// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
// REQUIRES: amdgpu-registered-target

// RUN: %clang_cc1 -include __clang_hip_runtime_wrapper.h      \
// RUN:   -internal-isystem %S/../../lib/Headers/cuda_wrappers \
// RUN:   -internal-isystem %S/Inputs/include \
// RUN:   -triple amdgcn-amd-amdhsa -aux-triple x86_64-unknown-unknown \
// RUN:   -target-cpu gfx906 -emit-llvm %s -fcuda-is-device -O1 -o - \
// RUN:   -D__HIPCC_RTC__ | FileCheck -check-prefixes=DEFAULT,CORRECT-DIV-SQRT %s

// Check that we end up with fast math flags set on intrinsic calls
// RUN: %clang_cc1 -include __clang_hip_runtime_wrapper.h \
// RUN:   -internal-isystem %S/../../lib/Headers/cuda_wrappers \
// RUN:   -internal-isystem %S/Inputs/include \
// RUN:   -triple amdgcn-amd-amdhsa -aux-triple x86_64-unknown-unknown \
// RUN:   -target-cpu gfx906 -emit-llvm %s -fcuda-is-device -O1 -menable-no-infs \
// RUN:   -menable-no-nans -o - \
// RUN:   -D__HIPCC_RTC__ | FileCheck -check-prefix=FINITEONLY %s

// Check that we end up with fpmath metadata set on sqrt calls
// RUN: %clang_cc1 -include __clang_hip_runtime_wrapper.h \
// RUN:   -internal-isystem %S/../../lib/Headers/cuda_wrappers \
// RUN:   -internal-isystem %S/Inputs/include \
// RUN:   -triple amdgcn-amd-amdhsa -aux-triple x86_64-unknown-unknown \
// RUN:   -target-cpu gfx906 -emit-llvm %s -fcuda-is-device -O1 \
// RUN:   -fno-hip-fp32-correctly-rounded-divide-sqrt -o - \
// RUN:   -D__HIPCC_RTC__ | FileCheck -check-prefixes=DEFAULT,NO-CORRECT-DIV-SQRT %s

// DEFAULT-LABEL: @test_fma_f16(
// DEFAULT-NEXT:  entry:
// DEFAULT-NEXT:    [[TMP0:%.*]] = tail call contract noundef half @llvm.fma.f16(half [[X:%.*]], half [[Y:%.*]], half [[Z:%.*]])
// DEFAULT-NEXT:    ret half [[TMP0]]
//
// FINITEONLY-LABEL: @test_fma_f16(
// FINITEONLY-NEXT:  entry:
// FINITEONLY-NEXT:    [[TMP0:%.*]] = tail call nnan ninf contract noundef nofpclass(nan inf) half @llvm.fma.f16(half nofpclass(nan inf) [[X:%.*]], half nofpclass(nan inf) [[Y:%.*]], half nofpclass(nan inf) [[Z:%.*]])
// FINITEONLY-NEXT:    ret half [[TMP0]]
//
extern "C" __device__ _Float16 test_fma_f16(_Float16 x, _Float16 y,
                                            _Float16 z) {
  return fma(x, y, z);
}

// DEFAULT-LABEL: @test_pow_f16(
// DEFAULT-NEXT:  entry:
// DEFAULT-NEXT:    [[CALL_I:%.*]] = tail call contract noundef half @__ocml_pown_f16(half noundef [[X:%.*]], i32 noundef [[Y:%.*]]) #[[ATTR9:[0-9]+]]
// DEFAULT-NEXT:    ret half [[CALL_I]]
//
// FINITEONLY-LABEL: @test_pow_f16(
// FINITEONLY-NEXT:  entry:
// FINITEONLY-NEXT:    [[CALL_I:%.*]] = tail call nnan ninf contract noundef nofpclass(nan inf) half @__ocml_pown_f16(half noundef nofpclass(nan inf) [[X:%.*]], i32 noundef [[Y:%.*]]) #[[ATTR9:[0-9]+]]
// FINITEONLY-NEXT:    ret half [[CALL_I]]
//
extern "C" __device__ _Float16 test_pow_f16(_Float16 x, int y) {
  return pow(x, y);
}

