Get Anvill building with LLVM 15

.github/workflows/build.yml

@@ -54,8 +54,8 @@ jobs:
           - { name: "ubuntu", tag: "20.04" }
         binja:
           - { channel: "headless", version: "3.1.3479" }
-        llvm: ["14"]
-        cxxcommon_version: ["v0.2.10"]
+        llvm: ["14", "15"]
+        cxxcommon_version: ["v0.2.16"]
 
     runs-on: ubuntu-20.04
     container:
@@ -432,8 +432,8 @@ jobs:
       fail-fast: false
       matrix:
         os: ["macos-11"]
-        llvm: ["14"]
-        cxxcommon_version: ["v0.2.10"]
+        llvm: ["14", "15"]
+        cxxcommon_version: ["v0.2.16"]
 
     runs-on: ${{ matrix.os }}
 
@@ -781,7 +781,7 @@ jobs:
 
     strategy:
       matrix:
-        llvm: ["14"]
+        llvm: ["14", "15"]
         ubuntu: ["20.04"]
         binja:
           - { channel: "headless", version: "3.1.3479" }

.remill_commit_id

@@ -1 +1 @@
-65817f05b86ed7c215ea5871325ff0461a467dea
+d5c50357287cae0072938477b3b525bafcce8069

Dockerfile

@@ -1,7 +1,7 @@
-ARG LLVM_VERSION=14
+ARG LLVM_VERSION=15
 ARG ARCH=amd64
 ARG UBUNTU_VERSION=20.04
-ARG CXX_COMMON_VERSION=0.2.10
+ARG CXX_COMMON_VERSION=0.2.16
 ARG DISTRO_BASE=ubuntu${UBUNTU_VERSION}
 ARG BUILD_BASE=ubuntu:${UBUNTU_VERSION}
 ARG LIBRARIES=/opt/trailofbits
@@ -13,7 +13,7 @@ ARG UBUNTU_VERSION
 ARG LIBRARIES
 ARG LLVM_VERSION
 ARG CXX_COMMON_VERSION
-ARG DEBIAN_FRONTEND=noninteractive 
+ARG DEBIAN_FRONTEND=noninteractive
 RUN apt-get update && \
     apt-get install -qqy --no-install-recommends git libdbus-1-3 curl unzip python3 python3-pip python3.8 python3.8-venv python3-setuptools xz-utils cmake && \
     rm -rf /var/lib/apt/lists/*
@@ -62,7 +62,7 @@ RUN cmake -G Ninja -B build -S  /dependencies/remill \
     -DCMAKE_VERBOSE_MAKEFILE=True \
     -DVCPKG_ROOT=/dependencies/vcpkg_ubuntu-${UBUNTU_VERSION}_llvm-${LLVM_VERSION}_amd64 \
     && \
-    cmake --build build --target install 
+    cmake --build build --target install
 
 # Source code build
 FROM deps AS build

bin/Decompile/Main.cpp

@@ -105,7 +105,9 @@ int main(int argc, char *argv[]) {
       remill::GetReference(maybe_buff);
 
   llvm::LLVMContext context;
+#if LLVM_VERSION_NUMBER < LLVM_VERSION(15, 0)
   context.enableOpaquePointers();
+#endif
   llvm::Module module("lifted_code", context);
 
   auto maybe_spec =

include/anvill/Transforms.h

@@ -190,26 +190,6 @@ void AddLowerRemillMemoryAccessIntrinsics(llvm::FunctionPassManager &fpm);
 // instructions.
 void AddLowerTypeHintIntrinsics(llvm::FunctionPassManager &fpm);
 
-//// Anvill-lifted bitcode operates at a very low level, swapping between integer
-//// and pointer representations. It is typically for just-lifted bitcode to
-//// perform integer arithmetic on addresses, then cast those integers into
-//// pointers in order to do a `load` or `store`. This happens because the bitcode
-//// we get from Remill uses memory access intrinsics, which abstract over the
-//// target program's address space and model memory loads/stores in terms of
-//// intrinsic function calls operating on integer addresses. When these intrinsic
-//// calls are lowered into `load` and `store` instructions by
-//// `LowerRemillMemoryAccessIntrinsics`, we are left with a mixed bag in integer
-//// arithmetic and then `inttoptr` casts.
-////
-//// Ideally, we want to comprehensively brighten all integer operations that
-//// produce pointers into pointer operations. For example, integer arithmetic
-//// should instead become `getelementptr` instructions, where possible, which
-//// model pointer arithmetic at a higher level.
-////
-//// This function attempts to apply a battery of pattern-based transforms to
-//// brighten integer operations into pointer operations.
-void AddConvertIntegerToPointerOperations(llvm::FunctionPassManager &fpm);
-
 // Transforms the bitcode to eliminate calls to `__remill_function_return`,
 // where appropriate. This will not succeed for all architectures, but is
 // likely to always succeed for x86(-64) and aarch64, due to their support

lib/CMakeLists.txt

@@ -31,7 +31,6 @@ protobuf_generate_cpp(
 # `.cpp` file.
 set(anvill_passes
   ConvertAddressesToEntityUses
-  ConvertIntegerToPointerOperations
   ConvertMasksToCasts
   ConvertSymbolicReturnAddressToConcreteReturnAddress
   ConvertXorsToCmps
@@ -217,7 +216,7 @@ if(ANVILL_ENABLE_INSTALL)
     PUBLIC_HEADER DESTINATION
       "${CMAKE_INSTALL_INCLUDEDIR}/anvill"
   )
-  
+
   install(
     FILES
       ${anvill_passes_HEADERS}

lib/Optimize.cpp

@@ -226,8 +226,6 @@ void OptimizeModule(const EntityLifter &lifter, llvm::Module &module) {
   // makes code easier to read and analyze. This is a fairly narrow optimization
   // but it comes up often enough for lifted code.
 
-  // TODO(alex): Need to rewrite this pass to somehow not rely on typed pointers.
-  // AddConvertIntegerToPointerOperations(fpm);
   AddConvertAddressesToEntityUses(fpm, xr, pc_metadata_id);
   AddBranchRecovery(fpm);