[CIR] Upstream FPToFPBuiltin CosOp (#158342)

Upstream support for FPToFPBuiltin CosOp

Author: AmrDeveloper

clang/include/clang/CIR/Dialect/IR/CIROps.td

@@ -3847,6 +3847,16 @@ def CIR_ATanOp : CIR_UnaryFPToFPBuiltinOp<"atan", "ATanOp"> {
   }];
 }
 
+def CIR_CosOp : CIR_UnaryFPToFPBuiltinOp<"cos", "CosOp"> {
+  let summary = "Computes the floating-point cosine value";
+  let description = [{
+    `cir.cos` computes the cosine of a floating-point operand and returns
+    a result of the same type.
+
+    Floating-point exceptions are ignored, and it does not set `errno`.
+  }];
+}
+
 def CIR_FAbsOp : CIR_UnaryFPToFPBuiltinOp<"fabs", "FAbsOp"> {
   let summary = "Computes the floating-point absolute value";
   let description = [{

clang/lib/CIR/CodeGen/CIRGenBuiltin.cpp

@@ -200,6 +200,17 @@ RValue CIRGenFunction::emitBuiltinExpr(const GlobalDecl &gd, unsigned builtinID,
         builder.createBitcast(allocaAddr, builder.getVoidPtrTy()));
   }
 
+  case Builtin::BIcos:
+  case Builtin::BIcosf:
+  case Builtin::BIcosl:
+  case Builtin::BI__builtin_cos:
+  case Builtin::BI__builtin_cosf:
+  case Builtin::BI__builtin_cosf16:
+  case Builtin::BI__builtin_cosl:
+  case Builtin::BI__builtin_cosf128:
+    assert(!cir::MissingFeatures::fastMathFlags());
+    return emitUnaryMaybeConstrainedFPBuiltin<cir::CosOp>(*this, *e);
+
   case Builtin::BIfabs:
   case Builtin::BIfabsf:
   case Builtin::BIfabsl:
@@ -415,6 +426,8 @@ RValue CIRGenFunction::emitBuiltinExpr(const GlobalDecl &gd, unsigned builtinID,
     return emitUnaryFPBuiltin<cir::ASinOp>(*this, *e);
   case Builtin::BI__builtin_elementwise_atan:
     return emitUnaryFPBuiltin<cir::ATanOp>(*this, *e);
+  case Builtin::BI__builtin_elementwise_cos:
+    return emitUnaryFPBuiltin<cir::CosOp>(*this, *e);
   }
 
   // If this is an alias for a lib function (e.g. __builtin_sin), emit

clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp

@@ -185,6 +185,14 @@ mlir::LogicalResult CIRToLLVMCopyOpLowering::matchAndRewrite(
   return mlir::success();
 }
 
+mlir::LogicalResult CIRToLLVMCosOpLowering::matchAndRewrite(
+    cir::CosOp op, OpAdaptor adaptor,
+    mlir::ConversionPatternRewriter &rewriter) const {
+  mlir::Type resTy = typeConverter->convertType(op.getType());
+  rewriter.replaceOpWithNewOp<mlir::LLVM::CosOp>(op, resTy, adaptor.getSrc());
+  return mlir::success();
+}
+
 static mlir::Value getLLVMIntCast(mlir::ConversionPatternRewriter &rewriter,
                                   mlir::Value llvmSrc, mlir::Type llvmDstIntTy,
                                   bool isUnsigned, uint64_t cirSrcWidth,
@@ -2498,6 +2506,7 @@ void ConvertCIRToLLVMPass::runOnOperation() {
                CIRToLLVMComplexRealPtrOpLowering,
                CIRToLLVMComplexSubOpLowering,
                CIRToLLVMCopyOpLowering,
+               CIRToLLVMCosOpLowering,
                CIRToLLVMConstantOpLowering,
                CIRToLLVMExpectOpLowering,
                CIRToLLVMFAbsOpLowering,

clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.h

@@ -189,6 +189,15 @@ class CIRToLLVMCopyOpLowering : public mlir::OpConversionPattern<cir::CopyOp> {
                   mlir::ConversionPatternRewriter &) const override;
 };
 
