[CIR] Extend support for floating point attributes

This commit extends the support for floating point attributes parsing by using the new `AsmParser::parseFloat(fltSemnatics, APFloat&)` interface. As a drive-by, this commit also harmonizes the cir.fp print/parse namespace usage, and adds the constraint of supporting only "CIRFPType"s for cir.fp in tablegen instead of verifying it manually in the parsing logic.
llvm · May 1, 2024 · d3d3988 · d3d3988
1 parent f780f38
commit d3d3988
Show file tree

Hide file tree

Showing 7 changed files with 203 additions and 41 deletions.
diff --git a/clang/include/clang/CIR/Dialect/IR/CIRAttrs.td b/clang/include/clang/CIR/Dialect/IR/CIRAttrs.td
@@ -223,13 +223,20 @@ def FPAttr : CIR_Attr<"FP", "fp", [TypedAttrInterface]> {
   let summary = "An attribute containing a floating-point value";
   let description = [{
     An fp attribute is a literal attribute that represents a floating-point
-    value of the specified floating-point type.
+    value of the specified floating-point type. Supporting only CIR FP types.
   }];
-  let parameters = (ins AttributeSelfTypeParameter<"">:$type, "APFloat":$value);
+  let parameters = (ins
+    AttributeSelfTypeParameter<"", "::mlir::cir::CIRFPTypeInterface">:$type,
+    APFloatParameter<"">:$value
+  );
   let builders = [
     AttrBuilderWithInferredContext<(ins "Type":$type,
                                         "const APFloat &":$value), [{
-      return $_get(type.getContext(), type, value);
+      return $_get(type.getContext(), type.cast<CIRFPTypeInterface>(), value);
+    }]>,
+    AttrBuilder<(ins "Type":$type,
+                     "const APFloat &":$value), [{
+      return $_get($_ctxt, type.cast<CIRFPTypeInterface>(), value);
     }]>,
   ];
   let extraClassDeclaration = [{

diff --git a/clang/lib/CIR/Dialect/IR/CIRAttrs.cpp b/clang/lib/CIR/Dialect/IR/CIRAttrs.cpp
@@ -41,7 +41,8 @@ static void printFloatLiteral(mlir::AsmPrinter &p, llvm::APFloat value,
                               mlir::Type ty);
 static mlir::ParseResult
 parseFloatLiteral(mlir::AsmParser &parser,
-                  mlir::FailureOr<llvm::APFloat> &value, mlir::Type ty);
+                  mlir::FailureOr<llvm::APFloat> &value,
+                  mlir::cir::CIRFPTypeInterface fpType);
 
 #define GET_ATTRDEF_CLASSES
 #include "clang/CIR/Dialect/IR/CIROpsAttributes.cpp.inc"
@@ -317,50 +318,30 @@ LogicalResult IntAttr::verify(function_ref<InFlightDiagnostic()> emitError,
 // FPAttr definitions
 //===----------------------------------------------------------------------===//
 
-static void printFloatLiteral(mlir::AsmPrinter &p, llvm::APFloat value,
-                              mlir::Type ty) {
+static void printFloatLiteral(AsmPrinter &p, APFloat value, Type ty) {
   p << value;
 }
 
-static mlir::ParseResult
-parseFloatLiteral(mlir::AsmParser &parser,
-                  mlir::FailureOr<llvm::APFloat> &value, mlir::Type ty) {
-  double rawValue;
-  if (parser.parseFloat(rawValue)) {
-    return parser.emitError(parser.getCurrentLocation(),
-                            "expected floating-point value");
-  }
-
-  auto losesInfo = false;
-  value.emplace(rawValue);
+static ParseResult parseFloatLiteral(AsmParser &parser,
+                                     FailureOr<APFloat> &value,
+                                     CIRFPTypeInterface fpType) {
 
-  auto tyFpInterface = ty.dyn_cast<cir::CIRFPTypeInterface>();
-  if (!tyFpInterface) {
-    // Parsing of the current floating-point literal has succeeded, but the
-    // given attribute type is invalid. This error will be reported later when
-    // the attribute is being verified.
-    return success();
-  }
+  APFloat parsedValue(0.0);
+  if (parser.parseFloat(fpType.getFloatSemantics(), parsedValue))
+    return failure();
 
-  value->convert(tyFpInterface.getFloatSemantics(),
-                 llvm::RoundingMode::TowardZero, &losesInfo);
+  value.emplace(parsedValue);
   return success();
 }
 
-cir::FPAttr cir::FPAttr::getZero(mlir::Type type) {
-  return get(type,
-             APFloat::getZero(
-                 type.cast<cir::CIRFPTypeInterface>().getFloatSemantics()));
+FPAttr FPAttr::getZero(Type type) {
+  return get(type, APFloat::getZero(
+                       type.cast<CIRFPTypeInterface>().getFloatSemantics()));
 }
 
