[llvm] fd84b1a - [M68k] Add new calling convention M68k_RTD

Min-Yih Hsu via llvm-commits llvm-commits at lists.llvm.org
Sun Oct 15 16:14:52 PDT 2023


Author: Min-Yih Hsu
Date: 2023-10-15T16:12:31-07:00
New Revision: fd84b1a99dfe37d4212be8afba2a93209679bc7f

URL: https://github.com/llvm/llvm-project/commit/fd84b1a99dfe37d4212be8afba2a93209679bc7f
DIFF: https://github.com/llvm/llvm-project/commit/fd84b1a99dfe37d4212be8afba2a93209679bc7f.diff

LOG: [M68k] Add new calling convention M68k_RTD

`M68k_RTD` is really similar to X86's stdcall, in which callee pops the
arguments from stack. In LLVM IR it can be written as `m68k_rtdcc`.
This patch also improves how ExpandPseudo Pass handles popping stack at
function returns in the absent of the RTD instruction.

Differential Revision: https://reviews.llvm.org/D149864

Added: 
    llvm/test/CodeGen/M68k/CConv/rtd-call.ll
    llvm/test/CodeGen/M68k/CConv/rtd-ret.ll

Modified: 
    llvm/include/llvm/AsmParser/LLToken.h
    llvm/include/llvm/IR/CallingConv.h
    llvm/lib/AsmParser/LLLexer.cpp
    llvm/lib/AsmParser/LLParser.cpp
    llvm/lib/IR/AsmWriter.cpp
    llvm/lib/Target/M68k/M68kExpandPseudo.cpp
    llvm/lib/Target/M68k/M68kISelLowering.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/include/llvm/AsmParser/LLToken.h b/llvm/include/llvm/AsmParser/LLToken.h
index 673dc58ce6451e3..2d6b8a19401d78d 100644
--- a/llvm/include/llvm/AsmParser/LLToken.h
+++ b/llvm/include/llvm/AsmParser/LLToken.h
@@ -175,6 +175,7 @@ enum Kind {
   kw_amdgpu_kernel,
   kw_amdgpu_gfx,
   kw_tailcc,
+  kw_m68k_rtdcc,
 
   // Attributes:
   kw_attributes,

diff  --git a/llvm/include/llvm/IR/CallingConv.h b/llvm/include/llvm/IR/CallingConv.h
index e97623b29f5230d..40222fa31d978bb 100644
--- a/llvm/include/llvm/IR/CallingConv.h
+++ b/llvm/include/llvm/IR/CallingConv.h
@@ -245,6 +245,9 @@ namespace CallingConv {
     /// placement. Preserves active lane values for input VGPRs.
     AMDGPU_CS_ChainPreserve = 105,
 
+    /// Used for M68k rtd-based CC (similar to X86's stdcall).
+    M68k_RTD = 106,
+
     /// The highest possible ID. Must be some 2^k - 1.
     MaxID = 1023
   };

diff  --git a/llvm/lib/AsmParser/LLLexer.cpp b/llvm/lib/AsmParser/LLLexer.cpp
index 466bdebc001f589..1402c152bb5c313 100644
--- a/llvm/lib/AsmParser/LLLexer.cpp
+++ b/llvm/lib/AsmParser/LLLexer.cpp
@@ -632,6 +632,7 @@ lltok::Kind LLLexer::LexIdentifier() {
   KEYWORD(amdgpu_kernel);
   KEYWORD(amdgpu_gfx);
   KEYWORD(tailcc);
+  KEYWORD(m68k_rtdcc);
 
   KEYWORD(cc);
   KEYWORD(c);

diff  --git a/llvm/lib/AsmParser/LLParser.cpp b/llvm/lib/AsmParser/LLParser.cpp
index 04eabc94cfc6abe..e104f8b3d1fdba5 100644
--- a/llvm/lib/AsmParser/LLParser.cpp
+++ b/llvm/lib/AsmParser/LLParser.cpp
@@ -1999,6 +1999,7 @@ void LLParser::parseOptionalDLLStorageClass(unsigned &Res) {
 ///   ::= 'amdgpu_cs_chain_preserve'
 ///   ::= 'amdgpu_kernel'
 ///   ::= 'tailcc'
+///   ::= 'm68k_rtdcc'
 ///   ::= 'cc' UINT
 ///
 bool LLParser::parseOptionalCallingConv(unsigned &CC) {
@@ -2067,6 +2068,7 @@ bool LLParser::parseOptionalCallingConv(unsigned &CC) {
     break;
   case lltok::kw_amdgpu_kernel:  CC = CallingConv::AMDGPU_KERNEL; break;
   case lltok::kw_tailcc:         CC = CallingConv::Tail; break;
+  case lltok::kw_m68k_rtdcc:     CC = CallingConv::M68k_RTD; break;
   case lltok::kw_cc: {
       Lex.Lex();
       return parseUInt32(CC);

