[TargetLoweringObjectFile] Handle riscv BE

[djtodoro]

Add DWARF exception handling support for riscv big-endian targets.
More CodeGen changes related to riscvbe are coming.

[llvmbot]

@llvm/pr-subscribers-backend-risc-v

Author: Djordje Todorovic (djtodoro)

Changes

Add DWARF exception handling support for riscv big-endian targets.
More CodeGen changes related to riscvbe are coming.

---

Full diff: https://github.com/llvm/llvm-project/pull/155166.diff

2 Files Affected:

• (modified) llvm/lib/CodeGen/TargetLoweringObjectFileImpl.cpp (+2)
• (modified) llvm/test/CodeGen/RISCV/dwarf-eh.ll (+16)

diff --git a/llvm/lib/CodeGen/TargetLoweringObjectFileImpl.cpp b/llvm/lib/CodeGen/TargetLoweringObjectFileImpl.cpp
index d19ef923ef740..739dcc7f4c868 100644
--- a/llvm/lib/CodeGen/TargetLoweringObjectFileImpl.cpp
+++ b/llvm/lib/CodeGen/TargetLoweringObjectFileImpl.cpp
@@ -247,6 +247,8 @@ void TargetLoweringObjectFileELF::Initialize(MCContext &Ctx,
     break;
   case Triple::riscv32:
   case Triple::riscv64:
+  case Triple::riscv32be:
+  case Triple::riscv64be:
     LSDAEncoding = dwarf::DW_EH_PE_pcrel | dwarf::DW_EH_PE_sdata4;
     PersonalityEncoding = dwarf::DW_EH_PE_indirect | dwarf::DW_EH_PE_pcrel |
                           dwarf::DW_EH_PE_sdata4;
diff --git a/llvm/test/CodeGen/RISCV/dwarf-eh.ll b/llvm/test/CodeGen/RISCV/dwarf-eh.ll
index 39cc0d52c9fe1..dd9e9e59206ac 100644
--- a/llvm/test/CodeGen/RISCV/dwarf-eh.ll
+++ b/llvm/test/CodeGen/RISCV/dwarf-eh.ll
@@ -14,6 +14,22 @@
 ; RUN:     | FileCheck %s
 ; RUN: llc -mtriple=riscv64 --code-model=medium -relocation-model=pic < %s \
 ; RUN:     | FileCheck %s
+; RUN: llc -mtriple=riscv32be --code-model=small  < %s \
+; RUN:     | FileCheck %s
+; RUN: llc -mtriple=riscv32be --code-model=medium < %s \
+; RUN:     | FileCheck %s
+; RUN: llc -mtriple=riscv32be --code-model=small  -relocation-model=pic < %s \
+; RUN:     | FileCheck %s
+; RUN: llc -mtriple=riscv32be --code-model=medium -relocation-model=pic < %s \
+; RUN:     | FileCheck %s
+; RUN: llc -mtriple=riscv64be --code-model=small  < %s \
+; RUN:     | FileCheck %s
+; RUN: llc -mtriple=riscv64be --code-model=medium < %s \
+; RUN:     | FileCheck %s
+; RUN: llc -mtriple=riscv64be --code-model=small  -relocation-model=pic < %s \
+; RUN:     | FileCheck %s
+; RUN: llc -mtriple=riscv64be --code-model=medium -relocation-model=pic < %s \
+; RUN:     | FileCheck %s
 
 declare void @throw_exception()
 

[lenary]

I'm a bit worries that this is the sort of place where a) big-endian data vs little-endian instructions will be a problem, and b) if we're using LLVM to both read and write in the test, and it is getting endianness wrong, then we won't necessarily see the issue in tests.

Have you separately cross-verified the reading and writing of dwarf data against binutils?

[djtodoro]

I'm a bit worries that this is the sort of place where a) big-endian data vs little-endian instructions will be a problem, and b) if we're using LLVM to both read and write in the test, and it is getting endianness wrong, then we won't necessarily see the issue in tests.

Have you separately cross-verified the reading and writing of dwarf data against binutils?

Yes, I checked it and it works. I am not sure if it can be incorporated into the llvm's testing infra somehow...