[MCA] Moving the target specific CustomBehaviour impl. from /tools/llvm-mca/ to /lib/Target/.

[holland11]

Summary

This patch is related to https://reviews.llvm.org/D104149 and https://reviews.llvm.org/D104730 . The first patch introduced the CustomBehaviour class to llvm-mca. To start, target specific CBs were meant to be placed within the llvm-mca directory. Since the class is only used by llvm-mca, this made sense and was the cleanest and most intuitive place to put them.

However, in the second patch, I attempted to implement some custom modelling for AMDGPU. This modelling required the use of some backend functions and symbols. This works fine for static builds and for shared builds that are not on linux. However, when building with shared libs on linux, there were build errors because the backend symbols were not accessible by AMDGPUCB. For more details on this, you can refer to the most recent comments within the second diff linked above.

To address these errors, I restructured the target specific CB implementations to be placed within the /lib/Target/<Target-Name>/MCA directory. (This MCA directory is new.) Since the implementations are now a part of the target's backend, they need to be initialized and registered similar to how many other target specific classes work (such as AsmParser). To accomplish this, I had to add some code to several cmake files throughout llvm. This is my first time working with general llvm config stuff so if I made any mistakes, hopefully someone can spot them and help me correct them.

This patch keeps the AMDGPUCB implementation empty. If/when this patch gets approved and pushed, I will submit a diff for the AMDGPUCB implementation (which is the same implementation seen in the second diff linked above).

I have tested several builds and run ninja check all on them on both macOS and linux. I have tested them with the future AMDGPUCB implementation as well to make sure that this solution does in fact fix the original error regarding unresolved symbols when building with shared libs on linux.

If anybody sees any issues or has any suggestions for ways that this can be done better, I'd be happy to tweak or modify this patch.

Thanks for your time!

[andreadb]

Looks good to me.

[foad]

llvm/lib/Target/AMDGPU/CMakeLists.txt

159

I don't speak cmake. What does this line do? Does it mean that the target MCA class will be pulled into the AMDGPUCodeGen library? That seems unnecessary, since it's not needed for codegen.

[andreadb]

llvm/lib/Target/AMDGPU/CMakeLists.txt

159

That's correct.

In https://reviews.llvm.org/D104730, Patrick has suggested three options for solving this issue. This was his comment:

@tstellar @andreadb I've been thinking about the problem for a bit now and have been brainstorming different ideas and digging through the code. I just made a post on the llvm discourse to hopefully get some feedback on the issue (and some potential solutions that I've come up with). If you guys are curious or would be willing to take a look, here's the post https://llvm.discourse.group/t/need-help-deciding-how-to-access-backend-functions-from-within-a-tool/3915 .

On that discourse link, both me and Tom have suggested to implement option 2. (i.e. this patch). We believe that it is the simplest solution, and we don't expect the CustomBehaviour code to ever be big enough in general.

I agree that this change is not strictly needed for codegen now. However, that might change in future.

During the development of MCA,it was requested to make MCA accessible from Codegen. That is because a few people (including @courbet 's team) expressed the intention of using MCA to define their own custom machine scheduler passes.

Unfortunately, we never really got to that point (mainly due to the lack of devs time). Nowadays, MCA is also built as a library. However, it is missing some lowering functionalities to translate from MachineInstr to mca::Instruction. However, that could change in future; if so, then this patch would make things much easier.

That being said, the main reason why we suggested to use this approach is mainly because we assumed that the impact on Codegen of CustomBehaviour objects would always be relatively small. Maybe Patrick could give a bit more details on this (since he presumably has also some numbers to show).

[foad]

llvm/lib/Target/AMDGPU/CMakeLists.txt

159

Thanks for explaining. I don't object to this, just wanted to make sure I understood how it worked.

[holland11]

I have just removed the AMDGPUTargetMCA line from the AMDGPUCodeGen cmake file. I was originally just following the patterns that I saw from the other folders and classes in that directory and then afterwards I went through and tried to remove things that didn't seem necessary. I guess I missed that one.

Tested it on both macOS and linux and it seems to still work properly. CodeGen definitely doesn't depend on TargetMCA so it was not necessary to have that line.

Will let the build test do its thing and if it looks good, I'll push this patch and watch the build bots closely in case I need to revert.

[lebedev.ri]

Have you looked at how per-target stuff of llvm-mca is handled?
Does that also need similar changes? If not, perhaps this can be changed to more closely mimic that instead?

[holland11]

In D106775#2908451, @lebedev.ri wrote:

Have you looked at how per-target stuff of llvm-mca is handled?
Does that also need similar changes? If not, perhaps this can be changed to more closely mimic that instead?

As far as MCA specific stuff goes, this is the only 'per-target' code that I'm aware of. MCA does use target specific classes (that aren't exclusive to mca) such as AsmParser, MCTargetStreamer, etc. and those are what I based this implementation off of.

[andreadb]

In D106775#2908396, @holland11 wrote:

I have just removed the AMDGPUTargetMCA line from the AMDGPUCodeGen cmake file. I was originally just following the patterns that I saw from the other folders and classes in that directory and then afterwards I went through and tried to remove things that didn't seem necessary. I guess I missed that one.

Tested it on both macOS and linux and it seems to still work properly. CodeGen definitely doesn't depend on TargetMCA so it was not necessary to have that line.

Will let the build test do its thing and if it looks good, I'll push this patch and watch the build bots closely in case I need to revert.