diff  --git a/llvm/lib/IR/AsmWriter.cpp b/llvm/lib/IR/AsmWriter.cpp
index e190d82127908db..bd8b3e9ad52215e 100644
--- a/llvm/lib/IR/AsmWriter.cpp
+++ b/llvm/lib/IR/AsmWriter.cpp
@@ -350,6 +350,7 @@ static void PrintCallingConv(unsigned cc, raw_ostream &Out) {
     break;
   case CallingConv::AMDGPU_KERNEL: Out << "amdgpu_kernel"; break;
   case CallingConv::AMDGPU_Gfx:    Out << "amdgpu_gfx"; break;
+  case CallingConv::M68k_RTD:      Out << "m68k_rtdcc"; break;
   }
 }
 

diff  --git a/llvm/lib/Target/M68k/M68kExpandPseudo.cpp b/llvm/lib/Target/M68k/M68kExpandPseudo.cpp
index 2f60fc834a18e3d..13268d754a9dde6 100644
--- a/llvm/lib/Target/M68k/M68kExpandPseudo.cpp
+++ b/llvm/lib/Target/M68k/M68kExpandPseudo.cpp
@@ -258,32 +258,22 @@ bool M68kExpandPseudo::ExpandMI(MachineBasicBlock &MBB,
 
     if (StackAdj == 0) {
       MIB = BuildMI(MBB, MBBI, DL, TII->get(M68k::RTS));
-    } else if (isUInt<16>(StackAdj)) {
-
-      if (STI->atLeastM68020()) {
-        llvm_unreachable("RTD is not implemented");
-      } else {
-        // Copy PC from stack to a free address(A0 or A1) register
-        // TODO check if pseudo expand uses free address register
-        BuildMI(MBB, MBBI, DL, TII->get(M68k::MOV32aj), M68k::A1)
-            .addReg(M68k::SP);
+    } else {
+      // Copy return address from stack to a free address(A0 or A1) register
+      // TODO check if pseudo expand uses free address register
+      BuildMI(MBB, MBBI, DL, TII->get(M68k::MOV32aj), M68k::A1)
+          .addReg(M68k::SP);
 
-        // Adjust SP
-        FL->emitSPUpdate(MBB, MBBI, StackAdj, /*InEpilogue=*/true);
+      // Adjust SP
+      FL->emitSPUpdate(MBB, MBBI, StackAdj, /*InEpilogue=*/true);
 
-        // Put the return address on stack
-        BuildMI(MBB, MBBI, DL, TII->get(M68k::MOV32ja))
-            .addReg(M68k::SP)
-            .addReg(M68k::A1);
+      // Put the return address on stack
+      BuildMI(MBB, MBBI, DL, TII->get(M68k::MOV32ja))
+          .addReg(M68k::SP)
+          .addReg(M68k::A1);
 
-        // RTS
-        BuildMI(MBB, MBBI, DL, TII->get(M68k::RTS));
-      }
-    } else {
-      // TODO: RTD can only handle immediates as big as 2**16-1.
-      // If we need to pop off bytes before the return address, we
-      // must do it manually.
-      llvm_unreachable("Stack adjustment size not supported");
+      // RTS
+      BuildMI(MBB, MBBI, DL, TII->get(M68k::RTS));
     }
 
     // FIXME: Can rest of the operands be ignored, if there is any?

diff  --git a/llvm/lib/Target/M68k/M68kISelLowering.cpp b/llvm/lib/Target/M68k/M68kISelLowering.cpp
index d1ed26457fbcffb..0830cc7feb220d7 100644
--- a/llvm/lib/Target/M68k/M68kISelLowering.cpp
+++ b/llvm/lib/Target/M68k/M68kISelLowering.cpp
@@ -3050,9 +3050,8 @@ M68kTargetLowering::getRegForInlineAsmConstraint(const TargetRegisterInfo *TRI,
 
 /// Determines whether the callee is required to pop its own arguments.
 /// Callee pop is necessary to support tail calls.
-bool M68k::isCalleePop(CallingConv::ID CallingConv, bool IsVarArg,
-                       bool GuaranteeTCO) {
-  return false;
+bool M68k::isCalleePop(CallingConv::ID CC, bool IsVarArg, bool GuaranteeTCO) {
+  return CC == CallingConv::M68k_RTD && !IsVarArg;
 }
 
