sglang_v0.5.2/pytorch_2.8.0/third_party/XNNPACK/test/f16-rsum.cc

// Copyright 2023 Google LLC
//
// This source code is licensed under the BSD-style license found in the
// LICENSE file in the root directory of this source tree.
//
// Auto-generated file. Do not edit!
//   Specification: test/f16-rsum.yaml
//   Generator: tools/generate-reduce-test.py


#include <gtest/gtest.h>
#include "xnnpack/common.h"
#include "xnnpack/isa-checks.h"
#include "xnnpack/microparams-init.h"
#include "xnnpack/reduce.h"
#include "rsum-microkernel-tester.h"


#if XNN_ENABLE_ARM_FP16_VECTOR && (XNN_ARCH_ARM || XNN_ARCH_ARM64)
  TEST(F16_RSUM__NEONFP16ARITH_U8, batch_eq_8) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    RSumMicrokernelTester()
      .batch_size(8)
      .Test(xnn_f16_rsum_ukernel__neonfp16arith_u8, xnn_init_f16_scale_scalar_params);
  }

  TEST(F16_RSUM__NEONFP16ARITH_U8, batch_div_8) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    for (size_t batch_size = 16; batch_size < 80; batch_size += 8) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__neonfp16arith_u8, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__NEONFP16ARITH_U8, batch_lt_8) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    for (size_t batch_size = 1; batch_size < 8; batch_size++) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__neonfp16arith_u8, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__NEONFP16ARITH_U8, batch_gt_8) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    for (size_t batch_size = 9; batch_size < 16; batch_size++) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__neonfp16arith_u8, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__NEONFP16ARITH_U8, scale) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    for (float scale = 0.3f; scale < 5.0f; scale *= 3.0f) {
      RSumMicrokernelTester()
        .batch_size(9)
        .scale(scale)
        .Test(xnn_f16_rsum_ukernel__neonfp16arith_u8, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__NEONFP16ARITH_U8, overflow_accumulator) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    RSumMicrokernelTester()
      .batch_size(1024)
      .Test(xnn_f16_rsum_ukernel__neonfp16arith_u8, xnn_init_f16_scale_scalar_params);
  }
#endif  // XNN_ENABLE_ARM_FP16_VECTOR && (XNN_ARCH_ARM || XNN_ARCH_ARM64)


#if XNN_ENABLE_ARM_FP16_VECTOR && (XNN_ARCH_ARM || XNN_ARCH_ARM64)
  TEST(F16_RSUM__NEONFP16ARITH_U16_ACC2, batch_eq_16) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    RSumMicrokernelTester()
      .batch_size(16)
      .Test(xnn_f16_rsum_ukernel__neonfp16arith_u16_acc2, xnn_init_f16_scale_scalar_params);
  }

  TEST(F16_RSUM__NEONFP16ARITH_U16_ACC2, batch_div_16) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    for (size_t batch_size = 32; batch_size < 160; batch_size += 16) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__neonfp16arith_u16_acc2, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__NEONFP16ARITH_U16_ACC2, batch_lt_16) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    for (size_t batch_size = 1; batch_size < 16; batch_size++) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__neonfp16arith_u16_acc2, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__NEONFP16ARITH_U16_ACC2, batch_gt_16) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    for (size_t batch_size = 17; batch_size < 32; batch_size++) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__neonfp16arith_u16_acc2, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__NEONFP16ARITH_U16_ACC2, scale) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    for (float scale = 0.3f; scale < 5.0f; scale *= 3.0f) {
      RSumMicrokernelTester()
        .batch_size(17)
        .scale(scale)
        .Test(xnn_f16_rsum_ukernel__neonfp16arith_u16_acc2, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__NEONFP16ARITH_U16_ACC2, overflow_accumulator) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    RSumMicrokernelTester()
      .batch_size(2048)
      .Test(xnn_f16_rsum_ukernel__neonfp16arith_u16_acc2, xnn_init_f16_scale_scalar_params);
  }
#endif  // XNN_ENABLE_ARM_FP16_VECTOR && (XNN_ARCH_ARM || XNN_ARCH_ARM64)