Sounds good. Thanks!

llvm/CMakeLists.txt

 ######
 
 # Configure all of the various header file fragments LLVM uses which depend on
 # configuration variables.
 set(LLVM_ENUM_TARGETS "")
 set(LLVM_ENUM_ASM_PRINTERS "")
 set(LLVM_ENUM_ASM_PARSERS "")
 set(LLVM_ENUM_DISASSEMBLERS "")
+set(LLVM_ENUM_TARGETMCAS "")
 foreach(t ${LLVM_TARGETS_TO_BUILD})
   set( td ${LLVM_MAIN_SRC_DIR}/lib/Target/${t} )
 
   list(FIND LLVM_ALL_TARGETS ${t} idx)
   list(FIND LLVM_EXPERIMENTAL_TARGETS_TO_BUILD ${t} idy)
   # At this point, LLVMBUILDTOOL already checked all the targets passed in
   # LLVM_TARGETS_TO_BUILD and LLVM_EXPERIMENTAL_TARGETS_TO_BUILD, so
   # this test just makes sure that any experimental targets were passed via
foreach(t ${LLVM_TARGETS_TO_BUILD})
   if( EXISTS ${td}/AsmParser/CMakeLists.txt )
     set(LLVM_ENUM_ASM_PARSERS
       "${LLVM_ENUM_ASM_PARSERS}LLVM_ASM_PARSER(${t})\n")
   endif()
   if( EXISTS ${td}/Disassembler/CMakeLists.txt )
     set(LLVM_ENUM_DISASSEMBLERS
       "${LLVM_ENUM_DISASSEMBLERS}LLVM_DISASSEMBLER(${t})\n")
   endif()
+    if( EXISTS ${td}/MCA/CMakeLists.txt )
+    set(LLVM_ENUM_TARGETMCAS
+      "${LLVM_ENUM_TARGETMCAS}LLVM_TARGETMCA(${t})\n")
+  endif()
 endforeach(t)
 
 # Produce the target definition files, which provide a way for clients to easily
 # include various classes of targets.
 configure_file(
   ${LLVM_MAIN_INCLUDE_DIR}/llvm/Config/AsmPrinters.def.in
   ${LLVM_INCLUDE_DIR}/llvm/Config/AsmPrinters.def
   )
 configure_file(
   ${LLVM_MAIN_INCLUDE_DIR}/llvm/Config/AsmParsers.def.in
   ${LLVM_INCLUDE_DIR}/llvm/Config/AsmParsers.def
   )
 configure_file(
   ${LLVM_MAIN_INCLUDE_DIR}/llvm/Config/Disassemblers.def.in
   ${LLVM_INCLUDE_DIR}/llvm/Config/Disassemblers.def
   )
 configure_file(
   ${LLVM_MAIN_INCLUDE_DIR}/llvm/Config/Targets.def.in
   ${LLVM_INCLUDE_DIR}/llvm/Config/Targets.def
   )
+configure_file(
+  ${LLVM_MAIN_INCLUDE_DIR}/llvm/Config/TargetMCAs.def.in
+  ${LLVM_INCLUDE_DIR}/llvm/Config/TargetMCAs.def
+  )
 
 # They are not referenced. See set_output_directory().
 set( CMAKE_RUNTIME_OUTPUT_DIRECTORY ${LLVM_BINARY_DIR}/bin )
 set( CMAKE_LIBRARY_OUTPUT_DIRECTORY ${LLVM_BINARY_DIR}/lib${LLVM_LIBDIR_SUFFIX} )
 set( CMAKE_ARCHIVE_OUTPUT_DIRECTORY ${LLVM_BINARY_DIR}/lib${LLVM_LIBDIR_SUFFIX} )
 
 if(LLVM_INCLUDE_TESTS)
   include(GetErrcMessages)

llvm/cmake/modules/LLVM-Config.cmake

elseif( c STREQUAL "AllTargetsDisassemblers" )
       endforeach(t)
     elseif( c STREQUAL "AllTargetsInfos" )
       # Link all the infos from all the targets
       foreach(t ${LLVM_TARGETS_TO_BUILD})
         if( TARGET LLVM${t}Info )
           list(APPEND expanded_components "${t}Info")
         endif()
       endforeach(t)
+    elseif( c STREQUAL "AllTargetsMCAs" )
+      # Link all the TargetMCAs from all the targets
+      foreach(t ${LLVM_TARGETS_TO_BUILD})
+        if( TARGET LLVM${t}TargetMCA )
+          list(APPEND expanded_components "${t}TargetMCA")
+        endif()
+      endforeach(t)
     else()
       list(APPEND expanded_components "${c}")
     endif()
   endforeach()
   set(${out_components} ${expanded_components} PARENT_SCOPE)
 endfunction(llvm_expand_pseudo_components out_components)
 
 # This is a variant intended for the final user:

llvm/docs/CommandGuide/llvm-mca.rst

.. option:: -json
   Print the requested views in valid JSON format. The instructions and the
   processor resources are printed as members of special top level JSON objects.
   The individual views refer to them by index. However, not all views are
   currently supported. For example, the report from the bottleneck analysis is
   not printed out in JSON. All the default views are currently supported.
 
 .. option:: -disable-cb
 
-  Force usage of the generic CustomBehaviour class rather than using the target
-  specific class. The generic class never detects any custom hazards.
+  Force usage of the generic CustomBehaviour and InstrPostProcess classes rather
+  than using the target specific implementation. The generic classes never
+  detect any custom hazards or make any post processing modifications to
+  instructions.
 
