[RISCV] CodeGen of RVE and ilp32e/lp64e ABIs (llvm#76777)

This commit includes the necessary changes to clang and LLVM to support
codegen of `RVE` and the `ilp32e`/`lp64e` ABIs.

The differences between `RVE` and `RVI` are:
* `RVE` reduces the integer register count to 16(x0-x16).
* The ABI should be `ilp32e` for 32 bits and `lp64e` for 64 bits.

`RVE` can be combined with all current standard extensions.

The central changes in ilp32e/lp64e ABI, compared to ilp32/lp64 are:
* Only 6 integer argument registers (rather than 8).
* Only 2 callee-saved registers (rather than 12).
* A Stack Alignment of 32bits (rather than 128bits).
* ilp32e isn't compatible with D ISA extension.

If `ilp32e` or `lp64` is used with an ISA that has any of the registers
x16-x31 and f0-f31, then these registers are considered temporaries.

To be compatible with the implementation of ilp32e in GCC, we don't use
aligned registers to pass variadic arguments and set stack alignment\
to 4-bytes for types with length of 2*XLEN.

FastCC is also supported on RVE, while GHC isn't since there is only one
avaiable register.

Differential Revision: https://reviews.llvm.org/D70401

clang/docs/ReleaseNotes.rst

@@ -1060,6 +1060,8 @@ RISC-V Support
 ^^^^^^^^^^^^^^
 - Unaligned memory accesses can be toggled by ``-m[no-]unaligned-access`` or the
   aliases ``-m[no-]strict-align``.
+- CodeGen of RV32E/RV64E was supported experimentally.
+- CodeGen of ilp32e/lp64e was supported experimentally.
 
 - Default ABI with F but without D was changed to ilp32f for RV32 and to lp64f
   for RV64.

clang/lib/Basic/Targets/RISCV.cpp

@@ -154,7 +154,7 @@ void RISCVTargetInfo::getTargetDefines(const LangOptions &Opts,
   else
     Builder.defineMacro("__riscv_float_abi_soft");
 
-  if (ABIName == "ilp32e")
+  if (ABIName == "ilp32e" || ABIName == "lp64e")
     Builder.defineMacro("__riscv_abi_rve");
 
   Builder.defineMacro("__riscv_arch_test");
@@ -214,6 +214,13 @@ void RISCVTargetInfo::getTargetDefines(const LangOptions &Opts,
     Builder.defineMacro("__riscv_misaligned_fast");
   else
     Builder.defineMacro("__riscv_misaligned_avoid");
+
+  if (ISAInfo->hasExtension("e")) {
+    if (Is64Bit)
+      Builder.defineMacro("__riscv_64e");
+    else
+      Builder.defineMacro("__riscv_32e");
+  }
 }
 
 static constexpr Builtin::Info BuiltinInfo[] = {
@@ -378,6 +385,11 @@ bool RISCVTargetInfo::handleTargetFeatures(std::vector<std::string> &Features,
   if (llvm::is_contained(Features, "+experimental"))
     HasExperimental = true;
 
+  if (ABI == "ilp32e" && ISAInfo->hasExtension("d")) {
+    Diags.Report(diag::err_invalid_feature_combination)
+        << "ILP32E cannot be used with the D ISA extension";
+    return false;
+  }
   return true;
 }
 

clang/lib/Basic/Targets/RISCV.h

@@ -132,6 +132,12 @@ class LLVM_LIBRARY_VISIBILITY RISCV32TargetInfo : public RISCVTargetInfo {
   }
 
   bool setABI(const std::string &Name) override {
+    if (Name == "ilp32e") {
+      ABI = Name;
+      resetDataLayout("e-m:e-p:32:32-i64:64-n32-S32");
+      return true;
+    }
+
     if (Name == "ilp32" || Name == "ilp32f" || Name == "ilp32d") {
       ABI = Name;
       return true;
@@ -156,6 +162,12 @@ class LLVM_LIBRARY_VISIBILITY RISCV64TargetInfo : public RISCVTargetInfo {
   }
 
   bool setABI(const std::string &Name) override {
+    if (Name == "lp64e") {
+      ABI = Name;
+      resetDataLayout("e-m:e-p:64:64-i64:64-i128:128-n32:64-S64");
+      return true;
+    }
+
     if (Name == "lp64" || Name == "lp64f" || Name == "lp64d") {
       ABI = Name;
       return true;

