[GENERAL] Removed deprecated driver files and added basic compatibility with rocm (#268)

- Removed driver module -- accelerator runtime is handled by pytorch
- Added basic support for ROCM based on @micmelesse 's PR -- now can execute empty kernel on AMD devices without any compile-time changes
- Now only using PREFER_SHARED for kernels when the size of shared memory is greater than 49k. Otherwise there can be poor L1 performance for broadcast tensors

CMakeLists.txt

@@ -31,7 +31,7 @@ set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -D__STDC_FORMAT_MACROS  -std=gnu++17")
 # LLVM
 ##########
 if("${LLVM_LIBRARY_DIR}" STREQUAL "")
-    find_package(LLVM 11 REQUIRED COMPONENTS "nvptx")
+    find_package(LLVM 11 REQUIRED COMPONENTS "nvptx;amdgpu")
     message(STATUS "Found LLVM ${LLVM_PACKAGE_VERSION}")
     if(APPLE)
       set(CMAKE_OSX_DEPLOYMENT_TARGET "10.14")
@@ -39,14 +39,52 @@ if("${LLVM_LIBRARY_DIR}" STREQUAL "")
 # sometimes we don't want to use llvm-config, since it may have been downloaded for some specific linux distros
 else()
     set(LLVM_LDFLAGS "-L${LLVM_LIBRARY_DIR}")
-    set(LLVM_LIBRARIES libLLVMNVPTXCodeGen.a libLLVMSelectionDAG.a libLLVMipo.a libLLVMInstrumentation.a
-                       libLLVMVectorize.a libLLVMLinker.a libLLVMIRReader.a libLLVMAsmParser.a libLLVMFrontendOpenMP.a
-                       libLLVMAsmPrinter.a libLLVMDebugInfoDWARF.a libLLVMCodeGen.a libLLVMTarget.a libLLVMScalarOpts.a
-                       libLLVMInstCombine.a libLLVMAggressiveInstCombine.a libLLVMTransformUtils.a libLLVMBitWriter.a
-                       libLLVMAnalysis.a libLLVMProfileData.a libLLVMObject.a libLLVMTextAPI.a libLLVMMCParser.a
-                       libLLVMBitReader.a libLLVMCore.a libLLVMRemarks.a libLLVMBitstreamReader.a libLLVMNVPTXDesc.a
-                       libLLVMMC.a libLLVMDebugInfoCodeView.a libLLVMDebugInfoMSF.a libLLVMBinaryFormat.a libLLVMNVPTXInfo.a
-                       libLLVMSupport.a libLLVMDemangle.a)
+    set(LLVM_LIBRARIES 
+libLLVMNVPTXCodeGen.a
+libLLVMNVPTXDesc.a
+libLLVMNVPTXInfo.a
+libLLVMAMDGPUDisassembler.a
+libLLVMMCDisassembler.a
+libLLVMAMDGPUCodeGen.a
+libLLVMMIRParser.a
+libLLVMGlobalISel.a
+libLLVMSelectionDAG.a
+libLLVMipo.a
+libLLVMInstrumentation.a
+libLLVMVectorize.a
+libLLVMLinker.a
+libLLVMIRReader.a
+libLLVMAsmParser.a
+libLLVMFrontendOpenMP.a
+libLLVMAsmPrinter.a
+libLLVMDebugInfoDWARF.a
+libLLVMCodeGen.a
+libLLVMTarget.a
+libLLVMScalarOpts.a
+libLLVMInstCombine.a
+libLLVMAggressiveInstCombine.a
+libLLVMTransformUtils.a
+libLLVMBitWriter.a
+libLLVMAnalysis.a
+libLLVMProfileData.a
+libLLVMObject.a
+libLLVMTextAPI.a
+libLLVMBitReader.a
+libLLVMAMDGPUAsmParser.a
+libLLVMMCParser.a
+libLLVMAMDGPUDesc.a
+libLLVMAMDGPUUtils.a
+libLLVMMC.a
+libLLVMDebugInfoCodeView.a
+libLLVMDebugInfoMSF.a
+libLLVMCore.a
+libLLVMRemarks.a
+libLLVMBitstreamReader.a
+libLLVMBinaryFormat.a
+libLLVMAMDGPUInfo.a
+libLLVMSupport.a
+libLLVMDemangle.a
+)
 endif()
 include_directories("${LLVM_INCLUDE_DIRS}")
 
@@ -82,4 +120,4 @@ if(BUILD_PYTHON_MODULE)
         set(PYTHON_LDFLAGS "-undefined dynamic_lookup -flto")
     endif()
     target_link_libraries(triton ${CUTLASS_LIBRARIES} ${PYTHON_LDFLAGS})
-endif()
+endif()