 
 EXIT STATUS
 -----------
 
 :program:`llvm-mca` returns 0 on success. Otherwise, an error message is printed
 to standard error, and the tool returns 1.
 

 current instruction and a list of all instructions still executing within
 the pipeline to determine if the current instruction should be dispatched.
 As output, the method returns an integer representing the number of cycles
 that the current instruction must stall for (this can be an underestimate
 if you don't know the exact number and a value of 0 represents no stall).
 
 If you'd like to add a CustomBehaviour class for a target that doesn't
 already have one, refer to an existing implementation to see how to set it
-up. Remember to look at (and add to) `/llvm-mca/lib/CMakeLists.txt`.
+up. The classes are implemented within the target specific backend (for
+example `/llvm/lib/Target/AMDGPU/MCA/`) so that they can access backend symbols.

llvm/include/llvm/Config/TargetMCAs.def.in

+/*===------ llvm/Config/TargetMCAs.def - LLVM Target MCAs -------*- C++ -*-===*\
+|*                                                                            *|
+|* Part of the LLVM Project, under the Apache License v2.0 with LLVM          *|
+|* Exceptions.                                                                *|
+|* See https://llvm.org/LICENSE.txt for license information.                  *|
+|* SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception                    *|
+|*                                                                            *|
+|*===----------------------------------------------------------------------===*|
+|*                                                                            *|
+|* This file enumerates all of the target MCAs                  *|
+|* supported by this build of LLVM. Clients of this file should define        *|
+|* the LLVM_TARGETMCA macro to be a function-like macro with a             *|
+|* single parameter (the name of the target whose assembly can be             *|
+|* generated); including this file will then enumerate all of the             *|
+|* targets with target MCAs.                                             *|
+|*                                                                            *|
+|* The set of targets supported by LLVM is generated at configuration         *|
+|* time, at which point this header is generated. Do not modify this          *|
+|* header directly.                                                           *|
+|*                                                                            *|
+\*===----------------------------------------------------------------------===*/
+
+#ifndef LLVM_TARGETMCA
+#  error Please define the macro LLVM_TARGETMCA(TargetName)
+#endif
+
+@LLVM_ENUM_TARGETMCAS@
+
+#undef LLVM_TARGETMCA

llvm/include/llvm/Config/llvm-config.h.cmake

 #cmakedefine LLVM_NATIVE_TARGET LLVMInitialize${LLVM_NATIVE_ARCH}Target
 
 /* LLVM name for the native TargetInfo init function, if available */
 #cmakedefine LLVM_NATIVE_TARGETINFO LLVMInitialize${LLVM_NATIVE_ARCH}TargetInfo
 
 /* LLVM name for the native target MC init function, if available */
 #cmakedefine LLVM_NATIVE_TARGETMC LLVMInitialize${LLVM_NATIVE_ARCH}TargetMC
 
+/* LLVM name for the native target MCA init function, if available */
+#cmakedefine LLVM_NATIVE_TARGETMCA LLVMInitialize${LLVM_NATIVE_ARCH}TargetMCA
+
 /* Define if this is Unixish platform */
 #cmakedefine LLVM_ON_UNIX ${LLVM_ON_UNIX}
 
 /* Define if we have the Intel JIT API runtime support library */
 #cmakedefine01 LLVM_USE_INTEL_JITEVENTS
 
 /* Define if we have the oprofile JIT-support library */
 #cmakedefine01 LLVM_USE_OPROFILE

llvm/include/llvm/MCA/CustomBehaviour.h

 /// dependencies and behaviours that aren't expressed well enough
 /// within the scheduling model for mca to automatically simulate
 /// them properly.
 /// If you implement this class for your target, make sure to also implement
 /// a target specific InstrPostProcess class as well.
 class CustomBehaviour {
 protected:
   const MCSubtargetInfo &STI;
-  const SourceMgr &SrcMgr;
+  const mca::SourceMgr &SrcMgr;
   const MCInstrInfo &MCII;
 
 public:
-  CustomBehaviour(const MCSubtargetInfo &STI, const SourceMgr &SrcMgr,
+  CustomBehaviour(const MCSubtargetInfo &STI, const mca::SourceMgr &SrcMgr,
                   const MCInstrInfo &MCII)
       : STI(STI), SrcMgr(SrcMgr), MCII(MCII) {}
 
   virtual ~CustomBehaviour();
 
   // Before the llvm-mca pipeline dispatches an instruction, it first checks
   // for any register or resource dependencies / hazards. If it doesn't find
   // any, this method will be invoked to determine if there are any custom

llvm/include/llvm/Support/TargetRegistry.h

 class MCSymbolizer;
 class MCTargetAsmParser;
 class MCTargetOptions;
 class MCTargetStreamer;
 class raw_ostream;
 class raw_pwrite_stream;
 class TargetMachine;
 class TargetOptions;
+namespace mca {
+class CustomBehaviour;
+class InstrPostProcess;
+class SourceMgr;
+} // namespace mca
 
 MCStreamer *createNullStreamer(MCContext &Ctx);
 // Takes ownership of \p TAB and \p CE.
 