// DEFAULT-LABEL: @test_fabs_f32(
// DEFAULT-NEXT:  entry:
// DEFAULT-NEXT:    [[TMP0:%.*]] = tail call contract noundef float @llvm.fabs.f32(float [[X:%.*]])
// DEFAULT-NEXT:    ret float [[TMP0]]
//
// FINITEONLY-LABEL: @test_fabs_f32(
// FINITEONLY-NEXT:  entry:
// FINITEONLY-NEXT:    [[TMP0:%.*]] = tail call nnan ninf contract noundef nofpclass(nan inf) float @llvm.fabs.f32(float nofpclass(nan inf) [[X:%.*]])
// FINITEONLY-NEXT:    ret float [[TMP0]]
//
extern "C" __device__ float test_fabs_f32(float x) {
  return fabs(x);
}

// DEFAULT-LABEL: @test_sin_f32(
// DEFAULT-NEXT:  entry:
// DEFAULT-NEXT:    [[CALL_I1:%.*]] = tail call contract noundef float @__ocml_sin_f32(float noundef [[X:%.*]]) #[[ATTR10:[0-9]+]]
// DEFAULT-NEXT:    ret float [[CALL_I1]]
//
// FINITEONLY-LABEL: @test_sin_f32(
// FINITEONLY-NEXT:  entry:
// FINITEONLY-NEXT:    [[CALL_I1:%.*]] = tail call nnan ninf contract noundef nofpclass(nan inf) float @__ocml_sin_f32(float noundef nofpclass(nan inf) [[X:%.*]]) #[[ATTR10:[0-9]+]]
// FINITEONLY-NEXT:    ret float [[CALL_I1]]
//
extern "C" __device__ float test_sin_f32(float x) {
  return sin(x);
}

// DEFAULT-LABEL: @test_cos_f32(
// DEFAULT-NEXT:  entry:
// DEFAULT-NEXT:    [[CALL_I1:%.*]] = tail call contract noundef float @__ocml_cos_f32(float noundef [[X:%.*]]) #[[ATTR10]]
// DEFAULT-NEXT:    ret float [[CALL_I1]]
//
// FINITEONLY-LABEL: @test_cos_f32(
// FINITEONLY-NEXT:  entry:
// FINITEONLY-NEXT:    [[CALL_I1:%.*]] = tail call nnan ninf contract noundef nofpclass(nan inf) float @__ocml_cos_f32(float noundef nofpclass(nan inf) [[X:%.*]]) #[[ATTR10]]
// FINITEONLY-NEXT:    ret float [[CALL_I1]]
//
extern "C" __device__ float test_cos_f32(float x) {
  return cos(x);
}

// Check user defined type which can be converted to float and double but not
// specializes __numeric_type will not cause ambiguity diagnostics.
struct user_bfloat16 {
  __host__ __device__ user_bfloat16(float);
  operator float();
  operator double();
};

namespace user_namespace {
// DEFAULT-LABEL: @_ZN14user_namespace3fmaE13user_bfloat16S0_S0_(
// DEFAULT-NEXT:  entry:
// DEFAULT-NEXT:    ret void
//
// FINITEONLY-LABEL: @_ZN14user_namespace3fmaE13user_bfloat16S0_S0_(
// FINITEONLY-NEXT:  entry:
// FINITEONLY-NEXT:    ret void
//
  __device__ user_bfloat16 fma(const user_bfloat16 a, const user_bfloat16 b, const user_bfloat16 c) {
    return a;
  }

