// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
// RUN: %clang_cc1 %s -triple x86_64-unknown-unknown -target-feature +avx512f -target-feature +avx512vl -target-feature +avx512fp16 -emit-llvm -o - | FileCheck %s

typedef double double8 __attribute__((ext_vector_type(8)));

// CHECK-LABEL: @fadd1(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[A_ADDR:%.*]] = alloca <8 x double>, align 64
// CHECK-NEXT:    [[B_ADDR:%.*]] = alloca double, align 8
// CHECK-NEXT:    store <8 x double> [[A:%.*]], ptr [[A_ADDR]], align 64
// CHECK-NEXT:    store double [[B:%.*]], ptr [[B_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load <8 x double>, ptr [[A_ADDR]], align 64
// CHECK-NEXT:    [[TMP1:%.*]] = call reassoc double @llvm.vector.reduce.fadd.v8f64(double 0.000000e+00, <8 x double> [[TMP0]])
// CHECK-NEXT:    [[TMP2:%.*]] = load double, ptr [[B_ADDR]], align 8
// CHECK-NEXT:    [[ADD:%.*]] = fadd double [[TMP1]], [[TMP2]]
// CHECK-NEXT:    ret double [[ADD]]
//
double fadd1(double8 a, double b) {
  return __builtin_ia32_reduce_fadd_pd512(0.0, a) + b;
}

#pragma clang fp reassociate(on)
// CHECK-LABEL: @fadd2(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[A_ADDR:%.*]] = alloca <8 x double>, align 64
// CHECK-NEXT:    [[B_ADDR:%.*]] = alloca double, align 8
// CHECK-NEXT:    store <8 x double> [[A:%.*]], ptr [[A_ADDR]], align 64
// CHECK-NEXT:    store double [[B:%.*]], ptr [[B_ADDR]], align 8
// CHECK-NEXT:    [[TMP0:%.*]] = load <8 x double>, ptr [[A_ADDR]], align 64
// CHECK-NEXT:    [[TMP1:%.*]] = call reassoc double @llvm.vector.reduce.fadd.v8f64(double 0.000000e+00, <8 x double> [[TMP0]])
// CHECK-NEXT:    [[TMP2:%.*]] = load double, ptr [[B_ADDR]], align 8
// CHECK-NEXT:    [[ADD:%.*]] = fadd reassoc double [[TMP1]], [[TMP2]]
// CHECK-NEXT:    ret double [[ADD]]
//
double fadd2(double8 a, double b) {
  return __builtin_ia32_reduce_fadd_pd512(0.0, a) + b;
}

typedef float float16 __attribute__((ext_vector_type(16)));

#pragma clang fp reassociate(off)
// CHECK-LABEL: @fmul1(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[A_ADDR:%.*]] = alloca <16 x float>, align 64
// CHECK-NEXT:    [[B_ADDR:%.*]] = alloca float, align 4
// CHECK-NEXT:    store <16 x float> [[A:%.*]], ptr [[A_ADDR]], align 64
// CHECK-NEXT:    store float [[B:%.*]], ptr [[B_ADDR]], align 4
// CHECK-NEXT:    [[TMP0:%.*]] = load <16 x float>, ptr [[A_ADDR]], align 64
// CHECK-NEXT:    [[TMP1:%.*]] = call reassoc float @llvm.vector.reduce.fmul.v16f32(float 1.000000e+00, <16 x float> [[TMP0]])
// CHECK-NEXT:    [[TMP2:%.*]] = load float, ptr [[B_ADDR]], align 4
// CHECK-NEXT:    [[ADD:%.*]] = fadd float [[TMP1]], [[TMP2]]
// CHECK-NEXT:    ret float [[ADD]]
//
float fmul1(float16 a, float b) {
  return __builtin_ia32_reduce_fmul_ps512(1.0f, a) + b;
}

typedef _Float16 half8 __attribute__((ext_vector_type(8)));

// CHECK-LABEL: @fmax1(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[A_ADDR:%.*]] = alloca <8 x half>, align 16
// CHECK-NEXT:    [[B_ADDR:%.*]] = alloca half, align 2
// CHECK-NEXT:    store <8 x half> [[A:%.*]], ptr [[A_ADDR]], align 16
// CHECK-NEXT:    store half [[B:%.*]], ptr [[B_ADDR]], align 2
// CHECK-NEXT:    [[TMP0:%.*]] = load <8 x half>, ptr [[A_ADDR]], align 16
// CHECK-NEXT:    [[TMP1:%.*]] = call nnan half @llvm.vector.reduce.fmax.v8f16(<8 x half> [[TMP0]])
// CHECK-NEXT:    [[TMP2:%.*]] = load half, ptr [[B_ADDR]], align 2
// CHECK-NEXT:    [[ADD:%.*]] = fadd half [[TMP1]], [[TMP2]]
// CHECK-NEXT:    ret half [[ADD]]
//
_Float16 fmax1(half8 a, _Float16 b) {
  return __builtin_ia32_reduce_fmax_ph128(a) + b;
}

typedef _Float16 half16 __attribute__((ext_vector_type(16)));

// CHECK-LABEL: @fmin1(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[A_ADDR:%.*]] = alloca <16 x half>, align 32
// CHECK-NEXT:    [[B_ADDR:%.*]] = alloca half, align 2
// CHECK-NEXT:    store <16 x half> [[A:%.*]], ptr [[A_ADDR]], align 32
// CHECK-NEXT:    store half [[B:%.*]], ptr [[B_ADDR]], align 2
// CHECK-NEXT:    [[TMP0:%.*]] = load <16 x half>, ptr [[A_ADDR]], align 32
// CHECK-NEXT:    [[TMP1:%.*]] = call nnan half @llvm.vector.reduce.fmin.v16f16(<16 x half> [[TMP0]])
// CHECK-NEXT:    [[TMP2:%.*]] = load half, ptr [[B_ADDR]], align 2
// CHECK-NEXT:    [[ADD:%.*]] = fadd half [[TMP1]], [[TMP2]]
// CHECK-NEXT:    ret half [[ADD]]
//
_Float16 fmin1(half16 a, _Float16 b) {
  return __builtin_ia32_reduce_fmin_ph256(a) + b;
}