clang/lib/CodeGen/CodeGenModule.cpp

@@ -229,7 +229,8 @@ createTargetCodeGenInfo(CodeGenModule &CGM) {
       ABIFLen = 32;
     else if (ABIStr.ends_with("d"))
       ABIFLen = 64;
-    return createRISCVTargetCodeGenInfo(CGM, XLen, ABIFLen);
+    bool EABI = ABIStr.ends_with("e");
+    return createRISCVTargetCodeGenInfo(CGM, XLen, ABIFLen, EABI);
   }
 
   case llvm::Triple::systemz: {

clang/lib/CodeGen/TargetInfo.h

@@ -496,7 +496,8 @@ createPPC64_SVR4_TargetCodeGenInfo(CodeGenModule &CGM, PPC64_SVR4_ABIKind Kind,
                                    bool SoftFloatABI);
 
 std::unique_ptr<TargetCodeGenInfo>
-createRISCVTargetCodeGenInfo(CodeGenModule &CGM, unsigned XLen, unsigned FLen);
+createRISCVTargetCodeGenInfo(CodeGenModule &CGM, unsigned XLen, unsigned FLen,
+                             bool EABI);
 
 std::unique_ptr<TargetCodeGenInfo>
 createCommonSPIRTargetCodeGenInfo(CodeGenModule &CGM);

clang/lib/CodeGen/Targets/RISCV.cpp

@@ -25,17 +25,20 @@ class RISCVABIInfo : public DefaultABIInfo {
   // ISA might have a wider FLen than the selected ABI (e.g. an RV32IF target
   // with soft float ABI has FLen==0).
   unsigned FLen;
-  static const int NumArgGPRs = 8;
-  static const int NumArgFPRs = 8;
+  const int NumArgGPRs;
+  const int NumArgFPRs;
+  const bool EABI;
   bool detectFPCCEligibleStructHelper(QualType Ty, CharUnits CurOff,
                                       llvm::Type *&Field1Ty,
                                       CharUnits &Field1Off,
                                       llvm::Type *&Field2Ty,
                                       CharUnits &Field2Off) const;
 
 public:
-  RISCVABIInfo(CodeGen::CodeGenTypes &CGT, unsigned XLen, unsigned FLen)
-      : DefaultABIInfo(CGT), XLen(XLen), FLen(FLen) {}
+  RISCVABIInfo(CodeGen::CodeGenTypes &CGT, unsigned XLen, unsigned FLen,
+               bool EABI)
+      : DefaultABIInfo(CGT), XLen(XLen), FLen(FLen), NumArgGPRs(EABI ? 6 : 8),
+        NumArgFPRs(FLen != 0 ? 8 : 0), EABI(EABI) {}
 
   // DefaultABIInfo's classifyReturnType and classifyArgumentType are
   // non-virtual, but computeInfo is virtual, so we overload it.
@@ -86,7 +89,7 @@ void RISCVABIInfo::computeInfo(CGFunctionInfo &FI) const {
   }
 
   int ArgGPRsLeft = IsRetIndirect ? NumArgGPRs - 1 : NumArgGPRs;
-  int ArgFPRsLeft = FLen ? NumArgFPRs : 0;
+  int ArgFPRsLeft = NumArgFPRs;
   int NumFixedArgs = FI.getNumRequiredArgs();
 
   int ArgNum = 0;
@@ -396,9 +399,12 @@ ABIArgInfo RISCVABIInfo::classifyArgumentType(QualType Ty, bool IsFixed,
   // Determine the number of GPRs needed to pass the current argument
   // according to the ABI. 2*XLen-aligned varargs are passed in "aligned"
   // register pairs, so may consume 3 registers.
+  // TODO: To be compatible with GCC's behaviors, we don't align registers
+  // currently if we are using ILP32E calling convention. This behavior may be
+  // changed when RV32E/ILP32E is ratified.
   int NeededArgGPRs = 1;
   if (!IsFixed && NeededAlign == 2 * XLen)
-    NeededArgGPRs = 2 + (ArgGPRsLeft % 2);
+    NeededArgGPRs = 2 + (EABI && XLen == 32 ? 0 : (ArgGPRsLeft % 2));
   else if (Size > XLen && Size <= 2 * XLen)
     NeededArgGPRs = 2;
 
@@ -480,6 +486,13 @@ Address RISCVABIInfo::EmitVAArg(CodeGenFunction &CGF, Address VAListAddr,
 
   auto TInfo = getContext().getTypeInfoInChars(Ty);
 
