[CIR][CodeGen] Support for __builtin_elementwise_atan (#1512)

Sub-issue of #1192. Adds
CIR_ATanOp and support for __builtin_elementwise_atan.

Author: ayokunle321

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

@@ -4737,9 +4737,10 @@ class UnaryFPToFPBuiltinOp<string mnemonic, string llvmOpName>
   let llvmOp = llvmOpName;
 }
 
+def ACosOp : UnaryFPToFPBuiltinOp<"acos", "ACosOp">;
 def ASinOp : UnaryFPToFPBuiltinOp<"asin", "ASinOp">;
+def ATanOp : UnaryFPToFPBuiltinOp<"atan", "ATanOp">;
 def CeilOp : UnaryFPToFPBuiltinOp<"ceil", "FCeilOp">;
-def ACosOp : UnaryFPToFPBuiltinOp<"acos", "ACosOp">;
 def CosOp : UnaryFPToFPBuiltinOp<"cos", "CosOp">;
 def ExpOp : UnaryFPToFPBuiltinOp<"exp", "ExpOp">;
 def Exp2Op : UnaryFPToFPBuiltinOp<"exp2", "Exp2Op">;

clang/lib/CIR/CodeGen/CIRGenBuiltin.cpp

@@ -1406,7 +1406,7 @@ RValue CIRGenFunction::emitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
   case Builtin::BI__builtin_elementwise_asin:
     return emitUnaryFPBuiltin<cir::ASinOp>(*this, *E);
   case Builtin::BI__builtin_elementwise_atan:
-    llvm_unreachable("BI__builtin_elementwise_atan NYI");
+    return emitUnaryFPBuiltin<cir::ATanOp>(*this, *E);
   case Builtin::BI__builtin_elementwise_atan2:
     llvm_unreachable("BI__builtin_elementwise_atan2 NYI");
   case Builtin::BI__builtin_elementwise_ceil:

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

@@ -60,23 +60,44 @@ void test_builtin_elementwise_acos(float f, double d, vfloat4 vf4,
 
 void test_builtin_elementwise_asin(float f, double d, vfloat4 vf4,
   vdouble4  vd4) {
-// CIR-LABEL: test_builtin_elementwise_asin
-// LLVM-LABEL: test_builtin_elementwise_asin
-// CIR: {{%.*}} = cir.asin {{%.*}} : !cir.float
-// LLVM: {{%.*}} = call float @llvm.asin.f32(float {{%.*}})
-f = __builtin_elementwise_asin(f);
-
-// CIR: {{%.*}} = cir.asin {{%.*}} : !cir.double
-// LLVM: {{%.*}} = call double @llvm.asin.f64(double {{%.*}})
-d = __builtin_elementwise_asin(d);
-
-// CIR: {{%.*}} = cir.asin {{%.*}} : !cir.vector<!cir.float x 4>
-// LLVM: {{%.*}} = call <4 x float> @llvm.asin.v4f32(<4 x float> {{%.*}})
-vf4 = __builtin_elementwise_asin(vf4);
-
-// CIR: {{%.*}} = cir.asin {{%.*}} : !cir.vector<!cir.double x 4>
-// LLVM: {{%.*}} = call <4 x double> @llvm.asin.v4f64(<4 x double> {{%.*}})
-vd4 = __builtin_elementwise_asin(vd4);
+  // CIR-LABEL: test_builtin_elementwise_asin
+  // LLVM-LABEL: test_builtin_elementwise_asin
+  // CIR: {{%.*}} = cir.asin {{%.*}} : !cir.float
+  // LLVM: {{%.*}} = call float @llvm.asin.f32(float {{%.*}})
+  f = __builtin_elementwise_asin(f);
+
+  // CIR: {{%.*}} = cir.asin {{%.*}} : !cir.double
+  // LLVM: {{%.*}} = call double @llvm.asin.f64(double {{%.*}})
+  d = __builtin_elementwise_asin(d);
+
+  // CIR: {{%.*}} = cir.asin {{%.*}} : !cir.vector<!cir.float x 4>
+  // LLVM: {{%.*}} = call <4 x float> @llvm.asin.v4f32(<4 x float> {{%.*}})
+  vf4 = __builtin_elementwise_asin(vf4);
+
+  // CIR: {{%.*}} = cir.asin {{%.*}} : !cir.vector<!cir.double x 4>
+  // LLVM: {{%.*}} = call <4 x double> @llvm.asin.v4f64(<4 x double> {{%.*}})
+  vd4 = __builtin_elementwise_asin(vd4);
+}
+
+void test_builtin_elementwise_atan(float f, double d, vfloat4 vf4,
+  vdouble4  vd4) {
+  // CIR-LABEL: test_builtin_elementwise_atan
+  // LLVM-LABEL: test_builtin_elementwise_atan
+  // CIR: {{%.*}} = cir.atan {{%.*}} : !cir.float
+  // LLVM: {{%.*}} = call float @llvm.atan.f32(float {{%.*}})
+  f = __builtin_elementwise_atan(f);
+
+  // CIR: {{%.*}} = cir.atan {{%.*}} : !cir.double
+  // LLVM: {{%.*}} = call double @llvm.atan.f64(double {{%.*}})
+  d = __builtin_elementwise_atan(d);
+
+  // CIR: {{%.*}} = cir.atan {{%.*}} : !cir.vector<!cir.float x 4>
+  // LLVM: {{%.*}} = call <4 x float> @llvm.atan.v4f32(<4 x float> {{%.*}})
+  vf4 = __builtin_elementwise_atan(vf4);
+
+  // CIR: {{%.*}} = cir.atan {{%.*}} : !cir.vector<!cir.double x 4>
+  // LLVM: {{%.*}} = call <4 x double> @llvm.atan.v4f64(<4 x double> {{%.*}})
+  vd4 = __builtin_elementwise_atan(vd4);
 }
 
 void test_builtin_elementwise_exp(float f, double d, vfloat4 vf4,