// DEFAULT-LABEL: @_ZN14user_namespace8test_fmaEv(
// DEFAULT-NEXT:  entry:
// DEFAULT-NEXT:    [[A:%.*]] = alloca [[STRUCT_USER_BFLOAT16:%.*]], align 1, addrspace(5)
// DEFAULT-NEXT:    [[B:%.*]] = alloca [[STRUCT_USER_BFLOAT16]], align 1, addrspace(5)
// DEFAULT-NEXT:    [[A_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[A]] to ptr
// DEFAULT-NEXT:    [[B_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[B]] to ptr
// DEFAULT-NEXT:    call void @llvm.lifetime.start.p5(ptr addrspace(5) [[A]]) #[[ATTR11:[0-9]+]]
// DEFAULT-NEXT:    call void @_ZN13user_bfloat16C1Ef(ptr noundef nonnull align 1 dereferenceable(1) [[A_ASCAST]], float noundef 1.000000e+00) #[[ATTR10]]
// DEFAULT-NEXT:    call void @llvm.lifetime.start.p5(ptr addrspace(5) [[B]]) #[[ATTR11]]
// DEFAULT-NEXT:    call void @_ZN13user_bfloat16C1Ef(ptr noundef nonnull align 1 dereferenceable(1) [[B_ASCAST]], float noundef 2.000000e+00) #[[ATTR10]]
// DEFAULT-NEXT:    call void @llvm.lifetime.end.p5(ptr addrspace(5) [[B]]) #[[ATTR11]]
// DEFAULT-NEXT:    call void @llvm.lifetime.end.p5(ptr addrspace(5) [[A]]) #[[ATTR11]]
// DEFAULT-NEXT:    ret void
//
// FINITEONLY-LABEL: @_ZN14user_namespace8test_fmaEv(
// FINITEONLY-NEXT:  entry:
// FINITEONLY-NEXT:    [[A:%.*]] = alloca [[STRUCT_USER_BFLOAT16:%.*]], align 1, addrspace(5)
// FINITEONLY-NEXT:    [[B:%.*]] = alloca [[STRUCT_USER_BFLOAT16]], align 1, addrspace(5)
// FINITEONLY-NEXT:    [[A_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[A]] to ptr
// FINITEONLY-NEXT:    [[B_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[B]] to ptr
// FINITEONLY-NEXT:    call void @llvm.lifetime.start.p5(ptr addrspace(5) [[A]]) #[[ATTR11:[0-9]+]]
// FINITEONLY-NEXT:    call void @_ZN13user_bfloat16C1Ef(ptr noundef nonnull align 1 dereferenceable(1) [[A_ASCAST]], float noundef nofpclass(nan inf) 1.000000e+00) #[[ATTR10]]
// FINITEONLY-NEXT:    call void @llvm.lifetime.start.p5(ptr addrspace(5) [[B]]) #[[ATTR11]]
// FINITEONLY-NEXT:    call void @_ZN13user_bfloat16C1Ef(ptr noundef nonnull align 1 dereferenceable(1) [[B_ASCAST]], float noundef nofpclass(nan inf) 2.000000e+00) #[[ATTR10]]
// FINITEONLY-NEXT:    call void @llvm.lifetime.end.p5(ptr addrspace(5) [[B]]) #[[ATTR11]]
// FINITEONLY-NEXT:    call void @llvm.lifetime.end.p5(ptr addrspace(5) [[A]]) #[[ATTR11]]
// FINITEONLY-NEXT:    ret void
//
  __global__ void test_fma() {
    user_bfloat16 a = 1.0f, b = 2.0f;
    fma(a, b, b);
  }
}

// CORRECT-DIV-SQRT-LABEL: @test_sqrt_f32(
// CORRECT-DIV-SQRT-NEXT:  entry:
// CORRECT-DIV-SQRT-NEXT:    [[TMP0:%.*]] = tail call contract noundef float @llvm.sqrt.f32(float [[X:%.*]])
// CORRECT-DIV-SQRT-NEXT:    ret float [[TMP0]]
//
// FINITEONLY-LABEL: @test_sqrt_f32(
// FINITEONLY-NEXT:  entry:
// FINITEONLY-NEXT:    [[TMP0:%.*]] = tail call nnan ninf contract noundef nofpclass(nan inf) float @llvm.sqrt.f32(float nofpclass(nan inf) [[X:%.*]])
// FINITEONLY-NEXT:    ret float [[TMP0]]
//
// NO-CORRECT-DIV-SQRT-LABEL: @test_sqrt_f32(
// NO-CORRECT-DIV-SQRT-NEXT:  entry:
// NO-CORRECT-DIV-SQRT-NEXT:    [[TMP0:%.*]] = tail call contract noundef float @llvm.sqrt.f32(float [[X:%.*]]), !fpmath [[META8:![0-9]+]]
// NO-CORRECT-DIV-SQRT-NEXT:    ret float [[TMP0]]
//
extern "C" __device__ float test_sqrt_f32(float x) {
  return sqrt(x);
}

// DEFAULT-LABEL: @test_sqrt_f64(
// DEFAULT-NEXT:  entry:
// DEFAULT-NEXT:    [[TMP0:%.*]] = tail call contract noundef double @llvm.sqrt.f64(double [[X:%.*]])
// DEFAULT-NEXT:    ret double [[TMP0]]
//
// FINITEONLY-LABEL: @test_sqrt_f64(
// FINITEONLY-NEXT:  entry:
// FINITEONLY-NEXT:    [[TMP0:%.*]] = tail call nnan ninf contract noundef nofpclass(nan inf) double @llvm.sqrt.f64(double nofpclass(nan inf) [[X:%.*]])
// FINITEONLY-NEXT:    ret double [[TMP0]]
//
extern "C" __device__ double test_sqrt_f64(double x) {
  return sqrt(x);
}