+class CIRToLLVMCosOpLowering : public mlir::OpConversionPattern<cir::CosOp> {
+public:
+  using mlir::OpConversionPattern<cir::CosOp>::OpConversionPattern;
+
+  mlir::LogicalResult
+  matchAndRewrite(cir::CosOp op, OpAdaptor,
+                  mlir::ConversionPatternRewriter &) const override;
+};
+
 class CIRToLLVMExpectOpLowering
     : public mlir::OpConversionPattern<cir::ExpectOp> {
 public:

clang/test/CIR/CodeGen/builtins-elementwise.c

@@ -89,3 +89,30 @@ void test_builtin_elementwise_atan(float f, double d, vfloat4 vf4,
   // OGCG: %{{.*}} = call <4 x double> @llvm.atan.v4f64(<4 x double> %{{.*}})
   vd4 = __builtin_elementwise_atan(vd4);
 }
+
+void test_builtin_elementwise_cos(float f, double d, vfloat4 vf4,
+                                     vdouble4 vd4) {
+  // CIR-LABEL: test_builtin_elementwise_cos
+  // LLVM-LABEL: test_builtin_elementwise_cos
+  // OGCG-LABEL: test_builtin_elementwise_cos
+
+  // CIR: {{%.*}} = cir.cos {{%.*}} : !cir.float
+  // LLVM: {{%.*}} = call float @llvm.cos.f32(float {{%.*}})
+  // OGCG: {{%.*}} = call float @llvm.cos.f32(float {{%.*}})
+  f = __builtin_elementwise_cos(f);
+
+  // CIR: {{%.*}} = cir.cos {{%.*}} : !cir.double
+  // LLVM: {{%.*}} = call double @llvm.cos.f64(double {{%.*}})
+  // OGCG: {{%.*}} = call double @llvm.cos.f64(double {{%.*}})
+  d = __builtin_elementwise_cos(d);
+
+  // CIR: {{%.*}} = cir.cos {{%.*}} : !cir.vector<4 x !cir.float>
+  // LLVM: {{%.*}} = call <4 x float> @llvm.cos.v4f32(<4 x float> {{%.*}})
+  // OGCG: {{%.*}} = call <4 x float> @llvm.cos.v4f32(<4 x float> {{%.*}})
+  vf4 = __builtin_elementwise_cos(vf4);
+
+  // CIR: {{%.*}} = cir.cos {{%.*}} : !cir.vector<4 x !cir.double>
+  // LLVM: {{%.*}} = call <4 x double> @llvm.cos.v4f64(<4 x double> {{%.*}})
+  // OGCG: {{%.*}} = call <4 x double> @llvm.cos.v4f64(<4 x double> {{%.*}})
+  vd4 = __builtin_elementwise_cos(vd4);
+}

clang/test/CIR/CodeGen/builtins-floating-point.c

@@ -0,0 +1,20 @@
+// RUN: %clang_cc1 -triple aarch64-none-linux-android24 -Wno-unused-value -fclangir -emit-cir %s -o %t.cir
+// RUN: FileCheck --input-file=%t.cir %s -check-prefix=CIR
+// RUN: %clang_cc1 -triple aarch64-none-linux-android24 -Wno-unused-value -fclangir -emit-llvm %s -o %t-cir.ll
+// RUN: FileCheck --input-file=%t-cir.ll %s -check-prefix=LLVM
+// RUN: %clang_cc1 -triple aarch64-none-linux-android24 -Wno-unused-value -emit-llvm %s -o %t.ll
+// RUN: FileCheck --input-file=%t.ll %s -check-prefix=OGCG
+
+float cosf(float f) {
+  return __builtin_cosf(f);
+  // CHECK: %{{.*}} = cir.cos {{.*}} : !cir.float
+  // LLVM: %{{.*}} = call float @llvm.cos.f32(float %{{.*}})
+  // OGCG: %{{.*}} = call float @llvm.cos.f32(float %{{.*}})
+}
+
+double cos(double f) {
+  return __builtin_cos(f);
+  // CIR: {{.+}} = cir.cos {{.+}} : !cir.double
+  // LLVM: %{{.*}} = call double @llvm.cos.f64(double %{{.*}})
+  // OGCG: %{{.*}} = call double @llvm.cos.f64(double %{{.*}})
+}