Starter bug hints: Checking endianness can be done given an IRGenModule &IGM by checking IGM.getModule()->getDataLayout().isBigEndian() or .isLittleEndian(), and bit width by IGM.SizeTy == IGM.Int64Ty or IGM.SizeTy == IGM.Int32Ty.

s390x is a big-endian 64-bit target maintained by IBM's Swift team. You should be able to use it as a test target. You can write IRGen tests for cross-compilation by providing a //RUN: line in the test file that specifies a target triple in the command line:

// RUN: %target-swift-frontend -parse-stdlib -emit-ir -target s390x-ibm-linux %s | %FileCheck --check-prefix=BE
// RUN: %target-swift-frontend -parse-stdlib -emit-ir -target x86_64-pc-linux %s | %FileCheck --check-prefix=LE

// BE: this will be checked for in big endian builds
// LE: this will be checked for in little endian builds

Note that I also included -parse-stdlib so the compiler won't try to load the standard library for the target platform, which might not be built on all hosts. You should be able to write tests that don't use anything from Swift's standard library, though.

Comment by Tapan Thaker (JIRA)

The places I found occurrences of _LITTLE_ENDIAN_ or _BIG_ENDIAN_ are the following:

• lib/IRGen/EnumPayload.cpp

• lib/IRGen/GenEnum.cpp

• stdlib/public/runtime/Enum.cpp

• stdlib/public/runtime/EnumImpl.h

• stdlib/public/stubs/MathStubs.cpp

• tools/swift-stdlib-tool/swift-stdlib-tool.mm

• unittests/runtime/Enum.cpp

If I understand correctly, it should be okay to have these macros in the runtime & is a problem only in lib/IRGen

You're correct. The compiler shouldn't change behavior based on host characteristics, but the runtime is compiled for the target(s).

Resetting assignee on all Starter Bugs last modified in 2018.