[SOL] Create explicit sign extension instruction (#114)

* Create explicit sign extension instruction
anza-xyz · Dec 3, 2024 · f602770 · f602770
1 parent a802634
commit f602770
Show file tree

Hide file tree

Showing 9 changed files with 196 additions and 20 deletions.
diff --git a/llvm/lib/Target/SBF/SBFISelLowering.cpp b/llvm/lib/Target/SBF/SBFISelLowering.cpp
@@ -933,24 +933,29 @@ SBFTargetLowering::EmitSubregExt(MachineInstr &MI, MachineBasicBlock *BB,
                                  unsigned Reg, bool isSigned) const {
   const TargetInstrInfo &TII = *BB->getParent()->getSubtarget().getInstrInfo();
   const TargetRegisterClass *RC = getRegClassFor(MVT::i64);
-  int RShiftOp = isSigned ? SBF::SRA_ri : SBF::SRL_ri;
   MachineFunction *F = BB->getParent();
   DebugLoc DL = MI.getDebugLoc();
 
   MachineRegisterInfo &RegInfo = F->getRegInfo();
 
   if (!isSigned) {
+    unsigned MovOp =
+        Subtarget->getHasExplicitSignExt()
+                         ? SBF::MOV_rr : SBF::MOV_32_64;
     Register PromotedReg0 = RegInfo.createVirtualRegister(RC);
-    BuildMI(BB, DL, TII.get(SBF::MOV_32_64), PromotedReg0).addReg(Reg);
+    BuildMI(BB, DL, TII.get(MovOp), PromotedReg0).addReg(Reg);
     return PromotedReg0;
   }
   Register PromotedReg0 = RegInfo.createVirtualRegister(RC);
+  BuildMI(BB, DL, TII.get(SBF::MOV_32_64), PromotedReg0).addReg(Reg);
+  if (Subtarget->getHasExplicitSignExt())
+    return PromotedReg0;
+
   Register PromotedReg1 = RegInfo.createVirtualRegister(RC);
   Register PromotedReg2 = RegInfo.createVirtualRegister(RC);
-  BuildMI(BB, DL, TII.get(SBF::MOV_32_64), PromotedReg0).addReg(Reg);
   BuildMI(BB, DL, TII.get(SBF::SLL_ri), PromotedReg1)
     .addReg(PromotedReg0).addImm(32);
-  BuildMI(BB, DL, TII.get(RShiftOp), PromotedReg2)
+  BuildMI(BB, DL, TII.get(SBF::SRA_ri), PromotedReg2)
     .addReg(PromotedReg1).addImm(32);
 
   return PromotedReg2;

diff --git a/llvm/lib/Target/SBF/SBFInstrInfo.cpp b/llvm/lib/Target/SBF/SBFInstrInfo.cpp
@@ -28,16 +28,24 @@ using namespace llvm;
 SBFInstrInfo::SBFInstrInfo()
     : SBFGenInstrInfo(SBF::ADJCALLSTACKDOWN, SBF::ADJCALLSTACKUP) {}
 
+void SBFInstrInfo::setHasExplicitSignExt(bool HasExplicitSext) {
+  this->HasExplicitSignExt = HasExplicitSext;
+}
+
 void SBFInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
                                MachineBasicBlock::iterator I,
                                const DebugLoc &DL, MCRegister DestReg,
                                MCRegister SrcReg, bool KillSrc) const {
   if (SBF::GPRRegClass.contains(DestReg, SrcReg))
     BuildMI(MBB, I, DL, get(SBF::MOV_rr), DestReg)
         .addReg(SrcReg, getKillRegState(KillSrc));
-  else if (SBF::GPR32RegClass.contains(DestReg, SrcReg))
-    BuildMI(MBB, I, DL, get(SBF::MOV_rr_32), DestReg)
+  else if (SBF::GPR32RegClass.contains(DestReg, SrcReg)) {
+    unsigned OpCode =
+        HasExplicitSignExt ? SBF::MOV_rr_32_no_sext_v2
+                           : SBF::MOV_rr_32_no_sext_v1;
+    BuildMI(MBB, I, DL, get(OpCode), DestReg)
         .addReg(SrcReg, getKillRegState(KillSrc));
+  }
   else
     llvm_unreachable("Impossible reg-to-reg copy");
 }