-LogicalResult cir::FPAttr::verify(function_ref<InFlightDiagnostic()> emitError,
-                                  Type type, APFloat value) {
-  auto fltTypeInterface = type.dyn_cast<cir::CIRFPTypeInterface>();
-  if (!fltTypeInterface) {
-    emitError() << "expected floating-point type";
-    return failure();
-  }
-  if (APFloat::SemanticsToEnum(fltTypeInterface.getFloatSemantics()) !=
+LogicalResult FPAttr::verify(function_ref<InFlightDiagnostic()> emitError,
+                             CIRFPTypeInterface fpType, APFloat value) {
+  if (APFloat::SemanticsToEnum(fpType.getFloatSemantics()) !=
       APFloat::SemanticsToEnum(value.getSemantics())) {
     emitError() << "floating-point semantics mismatch";
     return failure();

diff --git a/clang/test/CIR/IR/attribute.cir b/clang/test/CIR/IR/attribute.cir
@@ -0,0 +1,25 @@
+// RUN: cir-opt %s -split-input-file -allow-unregistered-dialect -verify-diagnostics | FileCheck %s
+
+cir.func @float_attrs_pass() {
+  "test.float_attrs"() {
+    // CHECK: float_attr = #cir.fp<2.000000e+00> : !cir.float
+    float_attr = #cir.fp<2.> : !cir.float
+  } : () -> ()
+  "test.float_attrs"() {
+    // CHECK: float_attr = #cir.fp<-2.000000e+00> : !cir.float
+    float_attr = #cir.fp<-2.> : !cir.float
+  } : () -> ()
+  "test.float_attrs"() {
+    // CHECK: float_attr = #cir.fp<2.000000e+00> : !cir.double
+    float_attr = #cir.fp<2.> : !cir.double
+  } : () -> ()
+  "test.float_attrs"() {
+    // CHECK: float_attr = #cir.fp<2.000000e+00> : !cir.long_double<!cir.f80>
+    float_attr = #cir.fp<2.> : !cir.long_double<!cir.f80>
+  } : () -> ()
+  "test.float_attrs"() {
+    // CHECK: float_attr = #cir.fp<2.000000e+00> : !cir.long_double<!cir.double>
+    float_attr = #cir.fp<2.> : !cir.long_double<!cir.double>
+  } : () -> ()
+  cir.return
+}
diff --git a/clang/test/CIR/IR/float.cir b/clang/test/CIR/IR/float.cir
@@ -0,0 +1,90 @@
+// RUN: cir-opt %s | FileCheck %s
+
+// Adapted from mlir/test/IR/parser.mlir
+
+// CHECK-LABEL: @f32_special_values
+cir.func @f32_special_values() {
+  // F32 signaling NaNs.
+  // CHECK: cir.const(#cir.fp<0x7F800001> : !cir.float) : !cir.float
+  %0 = cir.const(#cir.fp<0x7F800001> : !cir.float) : !cir.float
+  // CHECK: cir.const(#cir.fp<0x7FBFFFFF> : !cir.float) : !cir.float
+  %1 = cir.const(#cir.fp<0x7FBFFFFF> : !cir.float) : !cir.float
+
+  // F32 quiet NaNs.
+  // CHECK: cir.const(#cir.fp<0x7FC00000> : !cir.float) : !cir.float
+  %2 = cir.const(#cir.fp<0x7FC00000> : !cir.float) : !cir.float
+  // CHECK: cir.const(#cir.fp<0xFFFFFFFF> : !cir.float) : !cir.float
+  %3 = cir.const(#cir.fp<0xFFFFFFFF> : !cir.float) : !cir.float
+
+  // F32 positive infinity.
+  // CHECK: cir.const(#cir.fp<0x7F800000> : !cir.float) : !cir.float
+  %4 = cir.const(#cir.fp<0x7F800000> : !cir.float) : !cir.float
+  // F32 negative infinity.
+  // CHECK: cir.const(#cir.fp<0xFF800000> : !cir.float) : !cir.float
+  %5 = cir.const(#cir.fp<0xFF800000> : !cir.float) : !cir.float
+
+  cir.return
+}
+
+// CHECK-LABEL: @f64_special_values
+cir.func @f64_special_values() {
+  // F64 signaling NaNs.
+  // CHECK: cir.const(#cir.fp<0x7FF0000000000001> : !cir.double) : !cir.double
+  %0 = cir.const(#cir.fp<0x7FF0000000000001> : !cir.double) : !cir.double
+  // CHECK: cir.const(#cir.fp<0x7FF8000000000000> : !cir.double) : !cir.double
+  %1 = cir.const(#cir.fp<0x7FF8000000000000> : !cir.double) : !cir.double
