| // Auto-generated file. Do not edit! |
| // Template: src/f32-qs8-vcvt/neonv8.c.in |
| // Generator: tools/xngen |
| // |
| // Copyright 2021 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 <assert.h> |
| |
| #include <arm_neon.h> |
| |
| #include <xnnpack/common.h> |
| #include <xnnpack/intrinsics-polyfill.h> |
| #include <xnnpack/vcvt.h> |
| |
| |
| void xnn_f32_qs8_vcvt_ukernel__neonv8_x16( |
| size_t n, |
| const float* x, |
| int8_t* y, |
| const union xnn_f32_qs8_cvt_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS |
| { |
| assert(n != 0); |
| assert(n % sizeof(float) == 0); |
| assert(x != NULL); |
| assert(y != NULL); |
| |
| const float32x4_t vscale = vld1q_dup_f32(¶ms->neonv8.scale); |
| const int16x8_t voutput_zero_point = vld1q_dup_s16(¶ms->neonv8.output_zero_point); |
| const int8x16_t voutput_min = vld1q_dup_s8(¶ms->neonv8.output_min); |
| const int8x16_t voutput_max = vld1q_dup_s8(¶ms->neonv8.output_max); |
| for (; n >= 16 * sizeof(float); n -= 16 * sizeof(float)) { |
| float32x4_t vx0123 = vld1q_f32(x); x += 4; |
| float32x4_t vx4567 = vld1q_f32(x); x += 4; |
| float32x4_t vx89AB = vld1q_f32(x); x += 4; |
| float32x4_t vxCDEF = vld1q_f32(x); x += 4; |
| |
| vx0123 = vmulq_f32(vx0123, vscale); |
| vx4567 = vmulq_f32(vx4567, vscale); |
| vx89AB = vmulq_f32(vx89AB, vscale); |
| vxCDEF = vmulq_f32(vxCDEF, vscale); |
| |
| const int32x4_t vacc0123 = vcvtnq_s32_f32(vx0123); |
| const int32x4_t vacc4567 = vcvtnq_s32_f32(vx4567); |
| const int32x4_t vacc89AB = vcvtnq_s32_f32(vx89AB); |
| const int32x4_t vaccCDEF = vcvtnq_s32_f32(vxCDEF); |
| |
| int16x8_t vacc01234567 = vcombine_s16(vqmovn_s32(vacc0123), vqmovn_s32(vacc4567)); |
| int16x8_t vacc89ABCDEF = vcombine_s16(vqmovn_s32(vacc89AB), vqmovn_s32(vaccCDEF)); |
| |
| vacc01234567 = vqaddq_s16(vacc01234567, voutput_zero_point); |
| vacc89ABCDEF = vqaddq_s16(vacc89ABCDEF, voutput_zero_point); |
| |
| int8x16_t vy0123456789ABCDEF = vcombine_s8(vqmovn_s16(vacc01234567), vqmovn_s16(vacc89ABCDEF)); |
| |
| vy0123456789ABCDEF = vmaxq_s8(vy0123456789ABCDEF, voutput_min); |
| |
| vy0123456789ABCDEF = vminq_s8(vy0123456789ABCDEF, voutput_max); |
| |
| vst1q_s8(y, vy0123456789ABCDEF); y += 16; |
| } |
| for (; n >= 8 * sizeof(float); n -= 8 * sizeof(float)) { |
| float32x4_t vx_lo = vld1q_f32(x); x += 4; |
| float32x4_t vx_hi = vld1q_f32(x); x += 4; |
| |
| vx_lo = vmulq_f32(vx_lo, vscale); |
| vx_hi = vmulq_f32(vx_hi, vscale); |
| |
| const int32x4_t vacc_lo = vcvtnq_s32_f32(vx_lo); |
| const int32x4_t vacc_hi = vcvtnq_s32_f32(vx_hi); |
| |
| int16x8_t vacc = vcombine_s16(vqmovn_s32(vacc_lo), vqmovn_s32(vacc_hi)); |
| vacc = vqaddq_s16(vacc, voutput_zero_point); |
| |
| int8x8_t vy = vqmovn_s16(vacc); |
| vy = vmax_s8(vy, vget_low_s8(voutput_min)); |
| vy = vmin_s8(vy, vget_low_s8(voutput_max)); |
| vst1_s8(y, vy); y += 8; |
| } |
| if XNN_UNLIKELY(n != 0) { |
| assert(n >= 1 * sizeof(float)); |
| assert(n <= 7 * sizeof(float)); |
| float32x4_t vx_lo = vld1q_f32(x); |
| const float* x_hi = (const float*) ((uintptr_t) x + (n & (4 * sizeof(float)))); |
| float32x4_t vx_hi = vld1q_f32(x_hi); |
| |
| vx_lo = vmulq_f32(vx_lo, vscale); |
| vx_hi = vmulq_f32(vx_hi, vscale); |
| |
| const int32x4_t vacc_lo = vcvtnq_s32_f32(vx_lo); |
| const int32x4_t vacc_hi = vcvtnq_s32_f32(vx_hi); |
| |
| int16x8_t vacc = vcombine_s16(vqmovn_s32(vacc_lo), vqmovn_s32(vacc_hi)); |
| vacc = vqaddq_s16(vacc, voutput_zero_point); |
| |
| int8x8_t vy = vqmovn_s16(vacc); |
| vy = vmax_s8(vy, vget_low_s8(voutput_min)); |
| vy = vmin_s8(vy, vget_low_s8(voutput_max)); |
| |
| if (n & (4 * sizeof(float))) { |
| vst1_lane_u32((void*) y, vreinterpret_u32_s8(vy), 0); y += 4; |
| vy = vext_s8(vy, vy, 4); |
| } |
| if (n & (2 * sizeof(float))) { |
| vst1_lane_u16((void*) y, vreinterpret_u16_s8(vy), 0); y += 2; |
| vy = vext_s8(vy, vy, 2); |
| } |
| if (n & (1 * sizeof(float))) { |
| vst1_lane_s8(y, vy, 0); |
| } |
| } |
| } |