diff --git a/llvm/lib/Target/SBF/SBFInstrInfo.h b/llvm/lib/Target/SBF/SBFInstrInfo.h
@@ -58,9 +58,11 @@ class SBFInstrInfo : public SBFGenInstrInfo {
                         MachineBasicBlock *FBB, ArrayRef<MachineOperand> Cond,
                         const DebugLoc &DL,
                         int *BytesAdded = nullptr) const override;
+  void setHasExplicitSignExt(bool HasExplicitSext);
+
 private:
   void expandMEMCPY(MachineBasicBlock::iterator) const;
-
+  bool HasExplicitSignExt;
 };
 }
 

diff --git a/llvm/lib/Target/SBF/SBFInstrInfo.td b/llvm/lib/Target/SBF/SBFInstrInfo.td
@@ -65,6 +65,8 @@ def SBFNoCallxSrc : Predicate<"!Subtarget->getCallXRegSrc()">;
 def SBFPqrInstr : Predicate<"Subtarget->getHasPqrClass()">;
 def SBFNoPqrInstr : Predicate<"!Subtarget->getHasPqrClass()">;
 def SBFHasStoreImm : Predicate<"Subtarget->getHasStoreImm()">;
+def SBFExplicitSignExt : Predicate<"Subtarget->getHasExplicitSignExt()">;
+def SBFNoExplicitSignExt : Predicate<"!Subtarget->getHasExplicitSignExt()">;
 
 def brtarget : Operand<OtherVT> {
   let PrintMethod = "printBrTargetOperand";
@@ -415,16 +417,26 @@ def MOV_ri : MATH_RI<SBF_ALU64, SBF_MOV,
                     (ins i64imm:$imm),
                     "mov64 $dst, $imm",
                     [(set GPR:$dst, (i64 i64immSExt32:$imm))]>;
-def MOV_rr_32 : MATH_RR<SBF_ALU, SBF_MOV,
-                    (outs GPR32:$dst),
-                    (ins GPR32:$src),
-                    "mov32 $dst, $src",
-                    []>;
+
 def MOV_ri_32 : MATH_RI<SBF_ALU, SBF_MOV,
                     (outs GPR32:$dst),
                     (ins i32imm:$imm),
                     "mov32 $dst, $imm",
                     [(set GPR32:$dst, (i32 i32immSExt32:$imm))]>;
+
+let Predicates = [SBFExplicitSignExt], DecoderNamespace = "SBFv2" in {
+    def MOV_rr_32_no_sext_v2 : MATH_RR<SBF_ALU64, SBF_MOV,
+                                (outs GPR32:$dst),
+                                (ins GPR32:$src),
+                                "mov64 $dst, $src",
+                                []>;
+}
+
+def MOV_rr_32_no_sext_v1 : MATH_RR<SBF_ALU, SBF_MOV,
+                        (outs GPR32:$dst),
+                        (ins GPR32:$src),
+                        "mov32 $dst, $src",
+                        []>, Requires<[SBFNoExplicitSignExt]>;
 }
 
 def FI_ri
@@ -993,10 +1005,13 @@ let Constraints = "$dst = $src" in {
     }
 }
 
-let isCodeGenOnly = 1 in {
+let DecoderNamespace = "SBFv2" in {
   def MOV_32_64 : MATH_RR<SBF_ALU, SBF_MOV,
                          (outs GPR:$dst), (ins GPR32:$src),
                          "mov32 $dst, $src", []>;
+}
+
+let isCodeGenOnly = 1 in {
   def MOV_32_64_addr : MATH_RI<SBF_ALU, SBF_MOV,
                            (outs GPR:$dst), (ins u64imm:$imm),
                            "mov32 $dst, $imm", []>, Requires<[SBFNoLddw]>;
@@ -1023,19 +1038,39 @@ def : Pat<(SBFWrapper tglobaladdr:$in),
         (HOR_addr (MOV_32_64_addr tglobaladdr:$in),
                             tglobaladdr:$in)>, Requires<[SBFNoLddw]>;
 
