,6 +256,11 @@ private:
   ~PlatformState() {}
 public:
   static PlatformState& instance() {
+    if (platform_ == nullptr) {
+       // __hipRegisterFatBinary() will call this when app starts, thus
+       // there is no multiple entry issue here.
+       platform_ =  new PlatformState();
+    }
     return *platform_;
   }
 

projects/hip/vdi/hip_platform.cpp

@@ -30,7 +30,7 @@
 constexpr unsigned __hipFatMAGIC2 = 0x48495046; // "HIPF"
 
 thread_local std::stack<ihipExec_t> execStack_;
-PlatformState* PlatformState::platform_ = new PlatformState();
+PlatformState* PlatformState::platform_; // Initiaized as nullptr by default
 
 struct __CudaFatBinaryWrapper {
   unsigned int magic;