 /// Create a machine code streamer which will print out assembly for the native
 /// target, suitable for compiling with a native assembler.
 ///
 /// \param InstPrint - If given, the instruction printer to use. If not given

 
 MCRelocationInfo *createMCRelocationInfo(const Triple &TT, MCContext &Ctx);
 
 MCSymbolizer *createMCSymbolizer(const Triple &TT, LLVMOpInfoCallback GetOpInfo,
                                  LLVMSymbolLookupCallback SymbolLookUp,
                                  void *DisInfo, MCContext *Ctx,
                                  std::unique_ptr<MCRelocationInfo> &&RelInfo);
 
+mca::CustomBehaviour *createCustomBehaviour(const MCSubtargetInfo &STI,
+                                            const mca::SourceMgr &SrcMgr,
+                                            const MCInstrInfo &MCII);
+
+mca::InstrPostProcess *createInstrPostProcess(const MCSubtargetInfo &STI,
+                                              const MCInstrInfo &MCII);
+
 /// Target - Wrapper for Target specific information.
 ///
 /// For registration purposes, this is a POD type so that targets can be
 /// registered without the use of static constructors.
 ///
 /// Targets should implement a single global instance of this class (which
 /// will be zero initialized), and pass that instance to the TargetRegistry as
 /// part of their initialization.
using ObjectTargetStreamerCtorTy = MCTargetStreamer *(*)(
       MCStreamer &S, const MCSubtargetInfo &STI);
   using MCRelocationInfoCtorTy = MCRelocationInfo *(*)(const Triple &TT,
                                                        MCContext &Ctx);
   using MCSymbolizerCtorTy = MCSymbolizer *(*)(
       const Triple &TT, LLVMOpInfoCallback GetOpInfo,
       LLVMSymbolLookupCallback SymbolLookUp, void *DisInfo, MCContext *Ctx,
       std::unique_ptr<MCRelocationInfo> &&RelInfo);
 
+  using CustomBehaviourCtorTy =
+      mca::CustomBehaviour *(*)(const MCSubtargetInfo &STI,
+                                const mca::SourceMgr &SrcMgr,
+                                const MCInstrInfo &MCII);
+
+  using InstrPostProcessCtorTy =
+      mca::InstrPostProcess *(*)(const MCSubtargetInfo &STI,
+                                 const MCInstrInfo &MCII);
+
 private:
   /// Next - The next registered target in the linked list, maintained by the
   /// TargetRegistry.
   Target *Next;
 
   /// The target function for checking if an architecture is supported.
   ArchMatchFnTy ArchMatchFn;
 
private:
   /// MCRelocationInfoCtorFn - Construction function for this target's
   /// MCRelocationInfo, if registered (default = llvm::createMCRelocationInfo)
   MCRelocationInfoCtorTy MCRelocationInfoCtorFn = nullptr;
 
   /// MCSymbolizerCtorFn - Construction function for this target's
   /// MCSymbolizer, if registered (default = llvm::createMCSymbolizer)
   MCSymbolizerCtorTy MCSymbolizerCtorFn = nullptr;
 
+  /// CustomBehaviourCtorFn - Construction function for this target's
+  /// CustomBehaviour, if registered (default = nullptr).
+  CustomBehaviourCtorTy CustomBehaviourCtorFn = nullptr;
+
+  /// InstrPostProcessCtorFn - Construction function for this target's
+  /// InstrPostProcess, if registered (default = nullptr).
+  InstrPostProcessCtorTy InstrPostProcessCtorFn = nullptr;
+
 public:
   Target() = default;
 
   /// @name Target Information
   /// @{
 
   // getNext - Return the next registered target.
   const Target *getNext() const { return Next; }
createMCSymbolizer(StringRef TT, LLVMOpInfoCallback GetOpInfo,
                      MCContext *Ctx,
                      std::unique_ptr<MCRelocationInfo> &&RelInfo) const {
     MCSymbolizerCtorTy Fn =
         MCSymbolizerCtorFn ? MCSymbolizerCtorFn : llvm::createMCSymbolizer;
     return Fn(Triple(TT), GetOpInfo, SymbolLookUp, DisInfo, Ctx,
               std::move(RelInfo));
   }
 
+  /// createCustomBehaviour - Create a target specific CustomBehaviour.
+  /// This class is used by llvm-mca and requires backend functionality.
+  mca::CustomBehaviour *createCustomBehaviour(const MCSubtargetInfo &STI,
+                                              const mca::SourceMgr &SrcMgr,
+                                              const MCInstrInfo &MCII) const {
+    if (CustomBehaviourCtorFn)
+      return CustomBehaviourCtorFn(STI, SrcMgr, MCII);
+    return nullptr;
+  }
+
+  /// createInstrPostProcess - Create a target specific InstrPostProcess.
+  /// This class is used by llvm-mca and requires backend functionality.
+  mca::InstrPostProcess *createInstrPostProcess(const MCSubtargetInfo &STI,
+                                                const MCInstrInfo &MCII) const {
+    if (InstrPostProcessCtorFn)
+      return InstrPostProcessCtorFn(STI, MCII);
+    return nullptr;
+  }
+
   /// @}
 };
 
 /// TargetRegistry - Generic interface to target specific features.
 struct TargetRegistry {
   // FIXME: Make this a namespace, probably just move all the Register*
   // functions into Target (currently they all just set members on the Target
   // anyway, and Target friends this class so those functions can...
struct TargetRegistry {
   /// this is done by initializing all targets at program startup.
   ///
   /// @param T - The target being registered.
   /// @param Fn - A function to construct an MCSymbolizer for the target.
   static void RegisterMCSymbolizer(Target &T, Target::MCSymbolizerCtorTy Fn) {
     T.MCSymbolizerCtorFn = Fn;
   }
 