+// SBFv2 sign extension
+def : Pat<(i64 (sext GPR32:$src)),
+          (MOV_32_64 GPR32:$src)>, Requires<[SBFExplicitSignExt]>;
 
+// SBFv1 sign extension
 def : Pat<(i64 (sext GPR32:$src)),
-          (SRA_ri (SLL_ri (MOV_32_64 GPR32:$src), 32), 32)>;
+          (SRA_ri (SLL_ri (MOV_32_64 GPR32:$src), 32), 32)>,
+                Requires<[SBFNoExplicitSignExt]>;
+
+// SBFv2 zero extension
+def : Pat<(i64 (zext GPR32:$src)), (INSERT_SUBREG (i64 (IMPLICIT_DEF)),
+                                    GPR32:$src, sub_32)>, Requires<[SBFExplicitSignExt]>;
 
-def : Pat<(i64 (zext GPR32:$src)), (MOV_32_64 GPR32:$src)>;
+// SBFv1 zero extension
+def : Pat<(i64 (zext GPR32:$src)), (MOV_32_64 GPR32:$src)>,
+                                Requires<[SBFNoExplicitSignExt]>;
 
-// For i64 -> i32 truncation, use the 32-bit subregister directly.
+// SBFv2 truncation
 def : Pat<(i32 (trunc GPR:$src)),
-          (i32 (EXTRACT_SUBREG GPR:$src, sub_32))>;
+          (AND_ri_32 (i32 (EXTRACT_SUBREG GPR:$src, sub_32)),
+          (i32 0xffffffff))>, Requires<[SBFExplicitSignExt]>;
+
+// SBFv1 truncation
+def : Pat<(i32 (trunc GPR:$src)),
+          (i32 (EXTRACT_SUBREG GPR:$src, sub_32))>, Requires<[SBFNoExplicitSignExt]>;
+
+// SBFv2 anyext
+def : Pat<(i64 (anyext GPR32:$src)),
+          (MOV_32_64 GPR32:$src)>, Requires<[SBFExplicitSignExt]>;
 
-// For i32 -> i64 anyext, we don't care about the high bits.
+// SBFv1 anyext
 def : Pat<(i64 (anyext GPR32:$src)),
-          (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>;
+          (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPR32:$src, sub_32)>, Requires<[SBFNoExplicitSignExt]>;
 
 class STORE32<SBFWidthModifer SizeOp, string Mnemonic, list<dag> Pattern>
     : TYPE_LD_ST<SBF_MEM.Value, SizeOp.Value,

diff --git a/llvm/lib/Target/SBF/SBFSubtarget.cpp b/llvm/lib/Target/SBF/SBFSubtarget.cpp
@@ -30,6 +30,7 @@ SBFSubtarget &SBFSubtarget::initializeSubtargetDependencies(const Triple &TT,
                                                             StringRef FS) {
   initializeEnvironment(TT);
   initSubtargetFeatures(CPU, FS);
+  InstrInfo.setHasExplicitSignExt(HasExplicitSignExt);
   return *this;
 }
 

diff --git a/llvm/lib/Target/SBF/SBFSubtarget.h b/llvm/lib/Target/SBF/SBFSubtarget.h
@@ -80,6 +80,9 @@ class SBFSubtarget : public SBFGenSubtargetInfo {
   // Whether we have store imm instructions
   bool HasStoreImm;
 