+
+  // F64 quiet NaNs.
+  // CHECK: cir.const(#cir.fp<0x7FF0000001000000> : !cir.double) : !cir.double
+  %2 = cir.const(#cir.fp<0x7FF0000001000000> : !cir.double) : !cir.double
+  // CHECK: cir.const(#cir.fp<0xFFF0000001000000> : !cir.double) : !cir.double
+  %3 = cir.const(#cir.fp<0xFFF0000001000000> : !cir.double) : !cir.double
+
+  // F64 positive infinity.
+  // CHECK: cir.const(#cir.fp<0x7FF0000000000000> : !cir.double) : !cir.double
+  %4 = cir.const(#cir.fp<0x7FF0000000000000> : !cir.double) : !cir.double
+  // F64 negative infinity.
+  // CHECK: cir.const(#cir.fp<0xFFF0000000000000> : !cir.double) : !cir.double
+  %5 = cir.const(#cir.fp<0xFFF0000000000000> : !cir.double) : !cir.double
+
+  // Check that values that can't be represented with the default format, use
+  // hex instead.
+  // CHECK: cir.const(#cir.fp<0xC1CDC00000000000> : !cir.double) : !cir.double
+  %6 = cir.const(#cir.fp<0xC1CDC00000000000> : !cir.double) : !cir.double
+
+  cir.return
+}
+
+// CHECK-LABEL: @f80_special_values
+cir.func @f80_special_values() {
+  // F80 signaling NaNs.
+  // CHECK: cir.const(#cir.fp<0x7FFFE000000000000001> : !cir.long_double<!cir.f80>) : !cir.long_double<!cir.f80>
+  %0 = cir.const(#cir.fp<0x7FFFE000000000000001> : !cir.long_double<!cir.f80>) : !cir.long_double<!cir.f80>
+  // CHECK: cir.const(#cir.fp<0x7FFFB000000000000011> : !cir.long_double<!cir.f80>) : !cir.long_double<!cir.f80>
+  %1 = cir.const(#cir.fp<0x7FFFB000000000000011> : !cir.long_double<!cir.f80>) : !cir.long_double<!cir.f80>
+
+  // F80 quiet NaNs.
+  // CHECK: cir.const(#cir.fp<0x7FFFC000000000100000> : !cir.long_double<!cir.f80>) : !cir.long_double<!cir.f80>
+  %2 = cir.const(#cir.fp<0x7FFFC000000000100000> : !cir.long_double<!cir.f80>) : !cir.long_double<!cir.f80>
+  // CHECK: cir.const(#cir.fp<0x7FFFE000000001000000> : !cir.long_double<!cir.f80>) : !cir.long_double<!cir.f80>
+  %3 = cir.const(#cir.fp<0x7FFFE000000001000000> : !cir.long_double<!cir.f80>) : !cir.long_double<!cir.f80>
+
+  // F80 positive infinity.
+  // CHECK: cir.const(#cir.fp<0x7FFF8000000000000000> : !cir.long_double<!cir.f80>) : !cir.long_double<!cir.f80>
+  %4 = cir.const(#cir.fp<0x7FFF8000000000000000> : !cir.long_double<!cir.f80>) : !cir.long_double<!cir.f80>
+  // F80 negative infinity.
+  // CHECK: cir.const(#cir.fp<0xFFFF8000000000000000> : !cir.long_double<!cir.f80>) : !cir.long_double<!cir.f80>
+  %5 = cir.const(#cir.fp<0xFFFF8000000000000000> : !cir.long_double<!cir.f80>) : !cir.long_double<!cir.f80>
+
+  cir.return
+}
+
+// We want to print floats in exponential notation with 6 significant digits,
+// but it may lead to precision loss when parsing back, in which case we print
+// the decimal form instead.
+// CHECK-LABEL: @f32_potential_precision_loss()
+cir.func @f32_potential_precision_loss() {
+  // CHECK: cir.const(#cir.fp<1.23697901> : !cir.float) : !cir.float
+  %0 = cir.const(#cir.fp<1.23697901> : !cir.float) : !cir.float
+  cir.return
+}
diff --git a/clang/test/CIR/IR/invalid.cir b/clang/test/CIR/IR/invalid.cir
@@ -1044,7 +1044,7 @@ cir.func @bad_fetch(%x: !cir.ptr<!cir.float>, %y: !cir.float) -> () {
 
 cir.func @bad_operands_for_nowrap(%x: !cir.float, %y: !cir.float) {
   // expected-error@+1 {{only operations on integer values may have nsw/nuw flags}}
-  %0 = cir.binop(add, %x, %y) nsw : !cir.float 
+  %0 = cir.binop(add, %x, %y) nsw : !cir.float
 }
 