#if XNN_ENABLE_ARM_FP16_VECTOR && (XNN_ARCH_ARM || XNN_ARCH_ARM64)
  TEST(F16_RSUM__NEONFP16ARITH_U24_ACC3, batch_eq_24) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    RSumMicrokernelTester()
      .batch_size(24)
      .Test(xnn_f16_rsum_ukernel__neonfp16arith_u24_acc3, xnn_init_f16_scale_scalar_params);
  }

  TEST(F16_RSUM__NEONFP16ARITH_U24_ACC3, batch_div_24) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    for (size_t batch_size = 48; batch_size < 240; batch_size += 24) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__neonfp16arith_u24_acc3, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__NEONFP16ARITH_U24_ACC3, batch_lt_24) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    for (size_t batch_size = 1; batch_size < 24; batch_size++) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__neonfp16arith_u24_acc3, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__NEONFP16ARITH_U24_ACC3, batch_gt_24) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    for (size_t batch_size = 25; batch_size < 48; batch_size++) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__neonfp16arith_u24_acc3, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__NEONFP16ARITH_U24_ACC3, scale) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    for (float scale = 0.3f; scale < 5.0f; scale *= 3.0f) {
      RSumMicrokernelTester()
        .batch_size(25)
        .scale(scale)
        .Test(xnn_f16_rsum_ukernel__neonfp16arith_u24_acc3, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__NEONFP16ARITH_U24_ACC3, overflow_accumulator) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    RSumMicrokernelTester()
      .batch_size(3072)
      .Test(xnn_f16_rsum_ukernel__neonfp16arith_u24_acc3, xnn_init_f16_scale_scalar_params);
  }
#endif  // XNN_ENABLE_ARM_FP16_VECTOR && (XNN_ARCH_ARM || XNN_ARCH_ARM64)


#if XNN_ENABLE_ARM_FP16_VECTOR && (XNN_ARCH_ARM || XNN_ARCH_ARM64)
  TEST(F16_RSUM__NEONFP16ARITH_U32_ACC2, batch_eq_32) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    RSumMicrokernelTester()
      .batch_size(32)
      .Test(xnn_f16_rsum_ukernel__neonfp16arith_u32_acc2, xnn_init_f16_scale_scalar_params);
  }

  TEST(F16_RSUM__NEONFP16ARITH_U32_ACC2, batch_div_32) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    for (size_t batch_size = 64; batch_size < 320; batch_size += 32) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__neonfp16arith_u32_acc2, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__NEONFP16ARITH_U32_ACC2, batch_lt_32) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    for (size_t batch_size = 1; batch_size < 32; batch_size++) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__neonfp16arith_u32_acc2, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__NEONFP16ARITH_U32_ACC2, batch_gt_32) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    for (size_t batch_size = 33; batch_size < 64; batch_size++) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__neonfp16arith_u32_acc2, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__NEONFP16ARITH_U32_ACC2, scale) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    for (float scale = 0.3f; scale < 5.0f; scale *= 3.0f) {
      RSumMicrokernelTester()
        .batch_size(33)
        .scale(scale)
        .Test(xnn_f16_rsum_ukernel__neonfp16arith_u32_acc2, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__NEONFP16ARITH_U32_ACC2, overflow_accumulator) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    RSumMicrokernelTester()
      .batch_size(4096)
      .Test(xnn_f16_rsum_ukernel__neonfp16arith_u32_acc2, xnn_init_f16_scale_scalar_params);
  }
#endif  // XNN_ENABLE_ARM_FP16_VECTOR && (XNN_ARCH_ARM || XNN_ARCH_ARM64)


#if XNN_ENABLE_ARM_FP16_VECTOR && (XNN_ARCH_ARM || XNN_ARCH_ARM64)
  TEST(F16_RSUM__NEONFP16ARITH_U32_ACC4, batch_eq_32) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    RSumMicrokernelTester()
      .batch_size(32)
      .Test(xnn_f16_rsum_ukernel__neonfp16arith_u32_acc4, xnn_init_f16_scale_scalar_params);
  }

