[EVM] Add support for CLZ instruction

vladimirradosavljevic · akiramenai · commit 8e7767fdb8a1 · 2025-10-15T01:23:31.000+02:00
CLZ will be introduced in the fusaka update: https://eips.ethereum.org/EIPS/eip-7939 Beside adding support for CLZ instruction in this patch, support for EVM versioning is added, so CLZ instruction is generated for fusaka and later versions. isCheapToSpeculateCttz and isCheapToSpeculateCtlz are implemented so CGP doesn't generate check for zero to avoid calling the intrinsic in despeculateCountZeros. CTZ is generated via CLZ, and currently LLVM is generating: `256 - clz(~x & (x-1))` If x != 0 is proven, this can be optimized to: `255 - clz(x & -x)` (#918) Signed-off-by: Vladimir Radosavljevic <vr@matterlabs.dev>
diff --git a/llvm/lib/Target/EVM/Disassembler/EVMDisassembler.cpp b/llvm/lib/Target/EVM/Disassembler/EVMDisassembler.cpp
@@ -14,6 +14,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "MCTargetDesc/EVMMCExpr.h"
+#include "MCTargetDesc/EVMMCTargetDesc.h"
 #include "TargetInfo/EVMTargetInfo.h"
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/MC/MCContext.h"
diff --git a/llvm/lib/Target/EVM/EVM.td b/llvm/lib/Target/EVM/EVM.td
@@ -16,6 +16,13 @@
 
 include "llvm/Target/Target.td"
 
+//===----------------------------------------------------------------------===//
+// EVM Versions
+//===----------------------------------------------------------------------===//
+
+def FeatureFusaka : SubtargetFeature<"fusaka", "Version", "EVMVersion::Fusaka",
+                                     "Fusaka release">;
+def HasCLZ : Predicate<"Subtarget->hasCLZ()">;
 
 //===----------------------------------------------------------------------===//
 // Register File Description
diff --git a/llvm/lib/Target/EVM/EVMISelDAGToDAG.cpp b/llvm/lib/Target/EVM/EVMISelDAGToDAG.cpp
@@ -26,10 +26,17 @@ using namespace llvm;
 
 namespace {
 class EVMDAGToDAGISel final : public SelectionDAGISel {
+  const EVMSubtarget *Subtarget = nullptr;
+
 public:
   EVMDAGToDAGISel(EVMTargetMachine &TM, CodeGenOptLevel OptLevel)
       : SelectionDAGISel(TM, OptLevel) {}
 
+  bool runOnMachineFunction(MachineFunction &MF) override {
+    Subtarget = &MF.getSubtarget<EVMSubtarget>();
+    return SelectionDAGISel::runOnMachineFunction(MF);
+  }
+
 private:
 
   // Include the pieces autogenerated from the target description.
diff --git a/llvm/lib/Target/EVM/EVMISelLowering.cpp b/llvm/lib/Target/EVM/EVMISelLowering.cpp
@@ -54,6 +54,9 @@ EVMTargetLowering::EVMTargetLowering(const TargetMachine &TM,
                       ISD::FrameIndex},
                      MVT::i256, Legal);
 
+  if (Subtarget->hasCLZ())
+    setOperationAction(ISD::CTLZ, MVT::i256, Legal);
+
   for (auto CC : {ISD::SETULT, ISD::SETUGT, ISD::SETLT, ISD::SETGT, ISD::SETGE,
                   ISD::SETUGE, ISD::SETLE, ISD::SETULE, ISD::SETEQ, ISD::SETNE})
     setCondCodeAction(CC, MVT::i256, Legal);
@@ -129,6 +132,14 @@ bool EVMTargetLowering::isLegalAddressingMode(const DataLayout &DL,
   return true;
 }
 
+bool EVMTargetLowering::isCheapToSpeculateCtlz(Type *Ty) const {
+  return Subtarget->hasCLZ();
+}
+
+bool EVMTargetLowering::isCheapToSpeculateCttz(Type *Ty) const {
+  return Subtarget->hasCLZ();
+}
+
 //===----------------------------------------------------------------------===//
 // EVM Lowering private implementation.
 //===----------------------------------------------------------------------===//
diff --git a/llvm/lib/Target/EVM/EVMISelLowering.h b/llvm/lib/Target/EVM/EVMISelLowering.h
@@ -49,6 +49,10 @@ class EVMTargetLowering final : public TargetLowering {
     return MVT::i256;
   }
 
