// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
// RUN: %clang_cc1 -triple riscv32 -target-feature +v \
// RUN: -disable-O0-optnone \
// RUN:  -emit-llvm -o - %s | opt -S -passes=sroa | FileCheck %s
// RUN: %clang_cc1 -triple riscv64 -target-feature +v \
// RUN: -disable-O0-optnone \
// RUN:  -emit-llvm -o - %s | opt -S -passes=sroa | FileCheck %s

// REQUIRES: riscv-registered-target

#include <riscv_vector.h>

// Integer reductions — signed

// CHECK-LABEL: @reduce_add_i32(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[RDX_ADD:%.*]] = call i32 @llvm.vector.reduce.add.nxv2i32(<vscale x 2 x i32> [[A:%.*]])
// CHECK-NEXT:    ret i32 [[RDX_ADD]]
//
int reduce_add_i32(vint32m1_t a) {
  return __builtin_reduce_add(a);
}

// CHECK-LABEL: @reduce_and_i32(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[RDX_AND:%.*]] = call i32 @llvm.vector.reduce.and.nxv2i32(<vscale x 2 x i32> [[A:%.*]])
// CHECK-NEXT:    ret i32 [[RDX_AND]]
//
int reduce_and_i32(vint32m1_t a) {
  return __builtin_reduce_and(a);
}

// CHECK-LABEL: @reduce_or_i32(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[RDX_OR:%.*]] = call i32 @llvm.vector.reduce.or.nxv2i32(<vscale x 2 x i32> [[A:%.*]])
// CHECK-NEXT:    ret i32 [[RDX_OR]]
//
int reduce_or_i32(vint32m1_t a) {
  return __builtin_reduce_or(a);
}

// CHECK-LABEL: @reduce_xor_i32(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[RDX_XOR:%.*]] = call i32 @llvm.vector.reduce.xor.nxv2i32(<vscale x 2 x i32> [[A:%.*]])
// CHECK-NEXT:    ret i32 [[RDX_XOR]]
//
int reduce_xor_i32(vint32m1_t a) {
  return __builtin_reduce_xor(a);
}

// CHECK-LABEL: @reduce_smax_i32(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[RDX_MIN:%.*]] = call i32 @llvm.vector.reduce.smax.nxv2i32(<vscale x 2 x i32> [[A:%.*]])
// CHECK-NEXT:    ret i32 [[RDX_MIN]]
//
int reduce_smax_i32(vint32m1_t a) {
  return __builtin_reduce_max(a);
}

// CHECK-LABEL: @reduce_smin_i32(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[RDX_MIN:%.*]] = call i32 @llvm.vector.reduce.smin.nxv2i32(<vscale x 2 x i32> [[A:%.*]])
// CHECK-NEXT:    ret i32 [[RDX_MIN]]
//
int reduce_smin_i32(vint32m1_t a) {
  return __builtin_reduce_min(a);
}

// Integer reductions — unsigned

// CHECK-LABEL: @reduce_umax_u32(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[RDX_MIN:%.*]] = call i32 @llvm.vector.reduce.umax.nxv2i32(<vscale x 2 x i32> [[A:%.*]])
// CHECK-NEXT:    ret i32 [[RDX_MIN]]
//
unsigned reduce_umax_u32(vuint32m1_t a) {
  return __builtin_reduce_max(a);
}

// CHECK-LABEL: @reduce_umin_u32(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[RDX_MIN:%.*]] = call i32 @llvm.vector.reduce.umin.nxv2i32(<vscale x 2 x i32> [[A:%.*]])
// CHECK-NEXT:    ret i32 [[RDX_MIN]]
//
unsigned reduce_umin_u32(vuint32m1_t a) {
  return __builtin_reduce_min(a);
}

// Floating point reductions

// CHECK-LABEL: @reduce_fmax_f32(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[RDX_MIN:%.*]] = call float @llvm.vector.reduce.fmax.nxv2f32(<vscale x 2 x float> [[A:%.*]])
// CHECK-NEXT:    ret float [[RDX_MIN]]
//
float reduce_fmax_f32(vfloat32m1_t a) {
  return __builtin_reduce_max(a);
}

// CHECK-LABEL: @reduce_fmin_f32(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[RDX_MIN:%.*]] = call float @llvm.vector.reduce.fmin.nxv2f32(<vscale x 2 x float> [[A:%.*]])
// CHECK-NEXT:    ret float [[RDX_MIN]]
//
float reduce_fmin_f32(vfloat32m1_t a) {
  return __builtin_reduce_min(a);
}

// CHECK-LABEL: @reduce_fmaximum_f32(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[RDX_MAXIMUM:%.*]] = call float @llvm.vector.reduce.fmaximum.nxv2f32(<vscale x 2 x float> [[A:%.*]])
// CHECK-NEXT:    ret float [[RDX_MAXIMUM]]
//
float reduce_fmaximum_f32(vfloat32m1_t a) {
  return __builtin_reduce_maximum(a);
}

// CHECK-LABEL: @reduce_fminimum_f32(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[RDX_MINIMUM:%.*]] = call float @llvm.vector.reduce.fminimum.nxv2f32(<vscale x 2 x float> [[A:%.*]])
// CHECK-NEXT:    ret float [[RDX_MINIMUM]]
//
float reduce_fminimum_f32(vfloat32m1_t a) {
  return __builtin_reduce_minimum(a);
}

// Different LMUL

// CHECK-LABEL: @reduce_add_i32_m8(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[RDX_ADD:%.*]] = call i32 @llvm.vector.reduce.add.nxv16i32(<vscale x 16 x i32> [[A:%.*]])
// CHECK-NEXT:    ret i32 [[RDX_ADD]]
//
int reduce_add_i32_m8(vint32m8_t a) {
  return __builtin_reduce_add(a);
}

// Different element width

// CHECK-LABEL: @reduce_add_i64(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[RDX_ADD:%.*]] = call i64 @llvm.vector.reduce.add.nxv1i64(<vscale x 1 x i64> [[A:%.*]])
// CHECK-NEXT:    ret i64 [[RDX_ADD]]
//
long long reduce_add_i64(vint64m1_t a) {
  return __builtin_reduce_add(a);
}

// CHECK-LABEL: @reduce_mul_i32(
// CHECK-NEXT:  entry:
// CHECK-NEXT:    [[RDX_MUL:%.*]] = call i32 @llvm.vector.reduce.mul.nxv2i32(<vscale x 2 x i32> [[A:%.*]])
// CHECK-NEXT:    ret i32 [[RDX_MUL]]
//
int reduce_mul_i32(vint32m1_t a) {
  return __builtin_reduce_mul(a);
}