| Gian Marco | 58c5794 | 2017-11-28 09:10:03 +0000 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2017 ARM Limited. |
| 3 | * |
| 4 | * SPDX-License-Identifier: MIT |
| 5 | * |
| 6 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
| 7 | * of this software and associated documentation files (the "Software"), to |
| 8 | * deal in the Software without restriction, including without limitation the |
| 9 | * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or |
| 10 | * sell copies of the Software, and to permit persons to whom the Software is |
| 11 | * furnished to do so, subject to the following conditions: |
| 12 | * |
| 13 | * The above copyright notice and this permission notice shall be included in all |
| 14 | * copies or substantial portions of the Software. |
| 15 | * |
| 16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| 19 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| 20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| 21 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE |
| 22 | * SOFTWARE. |
| 23 | */ |
| 24 | namespace arm_compute |
| 25 | { |
| 26 | inline int32x4_t rounding_divide_by_pow2(int32x4_t x, int exponent) |
| 27 | { |
| 28 | const int32x4_t shift_vec = vdupq_n_s32(-exponent); |
| 29 | const int32x4_t fixup = vshrq_n_s32(vandq_s32(x, shift_vec), 31); |
| 30 | const int32x4_t fixed_up_x = vqaddq_s32(x, fixup); |
| 31 | return vrshlq_s32(fixed_up_x, shift_vec); |
| 32 | } |
| Michel Iwaniec | 5dfeae6 | 2017-11-29 10:48:23 +0000 | [diff] [blame] | 33 | |
| 34 | inline qasymm8x16_t vmlaq_qasymm8(qasymm8x16_t vd, float32x4_t vs, float32x4_t vo) |
| 35 | { |
| 36 | // Convert uint8 vectors to uint16 vectors |
| 37 | const uint8x8_t vd_low = vget_low_u8(vd); |
| 38 | const uint8x8_t vd_high = vget_high_u8(vd); |
| 39 | uint16x8_t vd_low_u16x8 = vmovl_u8(vd_low); |
| 40 | uint16x8_t vd_high_u16x8 = vmovl_u8(vd_high); |
| 41 | // Convert uint16 vectors to uint32 vectors |
| 42 | uint32x4_t A_u32x4 = vmovl_u16(vget_low_u16(vd_low_u16x8)); |
| 43 | uint32x4_t B_u32x4 = vmovl_u16(vget_high_u16(vd_low_u16x8)); |
| 44 | uint32x4_t C_u32x4 = vmovl_u16(vget_low_u16(vd_high_u16x8)); |
| 45 | uint32x4_t D_u32x4 = vmovl_u16(vget_high_u16(vd_high_u16x8)); |
| 46 | // Convert uint32 vectors to float32 vectors |
| 47 | float32x4_t A_f32x4 = vcvtq_f32_u32(A_u32x4); |
| 48 | float32x4_t B_f32x4 = vcvtq_f32_u32(B_u32x4); |
| 49 | float32x4_t C_f32x4 = vcvtq_f32_u32(C_u32x4); |
| 50 | float32x4_t D_f32x4 = vcvtq_f32_u32(D_u32x4); |
| 51 | // vd = vd*vs + vo |
| 52 | A_f32x4 = vmlaq_f32(vo, A_f32x4, vs); |
| 53 | B_f32x4 = vmlaq_f32(vo, B_f32x4, vs); |
| 54 | C_f32x4 = vmlaq_f32(vo, C_f32x4, vs); |
| 55 | D_f32x4 = vmlaq_f32(vo, D_f32x4, vs); |
| 56 | // Convert float32 vectors to uint32 vectors |
| 57 | A_u32x4 = vcvtq_u32_f32(A_f32x4); |
| 58 | B_u32x4 = vcvtq_u32_f32(B_f32x4); |
| 59 | C_u32x4 = vcvtq_u32_f32(C_f32x4); |
| 60 | D_u32x4 = vcvtq_u32_f32(D_f32x4); |
| 61 | // Convert uint32 vectors to uint16 vectors (with saturation) |
| 62 | vd_low_u16x8 = vcombine_u16(vqmovn_u32(A_u32x4), vqmovn_u32(B_u32x4)); |
| 63 | vd_high_u16x8 = vcombine_u16(vqmovn_u32(C_u32x4), vqmovn_u32(D_u32x4)); |
| 64 | // convert uint16 vectors to uint8 vectors (with saturation) |
| 65 | return vcombine_u8(vqmovn_u16(vd_low_u16x8), vqmovn_u16(vd_high_u16x8)); |
| 66 | } |
| 67 | } // namespace arm_compute |