// Auto-generated file. Do not edit! // Template: src/f16-dwconv2d-chw/3x3s2p1-neonfp16arith.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_f16_dwconv2d_chw_ukernel_3x3s2p1__neonfp16arith_2x4( size_t input_height, size_t input_width, const void* input, const void* weights, const void* zero, void* output, uint32_t padding_top, const union xnn_f16_chw_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS { assert(input_height != 0); assert(input_width != 0); assert(input_width % sizeof(__fp16) == 0); assert(padding_top >= 0); assert(padding_top <= 1); const uint16x4_t vmask_even = vld1_u16(params->neonfp16arith.mask_even); const uint16x4_t vmask_odd = vld1_u16(params->neonfp16arith.mask_odd); const float16x4_t vmax = vld1_dup_f16(¶ms->neonfp16arith.max); const float16x4_t vmin = vld1_dup_f16(¶ms->neonfp16arith.min); const __fp16* w0 = (const __fp16*)weights; const float16x8_t vw01234567 = vld1q_f16(w0); const float16x4_t vw89 = vreinterpret_f16_u32(vld1_lane_u32((const void*)(w0 + 8), vmov_n_u32(0), 0)); const size_t input_decrement = round_down_po2(input_width, 4 /* SIMD output width */ * 2 /* subsampling */ * sizeof(__fp16)); const size_t output_width = round_down_po2((input_width + (2 /* padding */ - 3 /* kernel size */ + 2 /* subsampling */) * sizeof(__fp16)) / 2, sizeof(__fp16)); const __fp16* i0 = (const __fp16*) ((uintptr_t) input - ((-padding_top) & input_width)); const __fp16* i1 = (const __fp16*) ((uintptr_t) i0 + input_width); if XNN_UNPREDICTABLE(padding_top != 0) { i0 = zero; } const __fp16* i2 = (const __fp16*) ((uintptr_t) i1 + input_width); const __fp16* i3 = (const __fp16*) ((uintptr_t) i2 + input_width); const __fp16* i4 = (const __fp16*) ((uintptr_t) i3 + input_width); __fp16* o0 = output; __fp16* o1 = (__fp16*) ((uintptr_t) o0 + output_width); size_t padded_input_height = input_height + padding_top + 1 /* padding bottom */; size_t output_height = (padded_input_height - 3 /* kernel size */ + 2 /* subsampling */) / 2; do { if XNN_UNPREDICTABLE(padded_input_height < 4) { i2 = zero; } if XNN_UNPREDICTABLE(padded_input_height < 5) { i3 = zero; o1 = o0; } if XNN_UNPREDICTABLE(padded_input_height < 6) { i4 = zero; } float16x4_t vi0x1357 = vmov_n_f16(0); float16x4_t vi1x1357 = vmov_n_f16(0); float16x4_t vi2x1357 = vmov_n_f16(0); float16x4_t vi3x1357 = vmov_n_f16(0); float16x4_t vi4x1357 = vmov_n_f16(0); size_t w = input_width; for (; w >= 8 * sizeof(__fp16); w -= 8 * sizeof(__fp16)) { float16x4_t vo0p0 = vdup_laneq_f16(vw01234567, 0); float16x4_t vo1p0 = vdup_laneq_f16(vw01234567, 0); const float16x4x2_t vi0x8ACE9BDF = vld2_f16(i0); i0 += 8; const float16x4x2_t vi1x8ACE9BDF = vld2_f16(i1); i1 += 8; const float16x4x2_t vi2x8ACE9BDF = vld2_f16(i2); i2 += 8; const float16x4x2_t vi3x8ACE9BDF = vld2_f16(i3); i3 += 8; const float16x4x2_t