// Auto-generated file. Do not edit! // Template: src/f32-clamp/neon.c.in // Generator: tools/xngen // // Copyright 2020 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. #include #include #include #include void xnn_f32_clamp_ukernel__neon_x4( size_t n, const float* x, float* y, const union xnn_f32_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_DISABLE_TSAN { assert(n != 0); assert(n % sizeof(float) == 0); assert(x != NULL); assert(y != NULL); const float32x4_t vy_min = vld1q_dup_f32(¶ms->scalar.min); const float32x4_t vy_max = vld1q_dup_f32(¶ms->scalar.max); for (; n >= 4 * sizeof(float); n -= 4 * sizeof(float)) { float32x4_t vacc0123 = vld1q_f32(x); x += 4; vacc0123 = vmaxq_f32(vacc0123, vy_min); vacc0123 = vminq_f32(vacc0123, vy_max); vst1q_f32(y, vacc0123); y += 4; } if XNN_UNLIKELY(n != 0) { float32x4_t vacc = vld1q_f32(x); vacc = vmaxq_f32(vacc, vy_min); vacc = vminq_f32(vacc, vy_max); float32x2_t vacc_lo = vget_low_f32(vacc); if (n & (2 * sizeof(float))) { vst1_f32(y, vacc_lo); y += 2; vacc_lo = vget_high_f32(vacc); } if (n & (1 * sizeof(float))) { vst1_lane_f32(y, vacc_lo, 0); } } }