+  // Whether we have the explicit sign extension instruction (mov32)
+  bool HasExplicitSignExt;
+
 public:
   // This constructor initializes the data members to match that
   // of the specified triple.
@@ -105,6 +108,7 @@ class SBFSubtarget : public SBFGenSubtargetInfo {
     return HasDynamicFrames && NewCallConvention;
   }
   bool getHasStoreImm() const { return HasStoreImm; }
+  bool getHasExplicitSignExt() const { return HasExplicitSignExt; }
   const SBFInstrInfo *getInstrInfo() const override { return &InstrInfo; }
   const SBFFrameLowering *getFrameLowering() const override {
     return &FrameLowering;

diff --git a/llvm/lib/Target/SBF/SBFTargetFeatures.td b/llvm/lib/Target/SBF/SBFTargetFeatures.td
@@ -46,6 +46,9 @@ def FeatureCallConv : SubtargetFeature<"new-call-conv", "NewCallConvention", "tr
 def FeatureStoreImm : SubtargetFeature<"store-imm", "HasStoreImm", "true",
                                            "Enable store imm instructions">;
 
+def FeatureExplicitSext : SubtargetFeature<"explicit-sext", "HasExplicitSignExt",
+                    "true", "Enable the explicit sign extension instruction mov32">;
+
 class Proc<string Name, list<SubtargetFeature> Features>
  : Processor<Name, NoItineraries, Features>;
 
@@ -55,4 +58,4 @@ def : Proc<"v2", []>;
 def : Proc<"v3", [ALU32]>;
 def : Proc<"sbfv2", [FeatureDynamicFrames, FeatureRelocAbs64, FeatureStaticSyscalls,
                         FeatureDisableNeg, FeatureReverseSubImm, FeatureDisableLddw, FeatureCallxRegSrc,
-                          FeaturePqrInstr, FeatureCallConv]>;
+                          FeaturePqrInstr, FeatureCallConv, FeatureExplicitSext]>;
diff --git a/llvm/test/CodeGen/SBF/explicit-sign-ext.ll b/llvm/test/CodeGen/SBF/explicit-sign-ext.ll
@@ -0,0 +1,110 @@
+; RUN: llc -march=sbf -mattr=+alu32 < %s | FileCheck --check-prefixes=CHECK-v1,CHECK %s
+; RUN: llc -march=sbf -mattr=+alu32,+explicit-sext < %s | FileCheck --check-prefixes=CHECK-v2,CHECK %s
+
+
+define dso_local i64 @my_sext(i32 %a) local_unnamed_addr #0 {
+entry:
+; CHECK-LABEL: my_sext
+  %res = sext i32 %a to i64
+
+; CHECK-v1: mov32 r0, w1
+; CHECK-v1: lsh64 r0, 32
+; CHECK-v1: arsh64 r0, 32
+
+; CHECK-v2: mov32 r0, w1
+  ret i64 %res
+}
+
+define dso_local i64 @my_zext(i32 %a) local_unnamed_addr #0 {
+entry:
+; CHECK-LABEL: my_zext
+  %res = zext i32 %a to i64
+
+; CHECK-v1: mov32 r0, w1
+; CHECK-v2: mov64 w0, w1
+
+  ret i64 %res
+}
+
+define dso_local i32 @my_trunc(i64 %a) local_unnamed_addr #0 {
+entry:
+; CHECK-LABEL: my_trunc
+  %res = trunc i64 %a to i32
+
+; CHECK-v1: mov64 r0, r1