  TEST(F16_RSUM__NEONFP16ARITH_U32_ACC4, batch_div_32) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    for (size_t batch_size = 64; batch_size < 320; batch_size += 32) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__neonfp16arith_u32_acc4, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__NEONFP16ARITH_U32_ACC4, batch_lt_32) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    for (size_t batch_size = 1; batch_size < 32; batch_size++) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__neonfp16arith_u32_acc4, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__NEONFP16ARITH_U32_ACC4, batch_gt_32) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    for (size_t batch_size = 33; batch_size < 64; batch_size++) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__neonfp16arith_u32_acc4, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__NEONFP16ARITH_U32_ACC4, scale) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    for (float scale = 0.3f; scale < 5.0f; scale *= 3.0f) {
      RSumMicrokernelTester()
        .batch_size(33)
        .scale(scale)
        .Test(xnn_f16_rsum_ukernel__neonfp16arith_u32_acc4, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__NEONFP16ARITH_U32_ACC4, overflow_accumulator) {
    TEST_REQUIRES_ARM_NEON_FP16_ARITH;
    RSumMicrokernelTester()
      .batch_size(4096)
      .Test(xnn_f16_rsum_ukernel__neonfp16arith_u32_acc4, xnn_init_f16_scale_scalar_params);
  }
#endif  // XNN_ENABLE_ARM_FP16_VECTOR && (XNN_ARCH_ARM || XNN_ARCH_ARM64)


#if XNN_ENABLE_AVX512FP16 && (XNN_ARCH_X86 || XNN_ARCH_X86_64)
  TEST(F16_RSUM__AVX512FP16_U32, batch_eq_32) {
    TEST_REQUIRES_X86_AVX512FP16;
    RSumMicrokernelTester()
      .batch_size(32)
      .Test(xnn_f16_rsum_ukernel__avx512fp16_u32, xnn_init_f16_scale_scalar_params);
  }

  TEST(F16_RSUM__AVX512FP16_U32, batch_div_32) {
    TEST_REQUIRES_X86_AVX512FP16;
    for (size_t batch_size = 64; batch_size < 320; batch_size += 32) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__avx512fp16_u32, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__AVX512FP16_U32, batch_lt_32) {
    TEST_REQUIRES_X86_AVX512FP16;
    for (size_t batch_size = 1; batch_size < 32; batch_size++) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__avx512fp16_u32, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__AVX512FP16_U32, batch_gt_32) {
    TEST_REQUIRES_X86_AVX512FP16;
    for (size_t batch_size = 33; batch_size < 64; batch_size++) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__avx512fp16_u32, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__AVX512FP16_U32, scale) {
    TEST_REQUIRES_X86_AVX512FP16;
    for (float scale = 0.3f; scale < 5.0f; scale *= 3.0f) {
      RSumMicrokernelTester()
        .batch_size(33)
        .scale(scale)
        .Test(xnn_f16_rsum_ukernel__avx512fp16_u32, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__AVX512FP16_U32, overflow_accumulator) {
    TEST_REQUIRES_X86_AVX512FP16;
    RSumMicrokernelTester()
      .batch_size(4096)
      .Test(xnn_f16_rsum_ukernel__avx512fp16_u32, xnn_init_f16_scale_scalar_params);
  }
#endif  // XNN_ENABLE_AVX512FP16 && (XNN_ARCH_X86 || XNN_ARCH_X86_64)


#if XNN_ENABLE_AVX512FP16 && (XNN_ARCH_X86 || XNN_ARCH_X86_64)
  TEST(F16_RSUM__AVX512FP16_U64_ACC2, batch_eq_64) {
    TEST_REQUIRES_X86_AVX512FP16;
    RSumMicrokernelTester()
      .batch_size(64)
      .Test(xnn_f16_rsum_ukernel__avx512fp16_u64_acc2, xnn_init_f16_scale_scalar_params);
  }