+  /// RegisterCustomBehaviour - Register a CustomBehaviour
+  /// implementation for the given target.
+  ///
+  /// Clients are responsible for ensuring that registration doesn't occur
+  /// while another thread is attempting to access the registry. Typically
+  /// this is done by initializing all targets at program startup.
+  ///
+  /// @param T - The target being registered.
+  /// @param Fn - A function to construct a CustomBehaviour for the target.
+  static void RegisterCustomBehaviour(Target &T,
+                                      Target::CustomBehaviourCtorTy Fn) {
+    T.CustomBehaviourCtorFn = Fn;
+  }
+
+  /// RegisterInstrPostProcess - Register an InstrPostProcess
+  /// implementation for the given target.
+  ///
+  /// Clients are responsible for ensuring that registration doesn't occur
+  /// while another thread is attempting to access the registry. Typically
+  /// this is done by initializing all targets at program startup.
+  ///
+  /// @param T - The target being registered.
+  /// @param Fn - A function to construct an InstrPostProcess for the target.
+  static void RegisterInstrPostProcess(Target &T,
+                                       Target::InstrPostProcessCtorTy Fn) {
+    T.InstrPostProcessCtorFn = Fn;
+  }
+
   /// @}
 };
 
 //===--------------------------------------------------------------------===//
 
 /// RegisterTarget - Helper template for registering a target, for use in the
 /// target's initialization function. Usage:
 ///

llvm/include/llvm/Support/TargetSelect.h

#include "llvm/Config/AsmPrinters.def"
   // Declare all of the available assembly parser initialization functions.
 #define LLVM_ASM_PARSER(TargetName) void LLVMInitialize##TargetName##AsmParser();
 #include "llvm/Config/AsmParsers.def"
 
   // Declare all of the available disassembler initialization functions.
 #define LLVM_DISASSEMBLER(TargetName) \
   void LLVMInitialize##TargetName##Disassembler();
 #include "llvm/Config/Disassemblers.def"
+
+// Declare all of the available TargetMCA initialization functions.
+#define LLVM_TARGETMCA(TargetName) void LLVMInitialize##TargetName##TargetMCA();
+#include "llvm/Config/TargetMCAs.def"
 }
 
 namespace llvm {
   /// InitializeAllTargetInfos - The main program should call this function if
   /// it wants access to all available targets that LLVM is configured to
   /// support, to make them available via the TargetRegistry.
   ///
   /// It is legal for a client to make multiple calls to this function.
#endif
   // If we have a native target, initialize the corresponding disassembler.
 #ifdef LLVM_NATIVE_DISASSEMBLER
     LLVM_NATIVE_DISASSEMBLER();
     return false;
 #else
     return true;
 #endif
   }
+
+  /// InitializeAllTargetMCAs - The main program should call
+  /// this function to initialize the target CustomBehaviour and
+  /// InstrPostProcess classes.
+  inline void InitializeAllTargetMCAs() {
+#define LLVM_TARGETMCA(TargetName) LLVMInitialize##TargetName##TargetMCA();
+#include "llvm/Config/TargetMCAs.def"
+  }
 }
 
 #endif

llvm/lib/Target/AMDGPU/CMakeLists.txt

add_llvm_target(AMDGPUCodeGen
   LINK_COMPONENTS
   Analysis
   AsmPrinter
   CodeGen
   Core
   IPO
   MC
   Passes
+  AMDGPUTargetMCA
   AMDGPUDesc
   AMDGPUInfo
   AMDGPUUtils
   Scalar
   SelectionDAG
   Support
   Target
   TransformUtils
   Vectorize
   GlobalISel
   BinaryFormat
   MIRParser
 
   ADD_TO_COMPONENT
   AMDGPU
   )
 
 add_subdirectory(AsmParser)
 add_subdirectory(Disassembler)
+add_subdirectory(MCA)
 add_subdirectory(MCTargetDesc)
 add_subdirectory(TargetInfo)
 add_subdirectory(Utils)

llvm/lib/Target/AMDGPU/MCA/AMDGPUCustomBehaviour.h

 //===------------------- AMDGPUCustomBehaviour.h ----------------*-C++ -* -===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 /// \file
 ///
 /// This file defines the AMDGPUCustomBehaviour class which inherits from
-/// CustomBehaviour.
+/// CustomBehaviour. This class is used by the tool llvm-mca to enforce
+/// target specific behaviour that is not expressed well enough in the
+/// scheduling model for mca to enforce it automatically.
 ///
 //===----------------------------------------------------------------------===//
 
-#ifndef LLVM_TOOLS_LLVM_MCA_LIB_AMDGPU_AMDGPUCUSTOMBEHAVIOUR_H
-#define LLVM_TOOLS_LLVM_MCA_LIB_AMDGPU_AMDGPUCUSTOMBEHAVIOUR_H
+#ifndef LLVM_LIB_TARGET_AMDGPU_MCA_AMDGPUCUSTOMBEHAVIOUR_H
+#define LLVM_LIB_TARGET_AMDGPU_MCA_AMDGPUCUSTOMBEHAVIOUR_H
 
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/MCA/CustomBehaviour.h"
 #include "llvm/Support/TargetParser.h"
 
 namespace llvm {
 namespace mca {
 
 class AMDGPUInstrPostProcess : public InstrPostProcess {
 public:
   AMDGPUInstrPostProcess(const MCSubtargetInfo &STI, const MCInstrInfo &MCII)
       : InstrPostProcess(STI, MCII) {}
 
   ~AMDGPUInstrPostProcess() {}
 
-  void postProcessInstruction(std::unique_ptr<Instruction> &Inst,
+  void postProcessInstruction(std::unique_ptr<mca::Instruction> &Inst,
                               const MCInst &MCI) override {}
 };
 
 class AMDGPUCustomBehaviour : public CustomBehaviour {
 public:
-  AMDGPUCustomBehaviour(const MCSubtargetInfo &STI, const SourceMgr &SrcMgr,
-                        const MCInstrInfo &MCII);
+  AMDGPUCustomBehaviour(const MCSubtargetInfo &STI,
+                        const mca::SourceMgr &SrcMgr, const MCInstrInfo &MCII);
 