 // Return true if it is OK for this CMOV pseudo-opcode to be cascaded

diff  --git a/llvm/test/CodeGen/M68k/CConv/rtd-call.ll b/llvm/test/CodeGen/M68k/CConv/rtd-call.ll
new file mode 100644
index 000000000000000..56f36efbe0fb967
--- /dev/null
+++ b/llvm/test/CodeGen/M68k/CConv/rtd-call.ll
@@ -0,0 +1,48 @@
+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -mtriple=m68k %s -stop-after=finalize-isel -o - | FileCheck %s
+
+; We want to make sure caller doesn't pop the stack for callees using
+; the M68k_RTD CC. However, we've implemented some frame optimization
+; techniques to eliminate as many as frame setup/destroy instructions.
+; Therefore, to make test case small and concise, we check the MIR generated
+; after ISel instead.
+
+declare dso_local m68k_rtdcc void @callee(i32 noundef)
+declare dso_local m68k_rtdcc void @va_callee(i32 noundef, ...)
+
+define dso_local i32 @caller(ptr noundef %y) {
+  ; CHECK-LABEL: name: caller
+  ; CHECK: bb.0.entry:
+  ; CHECK-NEXT:   [[MOV32rp:%[0-9]+]]:ar32 = MOV32rp 0, %fixed-stack.0, implicit-def dead $ccr :: (load (s32) from %fixed-stack.0, align 8)
+  ; CHECK-NEXT:   [[MOV32rj:%[0-9]+]]:xr32 = MOV32rj killed [[MOV32rp]], implicit-def dead $ccr :: (load (s32) from %ir.y)
+  ; CHECK-NEXT:   ADJCALLSTACKDOWN 4, 0, implicit-def dead $sp, implicit-def dead $ccr, implicit $sp
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:ar32 = COPY $sp
+  ; CHECK-NEXT:   MOV32jr [[COPY]], [[MOV32rj]], implicit-def dead $ccr :: (store (s32) into stack, align 2)
+  ; CHECK-NEXT:   CALLb @callee, csr_std, implicit $sp, implicit-def $sp
+  ; CHECK-NEXT:   ADJCALLSTACKUP 4, 4, implicit-def dead $sp, implicit-def dead $ccr, implicit $sp
+  ; CHECK-NEXT:   $d0 = COPY [[MOV32rj]]
+  ; CHECK-NEXT:   RET 0, $d0
+entry:
+  %0 = load i32, ptr %y, align 4
+  call m68k_rtdcc void @callee(i32 noundef %0)
+  ret i32 %0
+}
+
+define dso_local i32 @va_caller(ptr noundef %y) {
+  ; CHECK-LABEL: name: va_caller
+  ; CHECK: bb.0.entry:
+  ; CHECK-NEXT:   [[MOV32rp:%[0-9]+]]:ar32 = MOV32rp 0, %fixed-stack.0, implicit-def dead $ccr :: (load (s32) from %fixed-stack.0, align 8)
+  ; CHECK-NEXT:   [[MOV32rj:%[0-9]+]]:xr32 = MOV32rj killed [[MOV32rp]], implicit-def dead $ccr :: (load (s32) from %ir.y)
+  ; CHECK-NEXT:   ADJCALLSTACKDOWN 4, 0, implicit-def dead $sp, implicit-def dead $ccr, implicit $sp
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:ar32 = COPY $sp
+  ; CHECK-NEXT:   MOV32jr [[COPY]], [[MOV32rj]], implicit-def dead $ccr :: (store (s32) into stack, align 2)
+  ; CHECK-NEXT:   CALLb @va_callee, csr_std, implicit $sp, implicit-def $sp
+  ; CHECK-NEXT:   ADJCALLSTACKUP 4, 0, implicit-def dead $sp, implicit-def dead $ccr, implicit $sp
+  ; CHECK-NEXT:   $d0 = COPY [[MOV32rj]]
+  ; CHECK-NEXT:   RET 0, $d0
+entry:
+  %0 = load i32, ptr %y, align 4
+  call m68k_rtdcc void (i32, ...) @va_callee(i32 noundef %0)
+  ret i32 %0
+}
+

diff  --git a/llvm/test/CodeGen/M68k/CConv/rtd-ret.ll b/llvm/test/CodeGen/M68k/CConv/rtd-ret.ll
new file mode 100644
index 000000000000000..2dc5f2812fcea88
--- /dev/null
+++ b/llvm/test/CodeGen/M68k/CConv/rtd-ret.ll
@@ -0,0 +1,31 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -mtriple=m68k < %s | FileCheck %s
+
+define dso_local m68k_rtdcc i32 @ret(i32 noundef %a, i32 noundef %b, i32 noundef %c) nounwind {
+; CHECK-LABEL: ret:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    move.l (8,%sp), %d0
+; CHECK-NEXT:    add.l (4,%sp), %d0
+; CHECK-NEXT:    add.l (12,%sp), %d0
+; CHECK-NEXT:    move.l (%sp), %a1
+; CHECK-NEXT:    adda.l #12, %sp
+; CHECK-NEXT:    move.l %a1, (%sp)
+; CHECK-NEXT:    rts
+entry:
+  %add = add nsw i32 %b, %a
+  %add1 = add nsw i32 %add, %c
+  ret i32 %add1
+}
+
+define dso_local m68k_rtdcc i32 @va_ret(i32 noundef %a, i32 noundef %b, i32 noundef %c, ...) nounwind {
+; CHECK-LABEL: va_ret:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    move.l (8,%sp), %d0
+; CHECK-NEXT:    add.l (4,%sp), %d0
+; CHECK-NEXT:    add.l (12,%sp), %d0
+; CHECK-NEXT:    rts
+entry:
+  %add = add nsw i32 %b, %a
+  %add1 = add nsw i32 %add, %c
+  ret i32 %add1
+}


        


More information about the llvm-commits mailing list