  TEST(F16_RSUM__AVX512FP16_U64_ACC2, batch_div_64) {
    TEST_REQUIRES_X86_AVX512FP16;
    for (size_t batch_size = 128; batch_size < 640; batch_size += 64) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__avx512fp16_u64_acc2, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__AVX512FP16_U64_ACC2, batch_lt_64) {
    TEST_REQUIRES_X86_AVX512FP16;
    for (size_t batch_size = 1; batch_size < 64; batch_size++) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__avx512fp16_u64_acc2, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__AVX512FP16_U64_ACC2, batch_gt_64) {
    TEST_REQUIRES_X86_AVX512FP16;
    for (size_t batch_size = 65; batch_size < 128; batch_size++) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__avx512fp16_u64_acc2, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__AVX512FP16_U64_ACC2, scale) {
    TEST_REQUIRES_X86_AVX512FP16;
    for (float scale = 0.3f; scale < 5.0f; scale *= 3.0f) {
      RSumMicrokernelTester()
        .batch_size(65)
        .scale(scale)
        .Test(xnn_f16_rsum_ukernel__avx512fp16_u64_acc2, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__AVX512FP16_U64_ACC2, overflow_accumulator) {
    TEST_REQUIRES_X86_AVX512FP16;
    RSumMicrokernelTester()
      .batch_size(8192)
      .Test(xnn_f16_rsum_ukernel__avx512fp16_u64_acc2, xnn_init_f16_scale_scalar_params);
  }
#endif  // XNN_ENABLE_AVX512FP16 && (XNN_ARCH_X86 || XNN_ARCH_X86_64)


#if XNN_ENABLE_AVX512FP16 && (XNN_ARCH_X86 || XNN_ARCH_X86_64)
  TEST(F16_RSUM__AVX512FP16_U96_ACC3, batch_eq_96) {
    TEST_REQUIRES_X86_AVX512FP16;
    RSumMicrokernelTester()
      .batch_size(96)
      .Test(xnn_f16_rsum_ukernel__avx512fp16_u96_acc3, xnn_init_f16_scale_scalar_params);
  }

  TEST(F16_RSUM__AVX512FP16_U96_ACC3, batch_div_96) {
    TEST_REQUIRES_X86_AVX512FP16;
    for (size_t batch_size = 192; batch_size < 960; batch_size += 96) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__avx512fp16_u96_acc3, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__AVX512FP16_U96_ACC3, batch_lt_96) {
    TEST_REQUIRES_X86_AVX512FP16;
    for (size_t batch_size = 1; batch_size < 96; batch_size++) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__avx512fp16_u96_acc3, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__AVX512FP16_U96_ACC3, batch_gt_96) {
    TEST_REQUIRES_X86_AVX512FP16;
    for (size_t batch_size = 97; batch_size < 192; batch_size++) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__avx512fp16_u96_acc3, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__AVX512FP16_U96_ACC3, scale) {
    TEST_REQUIRES_X86_AVX512FP16;
    for (float scale = 0.3f; scale < 5.0f; scale *= 3.0f) {
      RSumMicrokernelTester()
        .batch_size(97)
        .scale(scale)
        .Test(xnn_f16_rsum_ukernel__avx512fp16_u96_acc3, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__AVX512FP16_U96_ACC3, overflow_accumulator) {
    TEST_REQUIRES_X86_AVX512FP16;
    RSumMicrokernelTester()
      .batch_size(12288)
      .Test(xnn_f16_rsum_ukernel__avx512fp16_u96_acc3, xnn_init_f16_scale_scalar_params);
  }
#endif  // XNN_ENABLE_AVX512FP16 && (XNN_ARCH_X86 || XNN_ARCH_X86_64)


#if XNN_ENABLE_AVX512FP16 && (XNN_ARCH_X86 || XNN_ARCH_X86_64)
  TEST(F16_RSUM__AVX512FP16_U128_ACC2, batch_eq_128) {
    TEST_REQUIRES_X86_AVX512FP16;
    RSumMicrokernelTester()
      .batch_size(128)
      .Test(xnn_f16_rsum_ukernel__avx512fp16_u128_acc2, xnn_init_f16_scale_scalar_params);
  }