   ~AMDGPUCustomBehaviour() {}
 
   /// This method is used to determine if an instruction
   /// should be allowed to be dispatched. The return value is
   /// how many cycles until the instruction can be dispatched.
   /// This method is called after MCA has already checked for
   /// register and hardware dependencies so this method should only
   /// implement custom behaviour and dependencies that are not picked up
   /// by MCA naturally.
-  unsigned checkCustomHazard(ArrayRef<InstRef> IssuedInst,
-                             const InstRef &IR) override;
+  unsigned checkCustomHazard(ArrayRef<mca::InstRef> IssuedInst,
+                             const mca::InstRef &IR) override;
 };
 
 } // namespace mca
 } // namespace llvm
 
-#endif /* LLVM_TOOLS_LLVM_MCA_LIB_AMDGPU_AMDGPUCUSTOMBEHAVIOUR_H */
+#endif

llvm/lib/Target/AMDGPU/MCA/AMDGPUCustomBehaviour.cpp

+//===------------------ AMDGPUCustomBehaviour.cpp ---------------*-C++ -* -===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+/// \file
+///
+/// This file implements methods from the AMDGPUCustomBehaviour class.
+///
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPUCustomBehaviour.h"
+#include "SIInstrInfo.h"
+#include "MCTargetDesc/AMDGPUMCTargetDesc.h"
+#include "llvm/Support/TargetRegistry.h"
+#include "llvm/Support/WithColor.h"
+#include "TargetInfo/AMDGPUTargetInfo.h"
+
+namespace llvm {
+namespace mca {
+
+AMDGPUCustomBehaviour::AMDGPUCustomBehaviour(const MCSubtargetInfo &STI,
+                                             const mca::SourceMgr &SrcMgr,
+                                             const MCInstrInfo &MCII)
+    : CustomBehaviour(STI, SrcMgr, MCII) {}
+
+unsigned
+AMDGPUCustomBehaviour::checkCustomHazard(ArrayRef<mca::InstRef> IssuedInst,
+                                         const mca::InstRef &IR) {
+  return 0;
+}
+
+} // namespace mca
+} // namespace llvm
+
+using namespace llvm;
+using namespace mca;
+
+static CustomBehaviour *
+createAMDGPUCustomBehaviour(const MCSubtargetInfo &STI,
+                            const mca::SourceMgr &SrcMgr,
+                            const MCInstrInfo &MCII) {
+  return new AMDGPUCustomBehaviour(STI, SrcMgr, MCII);
+}
+
+static InstrPostProcess *
+createAMDGPUInstrPostProcess(const MCSubtargetInfo &STI,
+                             const MCInstrInfo &MCII) {
+  return new AMDGPUInstrPostProcess(STI, MCII);
+}
+
+/// Extern function to initialize the targets for the AMDGPU backend
+
+extern "C" LLVM_EXTERNAL_VISIBILITY void LLVMInitializeAMDGPUTargetMCA() {
+  TargetRegistry::RegisterCustomBehaviour(getTheAMDGPUTarget(), createAMDGPUCustomBehaviour);
+  TargetRegistry::RegisterInstrPostProcess(getTheAMDGPUTarget(), createAMDGPUInstrPostProcess);
+  
+  TargetRegistry::RegisterCustomBehaviour(getTheGCNTarget(), createAMDGPUCustomBehaviour);
+  TargetRegistry::RegisterInstrPostProcess(getTheGCNTarget(), createAMDGPUInstrPostProcess);
+}

llvm/lib/Target/AMDGPU/MCA/CMakeLists.txt

+add_llvm_component_library(LLVMAMDGPUTargetMCA
+  AMDGPUCustomBehaviour.cpp
+
+  LINK_COMPONENTS
+  MC
+  MCParser
+  AMDGPUDesc
+  AMDGPUInfo
+  AMDGPUUtils
+  Support
+  MCA
+
+  ADD_TO_COMPONENT
+  AMDGPU
+  )
+
+add_dependencies(LLVMAMDGPUTargetMCA LLVMAMDGPUUtils)
 No newline at end of file

llvm/tools/llvm-mca/CMakeLists.txt

 include_directories(include)
 
-add_subdirectory(lib)
-
 set(LLVM_LINK_COMPONENTS
   AllTargetsAsmParsers
+  AllTargetsMCAs          # CustomBehaviour and InstrPostProcess
   AllTargetsDescs
   AllTargetsDisassemblers
   AllTargetsInfos
   MCA
   MC
   MCParser
   Support
   )
add_llvm_tool(llvm-mca
   Views/RetireControlUnitStatistics.cpp
   Views/SchedulerStatistics.cpp
   Views/SummaryView.cpp
   Views/TimelineView.cpp
   Views/View.cpp
   )
 
 set(LLVM_MCA_SOURCE_DIR ${CURRENT_SOURCE_DIR})
-
-target_link_libraries(llvm-mca PRIVATE
-  ${LLVM_MCA_CUSTOMBEHAVIOUR_TARGETS}
-  )
-
-add_definitions(${LLVM_MCA_MACROS_TO_DEFINE})

llvm/tools/llvm-mca/CodeRegionGenerator.h

 /// driver, and converting that into a CodeRegions instance.
 class CodeRegionGenerator {
 protected:
   CodeRegions Regions;
   CodeRegionGenerator(const CodeRegionGenerator &) = delete;
   CodeRegionGenerator &operator=(const CodeRegionGenerator &) = delete;
 
 public:
-  CodeRegionGenerator(SourceMgr &SM) : Regions(SM) {}
+  CodeRegionGenerator(llvm::SourceMgr &SM) : Regions(SM) {}
   virtual ~CodeRegionGenerator();
   virtual Expected<const CodeRegions &>
   parseCodeRegions(const std::unique_ptr<MCInstPrinter> &IP) = 0;
 };
 