+  bool isCheapToSpeculateCttz(Type *Ty) const override;
+
+  bool isCheapToSpeculateCtlz(Type *Ty) const override;
+
   /// Return true if it is profitable to move this shift by a constant amount
   /// through its operand, adjusting any immediate operands as necessary to
   /// preserve semantics. This transformation may not be desirable if it
diff --git a/llvm/lib/Target/EVM/EVMInstrInfo.td b/llvm/lib/Target/EVM/EVMInstrInfo.td
@@ -357,6 +357,11 @@ defm NOT
   : I<(outs GPR:$dst), (ins GPR:$src), [(set GPR:$dst, (not GPR:$src))],
       "NOT", "$dst, $src", 0x19, 3>;
 
+let Predicates = [HasCLZ] in
+defm CLZ
+  : I<(outs GPR:$dst), (ins GPR:$src), [(set GPR:$dst, (ctlz GPR:$src))],
+      "CLZ", "$dst, $src", 0x1e, 5>;
+
 let mayLoad = 1 in
 defm KECCAK256
   : I<(outs GPR:$dst), (ins GPR:$offset, GPR:$size),
diff --git a/llvm/lib/Target/EVM/EVMSubtarget.cpp b/llvm/lib/Target/EVM/EVMSubtarget.cpp
@@ -18,10 +18,18 @@ using namespace llvm;
 
 #define GET_SUBTARGETINFO_TARGET_DESC
 #define GET_SUBTARGETINFO_CTOR
+#define GET_SUBTARGETINFO_ENUM
 #include "EVMGenSubtargetInfo.inc"
 
+EVMSubtarget &EVMSubtarget::initializeSubtargetDependencies(StringRef CPU,
+                                                            StringRef FS) {
+  ParseSubtargetFeatures(CPU, /*TuneCPU*/ CPU, FS);
+  return *this;
+}
+
 EVMSubtarget::EVMSubtarget(const Triple &TT, const std::string &CPU,
                            const std::string &FS, const TargetMachine &TM)
-    : EVMGenSubtargetInfo(TT, CPU, /*TuneCPU*/ CPU, FS), TLInfo(TM, *this) {}
+    : EVMGenSubtargetInfo(TT, CPU, /*TuneCPU*/ CPU, FS),
+      TLInfo(TM, initializeSubtargetDependencies(CPU, FS)) {}
 
 bool EVMSubtarget::useAA() const { return true; }
diff --git a/llvm/lib/Target/EVM/EVMSubtarget.h b/llvm/lib/Target/EVM/EVMSubtarget.h
@@ -19,7 +19,6 @@
 #include "EVMRegisterInfo.h"
 #include "llvm/CodeGen/TargetSubtargetInfo.h"
 
-#define GET_SUBTARGETINFO_ENUM
 #define GET_SUBTARGETINFO_HEADER
 #include "EVMGenSubtargetInfo.inc"
 
@@ -28,10 +27,18 @@ class StringRef;
 
 class EVMSubtarget final : public EVMGenSubtargetInfo {
 private:
+  enum class EVMVersion : uint8_t {
+    Generic = 0, // Older versions of the EVM.
+    Fusaka,
+  };
+  EVMVersion Version = EVMVersion::Generic;
+
   EVMFrameLowering FrameLowering;
   EVMInstrInfo InstrInfo;
   EVMTargetLowering TLInfo;
 
+  EVMSubtarget &initializeSubtargetDependencies(StringRef CPU, StringRef FS);
+
 public:
   // This constructor initializes the data members to match that
   // of the specified triple.
@@ -57,6 +64,8 @@ class EVMSubtarget final : public EVMGenSubtargetInfo {
   bool useAA() const override;
 
   unsigned stackDepthLimit() const { return 16; }
+
+  bool hasCLZ() const { return Version >= EVMVersion::Fusaka; }
 };
 } // namespace llvm
 
diff --git a/llvm/lib/Target/EVM/MCTargetDesc/EVMMCTargetDesc.h b/llvm/lib/Target/EVM/MCTargetDesc/EVMMCTargetDesc.h
@@ -65,4 +65,7 @@ std::string getImmutableId(StringRef Name);
 #define GET_INSTRINFO_MC_HELPER_DECLS
 #include "EVMGenInstrInfo.inc"
 