  TEST(F16_RSUM__AVX512FP16_U128_ACC2, batch_div_128) {
    TEST_REQUIRES_X86_AVX512FP16;
    for (size_t batch_size = 256; batch_size < 1280; batch_size += 128) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__avx512fp16_u128_acc2, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__AVX512FP16_U128_ACC2, batch_lt_128) {
    TEST_REQUIRES_X86_AVX512FP16;
    for (size_t batch_size = 1; batch_size < 128; batch_size++) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__avx512fp16_u128_acc2, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__AVX512FP16_U128_ACC2, batch_gt_128) {
    TEST_REQUIRES_X86_AVX512FP16;
    for (size_t batch_size = 129; batch_size < 256; batch_size++) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__avx512fp16_u128_acc2, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__AVX512FP16_U128_ACC2, scale) {
    TEST_REQUIRES_X86_AVX512FP16;
    for (float scale = 0.3f; scale < 5.0f; scale *= 3.0f) {
      RSumMicrokernelTester()
        .batch_size(129)
        .scale(scale)
        .Test(xnn_f16_rsum_ukernel__avx512fp16_u128_acc2, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__AVX512FP16_U128_ACC2, overflow_accumulator) {
    TEST_REQUIRES_X86_AVX512FP16;
    RSumMicrokernelTester()
      .batch_size(16384)
      .Test(xnn_f16_rsum_ukernel__avx512fp16_u128_acc2, xnn_init_f16_scale_scalar_params);
  }
#endif  // XNN_ENABLE_AVX512FP16 && (XNN_ARCH_X86 || XNN_ARCH_X86_64)


#if XNN_ENABLE_AVX512FP16 && (XNN_ARCH_X86 || XNN_ARCH_X86_64)
  TEST(F16_RSUM__AVX512FP16_U128_ACC4, batch_eq_128) {
    TEST_REQUIRES_X86_AVX512FP16;
    RSumMicrokernelTester()
      .batch_size(128)
      .Test(xnn_f16_rsum_ukernel__avx512fp16_u128_acc4, xnn_init_f16_scale_scalar_params);
  }

  TEST(F16_RSUM__AVX512FP16_U128_ACC4, batch_div_128) {
    TEST_REQUIRES_X86_AVX512FP16;
    for (size_t batch_size = 256; batch_size < 1280; batch_size += 128) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__avx512fp16_u128_acc4, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__AVX512FP16_U128_ACC4, batch_lt_128) {
    TEST_REQUIRES_X86_AVX512FP16;
    for (size_t batch_size = 1; batch_size < 128; batch_size++) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__avx512fp16_u128_acc4, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__AVX512FP16_U128_ACC4, batch_gt_128) {
    TEST_REQUIRES_X86_AVX512FP16;
    for (size_t batch_size = 129; batch_size < 256; batch_size++) {
      RSumMicrokernelTester()
        .batch_size(batch_size)
        .Test(xnn_f16_rsum_ukernel__avx512fp16_u128_acc4, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__AVX512FP16_U128_ACC4, scale) {
    TEST_REQUIRES_X86_AVX512FP16;
    for (float scale = 0.3f; scale < 5.0f; scale *= 3.0f) {
      RSumMicrokernelTester()
        .batch_size(129)
        .scale(scale)
        .Test(xnn_f16_rsum_ukernel__avx512fp16_u128_acc4, xnn_init_f16_scale_scalar_params);
    }
  }

  TEST(F16_RSUM__AVX512FP16_U128_ACC4, overflow_accumulator) {
    TEST_REQUIRES_X86_AVX512FP16;
    RSumMicrokernelTester()
      .batch_size(16384)
      .Test(xnn_f16_rsum_ukernel__avx512fp16_u128_acc4, xnn_init_f16_scale_scalar_params);
  }
#endif  // XNN_ENABLE_AVX512FP16 && (XNN_ARCH_X86 || XNN_ARCH_X86_64)