dialects: (llvm) Add a bunch of float methods

tests/filecheck/dialects/llvm/arithmetic.mlir

@@ -1,6 +1,6 @@
 // RUN: XDSL_ROUNDTRIP
 
-%arg0, %arg1 = "test.op"() : () -> (i32, i32)
+%arg0, %arg1, %f1 = "test.op"() : () -> (i32, i32, f32)
 
 %add_both = llvm.add %arg0, %arg1 {"overflowFlags" = #llvm.overflow<nsw, nuw>} : i32
 // CHECK: %add_both = llvm.add %arg0, %arg1 {overflowFlags = #llvm.overflow<nsw,nuw>} : i32
@@ -121,3 +121,23 @@
 
 %icmp_uge = llvm.icmp "uge" %arg0, %arg1 : i32
 // CHECK: %icmp_uge = llvm.icmp "uge" %arg0, %arg1 : i32
+
+// float arith:
+
+%fmul = llvm.fmul %f1, %f1 : f32
+// CHECK: %fmul = llvm.fmul %f1, %f1 : f32
+
+%fmul_fast = llvm.fmul %f1, %f1 {fastmathFlags = #llvm.fastmath<fast>} : f32
+// CHECK: %fmul_fast = llvm.fmul %f1, %f1 {fastmathFlags = #llvm.fastmath<fast>} : f32
+
+%fdiv = llvm.fdiv %f1, %f1 : f32
+// CHECK: %fdiv = llvm.fdiv %f1, %f1 : f32
+
+%fadd = llvm.fadd %f1, %f1 : f32
+// CHECK: %fadd = llvm.fadd %f1, %f1 : f32
+
+%fsub = llvm.fsub %f1, %f1 : f32
+// CHECK: %fsub = llvm.fsub %f1, %f1 : f32
+
+%frem = llvm.frem %f1, %f1 : f32
+// CHECK: %frem = llvm.frem %f1, %f1 : f32

tests/filecheck/dialects/llvm/example.mlir

@@ -90,4 +90,12 @@ builtin.module {
 
 // CHECK:      %val = "test.op"() : () -> i32
 // CHECK-NEXT: %fval = llvm.bitcast %val : i32 to f32
+
+  %fval2 = llvm.sitofp %val : i32 to f32
+
+// CHECK-NEXT: %fval2 = llvm.sitofp %val : i32 to f32
+
+  %fval3 = llvm.fpext %fval : f32 to f64
+
+// CHECK-NEXT: %fval3 = llvm.fpext %fval : f32 to f64
 }

xdsl/dialects/llvm.py

@@ -13,6 +13,9 @@
     ContainerType,
     DenseArrayBase,
     DenseI64ArrayConstr,
+    Float16Type,
+    Float32Type,
+    Float64Type,
     IndexType,
     IntAttr,
     IntegerAttr,
@@ -25,7 +28,7 @@
     i32,
     i64,
 )
-from xdsl.dialects.utils import FastMathAttrBase
+from xdsl.dialects.utils import FastMathAttrBase, FastMathFlag
 from xdsl.ir import (
     Attribute,
     BitEnumAttribute,
@@ -46,6 +49,7 @@
     irdl_attr_definition,
     irdl_op_definition,
     operand_def,
+    opt_attr_def,
     opt_operand_def,
     opt_prop_def,
     opt_result_def,
@@ -54,6 +58,7 @@
     result_def,
     traits_def,
     var_operand_def,
+    AnyOf,
 )
 from xdsl.parser import AttrParser, Parser
 from xdsl.printer import Printer
@@ -1710,10 +1715,7 @@ class ZeroOp(IRDLOperation):
     res = result_def(LLVMTypeConstr)
 
 
-@irdl_op_definition
-class BitcastOp(IRDLOperation):
-    name = "llvm.bitcast"
-
+class GenericCastOp(IRDLOperation, ABC):
     arg = operand_def(Attribute)
     """
     LLVM-compatible non-aggregate type
@@ -1735,56 +1737,135 @@ def __init__(self, val: Operation | SSAValue, res_type: Attribute):
         )
 