+#define GET_SUBTARGETINFO_ENUM
+#include "EVMGenSubtargetInfo.inc"
+
 #endif // LLVM_LIB_TARGET_EVM_MCTARGETDESC_EVMMCTARGETDESC_H
diff --git a/llvm/test/CodeGen/EVM/bitmanipulation-intrinsics.ll b/llvm/test/CodeGen/EVM/bitmanipulation-intrinsics.ll
@@ -7,8 +7,6 @@ target triple = "evm"
 declare i256 @llvm.bitreverse.i256(i256)
 declare i256 @llvm.bswap.i256(i256)
 declare i256 @llvm.ctpop.i256(i256)
-declare i256 @llvm.ctlz.i256(i256, i1)
-declare i256 @llvm.cttz.i256(i256, i1)
 
 define i256 @bitreversetest(i256 %v) {
 ; CHECK-LABEL: bitreversetest:
@@ -583,188 +581,3 @@ define i256 @ctpoptest(i256 %v) {
   %res = call i256 @llvm.ctpop.i256(i256 %v)
   ret i256 %res
 }
-
-define i256 @ctlztest(i256 %v) {
-; CHECK-LABEL: ctlztest:
-; CHECK:       ; %bb.0:
-; CHECK-NEXT:    JUMPDEST
-; CHECK-NEXT:    PUSH32 0x101010101010101010101010101010101010101010101010101010101010101
-; CHECK-NEXT:    PUSH16 0xF0F0F0F0F0F0F0F0F0F0F0F0F0F0F0F
-; CHECK-NEXT:    DUP2
-; CHECK-NEXT:    DUP2
-; CHECK-NEXT:    PUSH32 0x3333333333333333333333333333333333333333333333333333333333333333
-; CHECK-NEXT:    DUP6
-; CHECK-NEXT:    PUSH16 0x55555555555555555555555555555555
-; CHECK-NEXT:    SWAP7
-; CHECK-NEXT:    PUSH1 0x1
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    OR
-; CHECK-NEXT:    DUP1
-; CHECK-NEXT:    PUSH1 0x2
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    OR
-; CHECK-NEXT:    DUP1
-; CHECK-NEXT:    PUSH1 0x4
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    OR
-; CHECK-NEXT:    DUP1
-; CHECK-NEXT:    PUSH1 0x8
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    OR
-; CHECK-NEXT:    DUP1
-; CHECK-NEXT:    PUSH1 0x10
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    OR
-; CHECK-NEXT:    DUP1
-; CHECK-NEXT:    PUSH1 0x20
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    OR
-; CHECK-NEXT:    DUP1
-; CHECK-NEXT:    PUSH1 0x40
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    OR
-; CHECK-NEXT:    DUP1
-; CHECK-NEXT:    PUSH1 0x80
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    OR
-; CHECK-NEXT:    NOT
-; CHECK-NEXT:    PUSH16 0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF
-; CHECK-NEXT:    DUP8
-; CHECK-NEXT:    DUP3
-; CHECK-NEXT:    PUSH1 0x81
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    AND
-; CHECK-NEXT:    DUP3
-; CHECK-NEXT:    PUSH1 0x80
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    SUB
-; CHECK-NEXT:    DUP4
-; CHECK-NEXT:    DUP1
-; CHECK-NEXT:    DUP3
-; CHECK-NEXT:    PUSH1 0x2
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    AND
-; CHECK-NEXT:    SWAP2
-; CHECK-NEXT:    AND
-; CHECK-NEXT:    ADD
-; CHECK-NEXT:    SWAP8
-; CHECK-NEXT:    DUP3
-; CHECK-NEXT:    PUSH1 0x1
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    AND
-; CHECK-NEXT:    SWAP2
-; CHECK-NEXT:    AND
-; CHECK-NEXT:    SUB
-; CHECK-NEXT:    SWAP1
-; CHECK-NEXT:    DUP1
-; CHECK-NEXT:    DUP3
-; CHECK-NEXT:    PUSH1 0x2
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    AND
-; CHECK-NEXT:    SWAP2