+  // TODO: To be compatible with GCC's behaviors, we force arguments with
+  // 2×XLEN-bit alignment and size at most 2×XLEN bits like `long long`,
+  // `unsigned long long` and `double` to have 4-byte alignment. This
+  // behavior may be changed when RV32E/ILP32E is ratified.
+  if (EABI && XLen == 32)
+    TInfo.Align = std::min(TInfo.Align, CharUnits::fromQuantity(4));
+
   // Arguments bigger than 2*Xlen bytes are passed indirectly.
   bool IsIndirect = TInfo.Width > 2 * SlotSize;
 
@@ -499,8 +512,9 @@ namespace {
 class RISCVTargetCodeGenInfo : public TargetCodeGenInfo {
 public:
   RISCVTargetCodeGenInfo(CodeGen::CodeGenTypes &CGT, unsigned XLen,
-                         unsigned FLen)
-      : TargetCodeGenInfo(std::make_unique<RISCVABIInfo>(CGT, XLen, FLen)) {}
+                         unsigned FLen, bool EABI)
+      : TargetCodeGenInfo(
+            std::make_unique<RISCVABIInfo>(CGT, XLen, FLen, EABI)) {}
 
   void setTargetAttributes(const Decl *D, llvm::GlobalValue *GV,
                            CodeGen::CodeGenModule &CGM) const override {
@@ -526,6 +540,7 @@ class RISCVTargetCodeGenInfo : public TargetCodeGenInfo {
 
 std::unique_ptr<TargetCodeGenInfo>
 CodeGen::createRISCVTargetCodeGenInfo(CodeGenModule &CGM, unsigned XLen,
-                                      unsigned FLen) {
-  return std::make_unique<RISCVTargetCodeGenInfo>(CGM.getTypes(), XLen, FLen);
+                                      unsigned FLen, bool EABI) {
+  return std::make_unique<RISCVTargetCodeGenInfo>(CGM.getTypes(), XLen, FLen,
+                                                  EABI);
 }

clang/lib/Driver/ToolChains/Arch/RISCV.cpp

@@ -210,6 +210,7 @@ StringRef riscv::getRISCVABI(const ArgList &Args, const llvm::Triple &Triple) {
   // rv32e -> ilp32e
   // rv32* -> ilp32
   // rv64g | rv64*d -> lp64d
+  // rv64e -> lp64e
   // rv64* -> lp64
   StringRef Arch = getRISCVArch(Args, Triple);
 
@@ -285,13 +286,16 @@ StringRef riscv::getRISCVArch(const llvm::opt::ArgList &Args,
   // 3. Choose a default based on `-mabi=`
   //
   // ilp32e -> rv32e
+  // lp64e -> rv64e
   // ilp32 | ilp32f | ilp32d -> rv32imafdc
   // lp64 | lp64f | lp64d -> rv64imafdc
   if (const Arg *A = Args.getLastArg(options::OPT_mabi_EQ)) {
     StringRef MABI = A->getValue();
 
     if (MABI.equals_insensitive("ilp32e"))
       return "rv32e";
+    else if (MABI.equals_insensitive("lp64e"))
+      return "rv64e";
     else if (MABI.starts_with_insensitive("ilp32"))
       return "rv32imafdc";
     else if (MABI.starts_with_insensitive("lp64")) {

clang/test/CodeGen/RISCV/riscv32-abi.c

@@ -5,6 +5,8 @@
 // RUN:   | FileCheck -check-prefixes=ILP32-ILP32F-ILP32D,ILP32F-ILP32D,ILP32-ILP32F,ILP32F %s
 // RUN: %clang_cc1 -triple riscv32 -target-feature +f -target-feature +d -target-abi ilp32d -emit-llvm %s -o - \
 // RUN:   | FileCheck -check-prefixes=ILP32-ILP32F-ILP32D,ILP32F-ILP32D,ILP32D %s
+// RUN: %clang_cc1 -triple riscv32 -emit-llvm -target-abi ilp32e %s -o - \
+// RUN:     | FileCheck -check-prefixes=ILP32-ILP32F-ILP32D,ILP32-ILP32F,ILP32,ILP32E %s
 
 #include <stddef.h>
 #include <stdint.h>
@@ -2064,4 +2066,5 @@ union float16_u f_ret_float16_u(void) {
 }
 
 //// NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+// ILP32E: {{.*}}
 // ILP32F: {{.*}}

clang/test/CodeGen/RISCV/riscv32-ilp32e-error.c

@@ -0,0 +1,4 @@
+// RUN: not %clang_cc1 -triple riscv32 -target-feature +d -emit-llvm -target-abi ilp32e %s 2>&1 \
+// RUN:     | FileCheck -check-prefix=ILP32E-WITH-FD %s
+
+// ILP32E-WITH-FD: error: invalid feature combination: ILP32E cannot be used with the D ISA extension