+; CHECK-v2: mov64 r0, r1
+; CHECK-v2: and32 w0, -1
+  ret i32 %res
+}
+
+define dso_local i32 @copy_phys(i32 %a) local_unnamed_addr #0 {
+entry:
+; CHECK-LABEL: copy_phys
+  %res = add i32 %a, 2
+  %b = sub i32 %res, 5
+; CHECK-v1: mov32 w0, w1
+; CHECK-v2: mov64 w0, w1
+  ret i32 %b
+}
+
+define dso_local i32 @select_cc_imm(i32 %a, i32 %c, i32 %d) local_unnamed_addr #0 {
+entry:
+; CHECK-LABEL: select_cc_imm
+  %cmp = icmp sgt i32 %a, 10
+; CHECK-v1: mov32 w0, w2
+; CHECK-v2: mov64 w0, w2
+
+; CHECK-v1: mov32 r1, w1
+; CHECK-v1: lsh64 r1, 32
+; CHECK-v1: arsh64 r1, 32
+; CHECK-v2: mov32 r1, w1
+
+  %c.d = select i1 %cmp, i32 %c, i32 %d
+; CHECK-v1: mov32 w0, w3
+; CHECK-v2: mov64 w0, w3
+
+  ret i32 %c.d
+}
+
+define dso_local i32 @select_cc_reg(i32 %a, i32 %b, i32 %c, i32 %d) local_unnamed_addr #0 {
+entry:
+; CHECK-LABEL: select_cc_reg
+  %cmp = icmp sgt i32 %a, %b
+; CHECK-v1: mov32 w0, w3
+; CHECK-v2: mov64 w0, w3
+
+; CHECK-v1: mov32 r1, w1
+; CHECK-v1: lsh64 r1, 32
+; CHECK-v1: arsh64 r1, 32
+; CHECK-v2: mov32 r1, w1
+
+; CHECK-v1: mov32 r2, w2
+; CHECK-v1: lsh64 r2, 32
+; CHECK-v1: arsh64 r2, 32
+; CHECK-v2: mov32 r2, w2
+
+  %c.d = select i1 %cmp, i32 %c, i32 %d
+; CHECK-v1: mov32 w0, w4
+; CHECK-v2: mov64 w0, w4
+  ret i32 %c.d
+}
+
+define dso_local i64 @select_cc_imm_64(i32 %a, i64 %c, i64 %d) local_unnamed_addr #0 {
+entry:
+; CHECK-LABEL: select_cc_imm_64
+; CHECK-v1: mov64 r0, r2
+  %cmp = icmp sgt i32 %a, 10
+  %c.d = select i1 %cmp, i64 %c, i64 %d
+; CHECK: mov64 r0, r3
+  ret i64 %c.d
+}
+
+define dso_local i64 @select_cc_reg_64(i32 %a, i32 %b, i64 %c, i64 %d) local_unnamed_addr #0 {
+entry:
+; CHECK-LABEL: select_cc_reg_64
+; CHECK: mov64 r0, r3
+  %cmp = icmp sgt i32 %a, %b
+  %c.d = select i1 %cmp, i64 %c, i64 %d
+; CHECK: mov64 r0, r4
+  ret i64 %c.d
+}
diff --git a/llvm/test/MC/SBF/sbf-alu.s b/llvm/test/MC/SBF/sbf-alu.s
@@ -291,6 +291,10 @@ mov64 r0, r9
 # CHECK-ASM-NEW: encoding: [0xbf,0x23,0x00,0x00,0x00,0x00,0x00,0x00]
 mov64 r3, r2
 
+# CHECK-OBJ-NEW: mov64 r3, r2
+# CHECK-ASM-NEW: encoding: [0xbf,0x23,0x00,0x00,0x00,0x00,0x00,0x00]
+mov64 w3, w2
+
 # CHECK-OBJ-NEW: mov64 r3, 0x7b
 # CHECK-ASM-NEW: encoding: [0xb7,0x03,0x00,0x00,0x7b,0x00,0x00,0x00]
 mov64 r3, 123
@@ -303,6 +307,10 @@ mov64 r5, -123
 # CHECK-ASM-NEW: encoding: [0xbc,0x26,0x00,0x00,0x00,0x00,0x00,0x00]
 mov32 w6, w2
 
+# CHECK-OBJ-NEW: mov32 w6, w2
+# CHECK-ASM-NEW: encoding: [0xbc,0x26,0x00,0x00,0x00,0x00,0x00,0x00]
+mov32 r6, w2
+
 # CHECK-OBJ-NEW: mov32 w5, -0x7b
 # CHECK-ASM-NEW: encoding: [0xb4,0x05,0x00,0x00,0x85,0xff,0xff,0xff]
 mov32 w5, -123