 /// This class is responsible for parsing input ASM and generating
 /// a CodeRegions instance.
 class AsmCodeRegionGenerator final : public CodeRegionGenerator {
   const Target &TheTarget;
   MCContext &Ctx;
   const MCAsmInfo &MAI;
   const MCSubtargetInfo &STI;
   const MCInstrInfo &MCII;
   unsigned AssemblerDialect; // This is set during parsing.
 
 public:
-  AsmCodeRegionGenerator(const Target &T, SourceMgr &SM, MCContext &C,
+  AsmCodeRegionGenerator(const Target &T, llvm::SourceMgr &SM, MCContext &C,
                          const MCAsmInfo &A, const MCSubtargetInfo &S,
                          const MCInstrInfo &I)
       : CodeRegionGenerator(SM), TheTarget(T), Ctx(C), MAI(A), STI(S), MCII(I),
         AssemblerDialect(0) {}
 
   unsigned getAssemblerDialect() const { return AssemblerDialect; }
   Expected<const CodeRegions &>
   parseCodeRegions(const std::unique_ptr<MCInstPrinter> &IP) override;
 };
 
 } // namespace mca
 } // namespace llvm
 
 #endif // LLVM_TOOLS_LLVM_MCA_CODEREGION_GENERATOR_H

llvm/tools/llvm-mca/lib/AMDGPU/AMDGPUCustomBehaviour.cpp

-//===------------------ AMDGPUCustomBehaviour.cpp ---------------*-C++ -* -===//
-//
-// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
-// See https://llvm.org/LICENSE.txt for license information.
-// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
-//
-//===----------------------------------------------------------------------===//
-/// \file
-///
-/// This file implements methods from the AMDGPUCustomBehaviour class.
-///
-//===----------------------------------------------------------------------===//
-
-#include "AMDGPUCustomBehaviour.h"
-#include "MCTargetDesc/AMDGPUMCTargetDesc.h"
-#include "SIInstrInfo.h"
-#include "llvm/Support/WithColor.h"
-
-namespace llvm {
-namespace mca {
-
-AMDGPUCustomBehaviour::AMDGPUCustomBehaviour(const MCSubtargetInfo &STI,
-                                             const SourceMgr &SrcMgr,
-                                             const MCInstrInfo &MCII)
-    : CustomBehaviour(STI, SrcMgr, MCII) {}
-
-unsigned AMDGPUCustomBehaviour::checkCustomHazard(ArrayRef<InstRef> IssuedInst,
-                                                  const InstRef &IR) {
-  return 0;
-}
-
-} // namespace mca
-} // namespace llvm

llvm/tools/llvm-mca/lib/AMDGPU/CMakeLists.txt

-include_directories(
-  ${LLVM_MAIN_SRC_DIR}/lib/Target/AMDGPU
-  ${LLVM_BINARY_DIR}/lib/Target/AMDGPU
-  )
-
-set(LLVM_LINK_COMPONENTS
-  AMDGPU
-  Core
-  MCA
-  Support
-  )
-
-add_llvm_library(LLVMMCACustomBehaviourAMDGPU
-  AMDGPUCustomBehaviour.cpp
-
-  DEPENDS
-  AMDGPUCommonTableGen
-  )

llvm/tools/llvm-mca/lib/CMakeLists.txt

-set(TARGETS_TO_APPEND "")
-set(MACROS_TO_APPEND "")
-
-if (LLVM_TARGETS_TO_BUILD MATCHES "AMDGPU")
-  add_subdirectory(AMDGPU)
-  list(APPEND TARGETS_TO_APPEND LLVMMCACustomBehaviourAMDGPU)
-  list(APPEND MACROS_TO_APPEND -DHAS_AMDGPU)
-endif()
-
-set(LLVM_MCA_CUSTOMBEHAVIOUR_TARGETS ${TARGETS_TO_APPEND} PARENT_SCOPE)
-set(LLVM_MCA_MACROS_TO_DEFINE ${MACROS_TO_APPEND} PARENT_SCOPE)

llvm/tools/llvm-mca/llvm-mca.cpp

 #include "Views/DispatchStatistics.h"
 #include "Views/InstructionInfoView.h"
 #include "Views/RegisterFileStatistics.h"
 #include "Views/ResourcePressureView.h"
 #include "Views/RetireControlUnitStatistics.h"
 #include "Views/SchedulerStatistics.h"
 #include "Views/SummaryView.h"
 #include "Views/TimelineView.h"
-#ifdef HAS_AMDGPU
-#include "lib/AMDGPU/AMDGPUCustomBehaviour.h"
-#endif
 #include "llvm/MC/MCAsmBackend.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCCodeEmitter.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCObjectFileInfo.h"
 #include "llvm/MC/MCRegisterInfo.h"
 #include "llvm/MC/MCSubtargetInfo.h"
 #include "llvm/MC/MCTargetOptionsCommandFlags.h"
const cl::opt<bool> &Default =
           : EnableAllViews;
   processOptionImpl(PrintRegisterFileStats, Default);
   processOptionImpl(PrintDispatchStats, Default);
   processOptionImpl(PrintSchedulerStats, Default);
   if (IsOutOfOrder)
     processOptionImpl(PrintRetireStats, Default);
 }
 
