From ddb93bbf12fc9d685e7ddbef703a886d67cbda9b Mon Sep 17 00:00:00 2001 From: Georgios Pinitas Date: Fri, 2 Oct 2020 16:38:59 +0100 Subject: COMPMID-3637: Move wrapper to src Signed-off-by: Georgios Pinitas Change-Id: I524b0c4b49c7a7035b7d078b9585d77b0d438e10 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4083 Reviewed-by: Michele Di Giorgio Reviewed-by: Michalis Spyrou Comments-Addressed: Arm Jenkins --- arm_compute/core/NEON/NEAsymm.inl | 92 --------------------------------------- 1 file changed, 92 deletions(-) delete mode 100644 arm_compute/core/NEON/NEAsymm.inl (limited to 'arm_compute/core/NEON/NEAsymm.inl') diff --git a/arm_compute/core/NEON/NEAsymm.inl b/arm_compute/core/NEON/NEAsymm.inl deleted file mode 100644 index d211382f7a..0000000000 --- a/arm_compute/core/NEON/NEAsymm.inl +++ /dev/null @@ -1,92 +0,0 @@ -/* - * Copyright (c) 2017-2019 Arm Limited. - * - * SPDX-License-Identifier: MIT - * - * Permission is hereby granted, free of charge, to any person obtaining a copy - * of this software and associated documentation files (the "Software"), to - * deal in the Software without restriction, including without limitation the - * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or - * sell copies of the Software, and to permit persons to whom the Software is - * furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included in all - * copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE - * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER - * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, - * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE - * SOFTWARE. - */ -namespace arm_compute -{ -inline qasymm8x16_t vmlaq_qasymm8(qasymm8x16_t vd, float32x4_t vs, float32x4_t vo) -{ - // Convert uint8 vectors to uint16 vectors - const uint8x8_t vd_low = vget_low_u8(vd); - const uint8x8_t vd_high = vget_high_u8(vd); - uint16x8_t vd_low_u16x8 = vmovl_u8(vd_low); - uint16x8_t vd_high_u16x8 = vmovl_u8(vd_high); - // Convert uint16 vectors to uint32 vectors - uint32x4_t A_u32x4 = vmovl_u16(vget_low_u16(vd_low_u16x8)); - uint32x4_t B_u32x4 = vmovl_u16(vget_high_u16(vd_low_u16x8)); - uint32x4_t C_u32x4 = vmovl_u16(vget_low_u16(vd_high_u16x8)); - uint32x4_t D_u32x4 = vmovl_u16(vget_high_u16(vd_high_u16x8)); - // Convert uint32 vectors to float32 vectors - float32x4_t A_f32x4 = vcvtq_f32_u32(A_u32x4); - float32x4_t B_f32x4 = vcvtq_f32_u32(B_u32x4); - float32x4_t C_f32x4 = vcvtq_f32_u32(C_u32x4); - float32x4_t D_f32x4 = vcvtq_f32_u32(D_u32x4); - // vd = vd*vs + vo - A_f32x4 = vmlaq_f32(vo, A_f32x4, vs); - B_f32x4 = vmlaq_f32(vo, B_f32x4, vs); - C_f32x4 = vmlaq_f32(vo, C_f32x4, vs); - D_f32x4 = vmlaq_f32(vo, D_f32x4, vs); - // Convert float32 vectors to uint32 vectors - A_u32x4 = vcvtq_u32_f32(A_f32x4); - B_u32x4 = vcvtq_u32_f32(B_f32x4); - C_u32x4 = vcvtq_u32_f32(C_f32x4); - D_u32x4 = vcvtq_u32_f32(D_f32x4); - // Convert uint32 vectors to uint16 vectors (with saturation) - vd_low_u16x8 = vcombine_u16(vqmovn_u32(A_u32x4), vqmovn_u32(B_u32x4)); - vd_high_u16x8 = vcombine_u16(vqmovn_u32(C_u32x4), vqmovn_u32(D_u32x4)); - // convert uint16 vectors to uint8 vectors (with saturation) - return vcombine_u8(vqmovn_u16(vd_low_u16x8), vqmovn_u16(vd_high_u16x8)); -} -inline qasymm8x16_signed_t vmlaq_qasymm8_signed(qasymm8x16_signed_t vd, float32x4_t vs, float32x4_t vo) -{ - // Convert uint8 vectors to int16 vectors - const int8x8_t vd_low = vget_low_s8(vd); - const int8x8_t vd_high = vget_high_s8(vd); - int16x8_t vd_low_s16x8 = vmovl_s8(vd_low); - int16x8_t vd_high_s16x8 = vmovl_s8(vd_high); - // Convert int16 vectors to int32 vectors - int32x4_t A_s32x4 = vmovl_s16(vget_low_s16(vd_low_s16x8)); - int32x4_t B_s32x4 = vmovl_s16(vget_high_s16(vd_low_s16x8)); - int32x4_t C_s32x4 = vmovl_s16(vget_low_s16(vd_high_s16x8)); - int32x4_t D_s32x4 = vmovl_s16(vget_high_s16(vd_high_s16x8)); - // Convert int32 vectors to float32 vectors - float32x4_t A_f32x4 = vcvtq_f32_s32(A_s32x4); - float32x4_t B_f32x4 = vcvtq_f32_s32(B_s32x4); - float32x4_t C_f32x4 = vcvtq_f32_s32(C_s32x4); - float32x4_t D_f32x4 = vcvtq_f32_s32(D_s32x4); - // vd = vd*vs + vo - A_f32x4 = vmlaq_f32(vo, A_f32x4, vs); - B_f32x4 = vmlaq_f32(vo, B_f32x4, vs); - C_f32x4 = vmlaq_f32(vo, C_f32x4, vs); - D_f32x4 = vmlaq_f32(vo, D_f32x4, vs); - // Convert float32 vectors to int32 vectors - A_s32x4 = vcvtq_s32_f32(A_f32x4); - B_s32x4 = vcvtq_s32_f32(B_f32x4); - C_s32x4 = vcvtq_s32_f32(C_f32x4); - D_s32x4 = vcvtq_s32_f32(D_f32x4); - // Convert int32 vectors to int16 vectors (with saturation) - vd_low_s16x8 = vcombine_s16(vqmovn_s32(A_s32x4), vqmovn_s32(B_s32x4)); - vd_high_s16x8 = vcombine_s16(vqmovn_s32(C_s32x4), vqmovn_s32(D_s32x4)); - // convert int16 vectors to int8 vectors (with saturation) - return vcombine_s8(vqmovn_s16(vd_low_s16x8), vqmovn_s16(vd_high_s16x8)); -} -} // namespace arm_compute -- cgit v1.2.1