| Published: | |
| Last Modified: | |
| Tags: | llvm C++ VM |
When writing my first llvm backend at my last job I quickly found out that there is only very material assisting the creation of a custom backend. The posts and tutorials only cover the basics and stop as soon as it starts to become really interesting. This series of posts serves to fill the void left by others and to teach how to customice and tailor different aspects of a backend to your needs.
Updates
| When | What |
|---|---|
| 2024-03-21 | Initial publication targeting llvm-16.0.6 |
| 2025-02-06 | Updated the diff to target llvm-19.1.7 |
Setting up the environment
Before we can get startet with writing our very own llvm backend we need to setup a couple of things first.
As we are writing a llvm backend the first thing we need is llvm itself. It can be obtained from github (llvm-project).
You should checkout a reasonably new stable release. Next you want to run cmake from a separate build directory.
I prefer the Ninja build system as parallel builds can easily be restricted when you are tight on RAM and it automatically runs jobs in parallel.
For LLVM adding -DLLVM_PARALLEL_LINK_JOBS=2 can help if you run into the oom killer.
git clone -b llvmorg-17.0.6 https://github.com/llvm/llvm-project.git
mkdir llvm-build
cd llvm-build
cmake -DLLVM_ENABLE_PROJECTS="lld;clang" \
-DLLVM_TARGETS_TO_BUILD="ARM;AArch64;X86" \
-DLLVM_ENABLE_ASSERTIONS=On \
-DCMAKE_BUILD_TYPE=Release \
-G Ninja \
../llvm-project/llvm
ninja
After making sure llvm can be build without any modifications create a new directory with the name
of your backend in llvm/lib/Target. I called mine LEG, so I created the direcotry llvm/lib/Target/LEG. Next a couple of files need to be created or modified. Please substitute your backend name for LEG in filenames, pathes, and file content.
diff --git a/llvm/include/llvm/BinaryFormat/ELF.h b/llvm/include/llvm/BinaryFormat/ELF.h
index 456cffff6..cb8948901 100644
--- a/llvm/include/llvm/BinaryFormat/ELF.h
+++ b/llvm/include/llvm/BinaryFormat/ELF.h
@@ -321,6 +321,7 @@ enum {
EM_VE = 251, // NEC SX-Aurora VE
EM_CSKY = 252, // C-SKY 32-bit processor
EM_LOONGARCH = 258, // LoongArch
+ EM_LEG = 512 // Ieads LEG architecture
};
// Object file classes.
@@ -1018,6 +1019,11 @@ enum {
#include "ELFRelocs/Xtensa.def"
};
+// ELF Relocation for LEG
+enum {
+#include "ELFRelocs/LEG.def"
+};
+
#undef ELF_RELOC
// Section header.
diff --git a/llvm/include/llvm/BinaryFormat/ELFRelocs/LEG.def b/llvm/include/llvm/BinaryFormat/ELFRelocs/LEG.def
new file mode 100644
index 000000000..23f71e57e
--- /dev/null
+++ b/llvm/include/llvm/BinaryFormat/ELFRelocs/LEG.def
@@ -0,0 +1,8 @@
+#ifndef ELF_RELOC
+#error "ELF_RELOC must be defined"
+#endif
+
+ELF_RELOC(R_LEG_NONE, 0)
+ELF_RELOC(R_LEG_IMM64, 1)
+ELF_RELOC(R_LEG_PCREL_BR22, 2)
+ELF_RELOC(R_LEG_PCREL_BR27, 3)
\ No newline at end of file
diff --git a/llvm/include/llvm/Object/ELFObjectFile.h b/llvm/include/llvm/Object/ELFObjectFile.h
index 811943dcd..4c61fd058 100644
--- a/llvm/include/llvm/Object/ELFObjectFile.h
+++ b/llvm/include/llvm/Object/ELFObjectFile.h
@@ -1348,6 +1348,8 @@ StringRef ELFObjectFile<ELFT>::getFileFormatName() const {
return "elf64-ve";
case ELF::EM_LOONGARCH:
return "elf64-loongarch";
+ case ELF::EM_LEG:
+ return "elf64-leg";
default:
return "elf64-unknown";
}
@@ -1401,6 +1403,8 @@ template <class ELFT> Triple::ArchType ELFObjectFile<ELFT>::getArch() const {
default:
report_fatal_error("Invalid ELFCLASS!");
}
+ case ELF::EM_LEG:
+ return Triple::leg;
case ELF::EM_S390:
return Triple::systemz;
diff --git a/llvm/include/llvm/TargetParser/Triple.h b/llvm/include/llvm/TargetParser/Triple.h
index d2126a03d..95fa926e7 100644
--- a/llvm/include/llvm/TargetParser/Triple.h
+++ b/llvm/include/llvm/TargetParser/Triple.h
@@ -74,6 +74,7 @@ public:
amdgcn, // AMDGCN: AMD GCN GPUs
riscv32, // RISC-V (32-bit): riscv32
riscv64, // RISC-V (64-bit): riscv64