-std::unique_ptr<mca::InstrPostProcess>
-createInstrPostProcess(const Triple &TheTriple, const MCSubtargetInfo &STI,
-                       const MCInstrInfo &MCII) {
-  // Might be a good idea to have a separate flag so that InstrPostProcess
-  // can be used with or without CustomBehaviour
-  if (DisableCustomBehaviour)
-    return std::make_unique<mca::InstrPostProcess>(STI, MCII);
-#ifdef HAS_AMDGPU
-  if (TheTriple.isAMDGPU())
-    return std::make_unique<mca::AMDGPUInstrPostProcess>(STI, MCII);
-#endif
-  return std::make_unique<mca::InstrPostProcess>(STI, MCII);
-}
-
-std::unique_ptr<mca::CustomBehaviour>
-createCustomBehaviour(const Triple &TheTriple, const MCSubtargetInfo &STI,
-                      const mca::SourceMgr &SrcMgr, const MCInstrInfo &MCII) {
-  // Build the appropriate CustomBehaviour object for the current target.
-  // The CustomBehaviour class should never depend on the source code,
-  // but it can depend on the list of mca::Instruction and any classes
-  // that can be built using just the target info. If you need extra
-  // information from the source code or the list of MCInst, consider
-  // adding that information to the mca::Instruction class and setting
-  // it during InstrBuilder::createInstruction().
-  if (DisableCustomBehaviour)
-    return std::make_unique<mca::CustomBehaviour>(STI, SrcMgr, MCII);
-#ifdef HAS_AMDGPU
-  if (TheTriple.isAMDGPU())
-    return std::make_unique<mca::AMDGPUCustomBehaviour>(STI, SrcMgr, MCII);
-#endif
-  return std::make_unique<mca::CustomBehaviour>(STI, SrcMgr, MCII);
-}
-
 // Returns true on success.
 static bool runPipeline(mca::Pipeline &P) {
   // Handle pipeline errors here.
   Expected<unsigned> Cycles = P.run();
   if (!Cycles) {
     WithColor::error() << toString(Cycles.takeError());
     return false;
   }
   return true;
 }
 
 int main(int argc, char **argv) {
   InitLLVM X(argc, argv);
 
   // Initialize targets and assembly parsers.
   InitializeAllTargetInfos();
   InitializeAllTargetMCs();
   InitializeAllAsmParsers();
+  InitializeAllTargetMCAs();
 
   // Enable printing of available targets when flag --version is specified.
   cl::AddExtraVersionPrinter(TargetRegistry::printRegisteredTargetsForVersion);
 
   cl::HideUnrelatedOptions({&ToolOptions, &ViewOptions});
 
   // Parse flags and initialize target options.
   cl::ParseCommandLineOptions(argc, argv,
for (const std::unique_ptr<mca::CodeRegion> &Region : Regions) {
     if (Region->empty())
       continue;
 
     IB.clear();
 
     // Lower the MCInst sequence into an mca::Instruction sequence.
     ArrayRef<MCInst> Insts = Region->getInstructions();
     mca::CodeEmitter CE(*STI, *MAB, *MCE, Insts);
-    std::unique_ptr<mca::InstrPostProcess> IPP =
-        createInstrPostProcess(TheTriple, *STI, *MCII);
+
+    std::unique_ptr<mca::InstrPostProcess> IPP;
+    if (!DisableCustomBehaviour) {
+      IPP = std::unique_ptr<mca::InstrPostProcess>(
+          TheTarget->createInstrPostProcess(*STI, *MCII));
+    }
+    if (!IPP)
+      // If the target doesn't have its own IPP implemented (or the
+      // -disable-cb flag is set) then we use the base class
+      // (which does nothing).
+      IPP = std::make_unique<mca::InstrPostProcess>(*STI, *MCII);
+
     std::vector<std::unique_ptr<mca::Instruction>> LoweredSequence;
     for (const MCInst &MCI : Insts) {
       Expected<std::unique_ptr<mca::Instruction>> Inst =
           IB.createInstruction(MCI);
       if (!Inst) {
         if (auto NewE = handleErrors(
                 Inst.takeError(),
                 [&IP, &STI](const mca::InstructionError<MCInst> &IE) {
for (const std::unique_ptr<mca::CodeRegion> &Region : Regions) {
     }
 
     // Create the CustomBehaviour object for enforcing Target Specific
     // behaviours and dependencies that aren't expressed well enough
     // in the tablegen. CB cannot depend on the list of MCInst or
     // the source code (but it can depend on the list of
     // mca::Instruction or any objects that can be reconstructed
     // from the target information).
-    std::unique_ptr<mca::CustomBehaviour> CB =
-        createCustomBehaviour(TheTriple, *STI, S, *MCII);
+    std::unique_ptr<mca::CustomBehaviour> CB;
+    if (!DisableCustomBehaviour)
+      CB = std::unique_ptr<mca::CustomBehaviour>(
+          TheTarget->createCustomBehaviour(*STI, S, *MCII));
+    if (!CB)
+      // If the target doesn't have its own CB implemented (or the -disable-cb
+      // flag is set) then we use the base class (which does nothing).
+      CB = std::make_unique<mca::CustomBehaviour>(*STI, S, *MCII);
 
     // Create a basic pipeline simulating an out-of-order backend.
     auto P = MCA.createDefaultPipeline(PO, S, *CB);
 
     mca::PipelinePrinter Printer(*P, *Region, RegionIdx, *STI, PO);
 
     // When we output JSON, we add a view that contains the instructions
     // and CPU resource information.