-; CHECK-NEXT:    AND
-; CHECK-NEXT:    ADD
-; CHECK-NEXT:    SWAP5
-; CHECK-NEXT:    DUP1
-; CHECK-NEXT:    PUSH1 0x4
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    ADD
-; CHECK-NEXT:    AND
-; CHECK-NEXT:    MUL
-; CHECK-NEXT:    PUSH1 0x10
-; CHECK-NEXT:    BYTE
-; CHECK-NEXT:    SWAP3
-; CHECK-NEXT:    DUP1
-; CHECK-NEXT:    PUSH1 0x4
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    ADD
-; CHECK-NEXT:    AND
-; CHECK-NEXT:    MUL
-; CHECK-NEXT:    PUSH1 0x10
-; CHECK-NEXT:    BYTE
-; CHECK-NEXT:    ADD
-; CHECK-NEXT:    SWAP1
-; CHECK-NEXT:    JUMP
-
-  %res = call i256 @llvm.ctlz.i256(i256 %v, i1 false)
-  ret i256 %res
-}
-
-define i256 @cttztest(i256 %v) {
-; CHECK-LABEL: cttztest:
-; CHECK:       ; %bb.0:
-; CHECK-NEXT:    JUMPDEST
-; CHECK-NEXT:    PUSH32 0x101010101010101010101010101010101010101010101010101010101010101
-; CHECK-NEXT:    PUSH16 0xF0F0F0F0F0F0F0F0F0F0F0F0F0F0F0F
-; CHECK-NEXT:    DUP2
-; CHECK-NEXT:    DUP2
-; CHECK-NEXT:    PUSH32 0x3333333333333333333333333333333333333333333333333333333333333333
-; CHECK-NEXT:    DUP6
-; CHECK-NEXT:    PUSH1 0x1
-; CHECK-NEXT:    PUSH16 0x55555555555555555555555555555555
-; CHECK-NEXT:    SWAP8
-; CHECK-NEXT:    SUB
-; CHECK-NEXT:    SWAP1
-; CHECK-NEXT:    NOT
-; CHECK-NEXT:    AND
-; CHECK-NEXT:    PUSH16 0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF
-; CHECK-NEXT:    DUP8
-; CHECK-NEXT:    DUP3
-; CHECK-NEXT:    PUSH1 0x81
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    AND
-; CHECK-NEXT:    DUP3
-; CHECK-NEXT:    PUSH1 0x80
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    SUB
-; CHECK-NEXT:    DUP4
-; CHECK-NEXT:    DUP1
-; CHECK-NEXT:    DUP3
-; CHECK-NEXT:    PUSH1 0x2
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    AND
-; CHECK-NEXT:    SWAP2
-; CHECK-NEXT:    AND
-; CHECK-NEXT:    ADD
-; CHECK-NEXT:    SWAP8
-; CHECK-NEXT:    DUP3
-; CHECK-NEXT:    PUSH1 0x1
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    AND
-; CHECK-NEXT:    SWAP2
-; CHECK-NEXT:    AND
-; CHECK-NEXT:    SUB
-; CHECK-NEXT:    SWAP1
-; CHECK-NEXT:    DUP1
-; CHECK-NEXT:    DUP3
-; CHECK-NEXT:    PUSH1 0x2
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    AND
-; CHECK-NEXT:    SWAP2
-; CHECK-NEXT:    AND
-; CHECK-NEXT:    ADD
-; CHECK-NEXT:    SWAP5
-; CHECK-NEXT:    DUP1
-; CHECK-NEXT:    PUSH1 0x4
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    ADD
-; CHECK-NEXT:    AND
-; CHECK-NEXT:    MUL
-; CHECK-NEXT:    PUSH1 0x10
-; CHECK-NEXT:    BYTE
-; CHECK-NEXT:    SWAP3
-; CHECK-NEXT:    DUP1
-; CHECK-NEXT:    PUSH1 0x4
-; CHECK-NEXT:    SHR
-; CHECK-NEXT:    ADD
-; CHECK-NEXT:    AND
-; CHECK-NEXT:    MUL
-; CHECK-NEXT:    PUSH1 0x10
-; CHECK-NEXT:    BYTE
-; CHECK-NEXT:    ADD
-; CHECK-NEXT:    SWAP1
-; CHECK-NEXT:    JUMP
-
-  %res = call i256 @llvm.cttz.i256(i256 %v, i1 false)
-  ret i256 %res
-}
diff --git a/llvm/test/CodeGen/EVM/clz-ctz.ll b/llvm/test/CodeGen/EVM/clz-ctz.ll
diff --git a/llvm/test/MC/EVM/clz.ll b/llvm/test/MC/EVM/clz.ll