 // -----
@@ -1119,3 +1119,62 @@ module {
     %0 = cir.dyn_cast(ptr, %arg0 : !cir.ptr<!Base>, #cir.dyn_cast_info<#cir.global_view<@_ZTI4Base> : !cir.ptr<!u8i>, #cir.global_view<@_ZTI7Derived> : !cir.ptr<!u32i>, @__dynamic_cast, @__cxa_bad_cast, #cir.int<0> : !s64i>) -> !cir.ptr<!Derived>
   }
 }
+// -----
+
+// Type of the attribute must be a CIR floating point type
+
+// expected-error @below {{invalid kind of type specified}}
+cir.global external @f = #cir.fp<0.5> : !cir.int<s, 32>
+
+// -----
+
+// Value must be a floating point literal or integer literal
+
+// expected-error @below {{expected floating point literal}}
+cir.global external @f = #cir.fp<"blabla"> : !cir.float
+
+// -----
+
+// Integer value must be in the width of the floating point type
+
+// expected-error @below {{hexadecimal float constant out of range for type}}
+cir.global external @f = #cir.fp<0x7FC000000> : !cir.float
+
+// -----
+
+// Integer value must be in the width of the floating point type
+
+// expected-error @below {{hexadecimal float constant out of range for type}}
+cir.global external @f = #cir.fp<0x7FC000007FC0000000> : !cir.double
+
+// -----
+
+// Integer value must be in the width of the floating point type
+
+// expected-error @below {{hexadecimal float constant out of range for type}}
+cir.global external @f = #cir.fp<0x7FC0000007FC0000007FC000000> : !cir.long_double<!cir.f80>
+
+// -----
+
+// Long double with `double` semnatics should have a value that fits in a double.
+
+// CHECK: cir.global external @f = #cir.fp<0x7FC000007FC000000000> : !cir.long_double<!cir.f80>
+cir.global external @f = #cir.fp<0x7FC000007FC000000000> : !cir.long_double<!cir.f80>
+
+// expected-error @below {{hexadecimal float constant out of range for type}}
+cir.global external @f = #cir.fp<0x7FC000007FC000000000> : !cir.long_double<!cir.double>
+
+// -----
+
+// Verify no need for type inside the attribute
+
+// expected-error @below {{expected '>'}}
+cir.global external @f = #cir.fp<0x7FC00000 : !cir.float> : !cir.float
+
+// -----
+
+// Verify literal must be hex or float
+
+// expected-error @below {{unexpected decimal integer literal for a floating point value}}
+// expected-note @below {{add a trailing dot to make the literal a float}}
+cir.global external @f = #cir.fp<42> : !cir.float
diff --git a/clang/test/CIR/Lowering/class.cir b/clang/test/CIR/Lowering/class.cir
@@ -44,7 +44,7 @@ module {
   // CHECK:   %0 = llvm.mlir.undef : !llvm.struct<"class.S1", (i32, f32, ptr)>
   // CHECK:   %1 = llvm.mlir.constant(1 : i32) : i32
   // CHECK:   %2 = llvm.insertvalue %1, %0[0] : !llvm.struct<"class.S1", (i32, f32, ptr)>
-  // CHECK:   %3 = llvm.mlir.constant(0.099999994 : f32) : f32
+  // CHECK:   %3 = llvm.mlir.constant(1.000000e-01 : f32) : f32
   // CHECK:   %4 = llvm.insertvalue %3, %2[1] : !llvm.struct<"class.S1", (i32, f32, ptr)>
   // CHECK:   %5 = llvm.mlir.zero : !llvm.ptr
   // CHECK:   %6 = llvm.insertvalue %5, %4[2] : !llvm.struct<"class.S1", (i32, f32, ptr)>

diff --git a/clang/test/CIR/Lowering/struct.cir b/clang/test/CIR/Lowering/struct.cir
@@ -44,7 +44,7 @@ module {
   // CHECK:   %0 = llvm.mlir.undef : !llvm.struct<"struct.S1", (i32, f32, ptr)>
   // CHECK:   %1 = llvm.mlir.constant(1 : i32) : i32
   // CHECK:   %2 = llvm.insertvalue %1, %0[0] : !llvm.struct<"struct.S1", (i32, f32, ptr)>
-  // CHECK:   %3 = llvm.mlir.constant(0.099999994 : f32) : f32
+  // CHECK:   %3 = llvm.mlir.constant(1.000000e-01 : f32) : f32
   // CHECK:   %4 = llvm.insertvalue %3, %2[1] : !llvm.struct<"struct.S1", (i32, f32, ptr)>
   // CHECK:   %5 = llvm.mlir.zero : !llvm.ptr
   // CHECK:   %6 = llvm.insertvalue %5, %4[2] : !llvm.struct<"struct.S1", (i32, f32, ptr)>