+ leg, // Ieads LEG architecture
sparc, // Sparc: sparc
sparcv9, // Sparcv9: Sparcv9
sparcel, // Sparc: (endianness = little). NB: 'Sparcle' is a CPU variant
diff --git a/llvm/lib/BinaryFormat/ELF.cpp b/llvm/lib/BinaryFormat/ELF.cpp
index 9878f5769..e58fb5bf7 100644
--- a/llvm/lib/BinaryFormat/ELF.cpp
+++ b/llvm/lib/BinaryFormat/ELF.cpp
@@ -192,6 +192,7 @@ uint16_t ELF::convertArchNameToEMachine(StringRef Arch) {
.Case("csr_kalimba", EM_CSR_KALIMBA)
.Case("amdgpu", EM_AMDGPU)
.Case("riscv", EM_RISCV)
+ .Case("leg", EM_LEG)
.Case("lanai", EM_LANAI)
.Case("bpf", EM_BPF)
.Case("ve", EM_VE)
@@ -553,6 +554,8 @@ StringRef ELF::convertEMachineToArchName(uint16_t EMachine) {
return "amdgpu";
case EM_RISCV:
return "riscv";
+ case EM_LEG:
+ return "leg";
case EM_LANAI:
return "lanai";
case EM_BPF:
diff --git a/llvm/lib/Object/ELF.cpp b/llvm/lib/Object/ELF.cpp
index e47a40b87..a93104ddc 100644
--- a/llvm/lib/Object/ELF.cpp
+++ b/llvm/lib/Object/ELF.cpp
@@ -116,6 +116,13 @@ StringRef llvm::object::getELFRelocationTypeName(uint32_t Machine,
break;
}
break;
+ case ELF::EM_LEG:
+ switch (Type) {
+#include "llvm/BinaryFormat/ELFRelocs/LEG.def"
+ default:
+ break;
+ }
+ break;
case ELF::EM_S390:
switch (Type) {
#include "llvm/BinaryFormat/ELFRelocs/SystemZ.def"
diff --git a/llvm/lib/TargetParser/Triple.cpp b/llvm/lib/TargetParser/Triple.cpp
index 21d6c74b5..3109e7a27 100644
--- a/llvm/lib/TargetParser/Triple.cpp
+++ b/llvm/lib/TargetParser/Triple.cpp
@@ -65,6 +65,7 @@ StringRef Triple::getArchTypeName(ArchType Kind) {
case renderscript64: return "renderscript64";
case riscv32: return "riscv32";
case riscv64: return "riscv64";
+ case leg: return "leg";
case shave: return "shave";
case sparc: return "sparc";
case sparcel: return "sparcel";
@@ -224,6 +225,8 @@ StringRef Triple::getArchTypePrefix(ArchType Kind) {
case riscv32:
case riscv64: return "riscv";
+ case leg: return "leg";
+
case ve: return "ve";
case csky: return "csky";
@@ -421,6 +424,7 @@ Triple::ArchType Triple::getArchTypeForLLVMName(StringRef Name) {
.Case("amdgcn", amdgcn)
.Case("riscv32", riscv32)
.Case("riscv64", riscv64)
+ .Case("leg", leg)
.Case("hexagon", hexagon)
.Case("sparc", sparc)
.Case("sparcel", sparcel)
@@ -569,6 +573,7 @@ static Triple::ArchType parseArch(StringRef ArchName) {
.Case("amdgcn", Triple::amdgcn)
.Case("riscv32", Triple::riscv32)
.Case("riscv64", Triple::riscv64)
+ .Case("leg", Triple::leg)
.Case("hexagon", Triple::hexagon)
.Cases("s390x", "systemz", Triple::systemz)
.Case("sparc", Triple::sparc)
@@ -933,6 +938,7 @@ static Triple::ObjectFormatType getDefaultFormat(const Triple &T) {
case Triple::renderscript64:
case Triple::riscv32:
case Triple::riscv64:
+ case Triple::leg:
case Triple::shave:
case Triple::sparc:
case Triple::sparcel:
@@ -1654,6 +1660,7 @@ unsigned Triple::getArchPointerBitWidth(llvm::Triple::ArchType Arch) {
case llvm::Triple::ppc64le:
case llvm::Triple::renderscript64:
case llvm::Triple::riscv64:
+ case llvm::Triple::leg:
case llvm::Triple::sparcv9:
case llvm::Triple::spirv:
case llvm::Triple::spir64:
@@ -1685,6 +1692,7 @@ Triple Triple::get32BitArchVariant() const {
case Triple::UnknownArch:
case Triple::amdgcn:
case Triple::avr:
+ case Triple::leg:
case Triple::bpfeb:
case Triple::bpfel:
case Triple::msp430:
@@ -1799,6 +1807,7 @@ Triple Triple::get64BitArchVariant() const {
case Triple::ppc64le:
case Triple::renderscript64:
case Triple::riscv64:
+ case Triple::leg:
case Triple::sparcv9:
case Triple::spir64:
case Triple::spirv64:
@@ -1869,6 +1878,7 @@ Triple Triple::getBigEndianArchVariant() const {
case Triple::renderscript64:
case Triple::riscv32:
case Triple::riscv64:
+ case Triple::leg:
case Triple::shave:
case Triple::spir64:
case Triple::spir:
@@ -1978,6 +1988,7 @@ bool Triple::isLittleEndian() const {
case Triple::renderscript64:
case Triple::riscv32:
case Triple::riscv64:
+ case Triple::leg:
case Triple::shave:
case Triple::sparcel:
case Triple::spir64: