[ROCM] enable matmul(dot) and others (#391)

CMakeLists.txt

@@ -7,6 +7,8 @@ if(NOT TRITON_LLVM_BUILD_DIR)
     set(TRITON_LLVM_BUILD_DIR ${CMAKE_BINARY_DIR})
 endif()
 
+set(TRITON_USE_ROCM "$ENV{TRITON_USE_ROCM}")
+set(TRITON_ROCM_DEBUG "$ENV{TRITON_ROCM_DEBUG}")
 
 project(triton)
 include(CTest)
@@ -37,7 +39,11 @@ if(WIN32)
     add_subdirectory(deps/dlfcn-win32/src ${CMAKE_BINARY_DIR}/dlfcn-win32)
 endif()
 
-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -D__STDC_FORMAT_MACROS  -std=gnu++17")
+if (TRITON_USE_ROCM)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -D__STDC_FORMAT_MACROS  -std=gnu++17 -Wno-unused-result -Wno-attributes")
+else()
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -D__STDC_FORMAT_MACROS  -std=gnu++17")
+endif()
 
 
 ##########
@@ -110,6 +116,15 @@ libLLVMBinaryFormat.a
 libLLVMAMDGPUInfo.a
 libLLVMSupport.a
 libLLVMDemangle.a
+libLLVMPasses.a
+libLLVMAnalysis.a
+libLLVMTransformUtils.a
+libLLVMScalarOpts.a
+libLLVMTransformUtils.a
+libLLVMipo.a
+libLLVMObjCARCOpts.a
+libLLVMCoroutines.a
+libLLVMAnalysis.a
 )
 endif()
 include_directories("${LLVM_INCLUDE_DIRS}")
@@ -128,6 +143,13 @@ if(BUILD_PYTHON_MODULE)
     endif()
     include_directories("." ${PYTHON_SRC_PATH} ${PYTHON_INCLUDE_DIRS} ${CUTLASS_INCLUDE_DIR})
     link_directories(${PYTHON_LINK_DIRS} ${CUTLASS_LIBRARY_DIR})
+    if (TRITON_USE_ROCM)
+        add_definitions(-DUSE_ROCM)
+    endif()
+    if (TRITON_ROCM_DEBUG)
+        add_definitions(-DDEBUG_ROCM)
+    endif()
+
     set(PYTHON_SRC ${PYTHON_SRC_PATH}/main.cc ${PYTHON_SRC_PATH}/triton.cc  ${PYTHON_SRC_PATH}/superblock.cc ${CUTLASS_SRC})
 endif()