 
+floatingPointLike = AnyOf([Float16Type, Float32Type, Float64Type])
+
+
+class AbstractFloatArithOp(IRDLOperation, ABC):
+    T: ClassVar = VarConstraint("T", floatingPointLike)
+
+    lhs = operand_def(T)
+    rhs = operand_def(T)
+    res = result_def(T)
+
+    fastmathFlags = opt_attr_def(FastMathAttr)
+
+    traits = traits_def(NoMemoryEffect())
+
+    assembly_format = "$lhs `,` $rhs attr-dict `:` type($lhs)"
+
+    def __init__(
+        self,
+        lhs: SSAValue | Operation,
+        rhs: SSAValue | Operation,
+        fast_math: FastMathAttr | FastMathFlag | None,
+    ):
+        if isinstance(fast_math, FastMathFlag | str):
+            fast_math = FastMathAttr(fast_math)
+
+        super().__init__(
+            operands=[lhs, rhs],
+            result_types=[SSAValue.get(lhs).type],
+            attributes={"fastmathFlags": fast_math},
+        )
+
+
+@irdl_op_definition
+class FAddOp(AbstractFloatArithOp):
+    name = "llvm.fadd"
+
+
+@irdl_op_definition
+class FMulOp(AbstractFloatArithOp):
+    name = "llvm.fmul"
+
+
+@irdl_op_definition
+class FDivOp(AbstractFloatArithOp):
+    name = "llvm.fdiv"
+
+
+@irdl_op_definition
+class FSubOp(AbstractFloatArithOp):
+    name = "llvm.fsub"
+
+
+@irdl_op_definition
+class FRemOp(AbstractFloatArithOp):
+    name = "llvm.frem"
+
+
+@irdl_op_definition
+class BitcastOp(GenericCastOp):
+    name = "llvm.bitcast"
+
+
+@irdl_op_definition
+class SIToFPOp(GenericCastOp):
+    name = "llvm.sitofp"
+
+
+@irdl_op_definition
+class FPExtOp(GenericCastOp):
+    name = "llvm.fpext"
+
+
 LLVM = Dialect(
     "llvm",
     [
+        AShrOp,
         AddOp,
+        AddressOfOp,
+        AllocaOp,
+        AndOp,
         BitcastOp,
-        SubOp,
+        CallIntrinsicOp,
+        CallOp,
+        ConstantOp,
+        ExtractValueOp,
+        FAddOp,
+        FDivOp,
+        FMulOp,
+        FPExtOp,
+        FRemOp,
+        FSubOp,
+        FuncOp,
+        GEPOp,
+        GlobalOp,
+        ICmpOp,
+        InlineAsmOp,
+        InsertValueOp,
+        IntToPtrOp,
+        LShrOp,
+        LoadOp,
         MulOp,
-        UDivOp,
+        NullOp,
+        OrOp,
+        ReturnOp,
         SDivOp,
-        URemOp,
+        SExtOp,
+        SIToFPOp,
         SRemOp,
-        AndOp,
-        OrOp,
-        XOrOp,
         ShlOp,
-        LShrOp,
-        AShrOp,
+        StoreOp,
+        SubOp,
         TruncOp,
-        ZExtOp,
-        SExtOp,
-        ICmpOp,
-        ExtractValueOp,
-        InsertValueOp,
-        InlineAsmOp,
+        UDivOp,
+        URemOp,
         UndefOp,
-        AllocaOp,
-        GEPOp,
-        IntToPtrOp,
-        NullOp,
-        LoadOp,
-        StoreOp,
-        GlobalOp,
-        AddressOfOp,
-        FuncOp,
-        CallOp,
-        ReturnOp,
-        ConstantOp,
-        CallIntrinsicOp,
+        XOrOp,
+        ZExtOp,
         ZeroOp,
     ],
     [
-        LLVMStructType,
-        LLVMPointerType,
+        CallingConventionAttr,
+        FastMathAttr,
         LLVMArrayType,
-        LLVMVoidType,
         LLVMFunctionType,
+        LLVMPointerType,
+        LLVMStructType,
+        LLVMVoidType,
         LinkageAttr,
-        CallingConventionAttr,
-        TailCallKindAttr,
-        FastMathAttr,
         OverflowAttr,
+        TailCallKindAttr,
     ],
 )