vi4x8ACE9BDF = vld2_f16(i4); i4 += 8; vo0p0 = vfma_laneq_f16(vo0p0, vi0x8ACE9BDF.val[0], vw01234567, 2); vo1p0 = vfma_laneq_f16(vo1p0, vi2x8ACE9BDF.val[0], vw01234567, 2); vo0p0 = vfma_laneq_f16(vo0p0, vi1x8ACE9BDF.val[0], vw01234567, 5); vo1p0 = vfma_laneq_f16(vo1p0, vi3x8ACE9BDF.val[0], vw01234567, 5); vo0p0 = vfma_lane_f16(vo0p0, vi2x8ACE9BDF.val[0], vw89, 0); vo1p0 = vfma_lane_f16(vo1p0, vi4x8ACE9BDF.val[0], vw89, 0); const float16x4_t vi0x7BDF = vext_f16(vi0x1357, vi0x8ACE9BDF.val[1], 3); vi0x1357 = vi0x8ACE9BDF.val[1]; const float16x4_t vi1x7BDF = vext_f16(vi1x1357, vi1x8ACE9BDF.val[1], 3); vi1x1357 = vi1x8ACE9BDF.val[1]; const float16x4_t vi2x7BDF = vext_f16(vi2x1357, vi2x8ACE9BDF.val[1], 3); vi2x1357 = vi2x8ACE9BDF.val[1]; const float16x4_t vi3x7BDF = vext_f16(vi3x1357, vi3x8ACE9BDF.val[1], 3); vi3x1357 = vi3x8ACE9BDF.val[1]; const float16x4_t vi4x7BDF = vext_f16(vi4x1357, vi4x8ACE9BDF.val[1], 3); vi4x1357 = vi4x8ACE9BDF.val[1]; vo0p0 = vfma_laneq_f16(vo0p0, vi0x7BDF, vw01234567, 1); vo1p0 = vfma_laneq_f16(vo1p0, vi2x7BDF, vw01234567, 1); vo0p0 = vfma_laneq_f16(vo0p0, vi1x7BDF, vw01234567, 4); vo1p0 = vfma_laneq_f16(vo1p0, vi3x7BDF, vw01234567, 4); vo0p0 = vfma_laneq_f16(vo0p0, vi2x7BDF, vw01234567, 7); vo1p0 = vfma_laneq_f16(vo1p0, vi4x7BDF, vw01234567, 7); vo0p0 = vfma_laneq_f16(vo0p0, vi0x8ACE9BDF.val[1], vw01234567, 3); vo1p0 = vfma_laneq_f16(vo1p0, vi2x8ACE9BDF.val[1], vw01234567, 3); vo0p0 = vfma_laneq_f16(vo0p0, vi1x8ACE9BDF.val[1], vw01234567, 6); vo1p0 = vfma_laneq_f16(vo1p0, vi3x8ACE9BDF.val[1], vw01234567, 6); vo0p0 = vfma_lane_f16(vo0p0, vi2x8ACE9BDF.val[1], vw89, 1); vo1p0 = vfma_lane_f16(vo1p0, vi4x8ACE9BDF.val[1], vw89, 1); float16x4_t vo0 = vmax_f16(vo0p0, vmin); float16x4_t vo1 = vmax_f16(vo1p0, vmin); vo0 = vmin_f16(vo0, vmax); vo1 = vmin_f16(vo1, vmax); vst1_f16(o1, vo1); o1 += 4; vst1_f16(o0, vo0); o0 += 4; } // Last block has 0-7 pixels to process. assert(w < 8 * sizeof(__fp16)); if XNN_LIKELY(w != 0) { float16x4_t vo0p0 = vdup_laneq_f16(vw01234567, 0); float16x4_t vo1p0 = vdup_laneq_f16(vw01234567, 0); const float16x4x2_t vi0x8ACE9BDF = vld2_f16(i0); const float16x4x2_t vi1x8ACE9BDF = vld2_f16(i1); const float16x4x2_t vi2x8ACE9BDF = vld2_f16(i2); const float16x4x2_t vi3x8ACE9BDF = vld2_f16(i3); const float16x4x2_t vi4x8ACE9BDF = vld2_f16(i4); const float16x4_t vi0x8ACE = vreinterpret_f16_u16(vand_u16(vmask_even, vreinterpret_u16_f16(vi0x8ACE9BDF.val[0]))); const float16x4_t vi0x9BDF = vreinterpret_f16_u16(vand_u16(vmask_odd, vreinterpret_u16_f16(vi0x8ACE9BDF.val[1]))); const float16x4_t vi1x8ACE = vreinterpret_f16_u16(vand_u16(vmask_even, vreinterpret_u16_f16(vi1x8ACE9BDF.val[0]))); const float16x4_t vi1x9BDF = vreinterpret_f16_u16(vand_u16(vmask_odd, vreinterpret_u16_f16(vi1x8ACE9BDF.val[1]))); const float16x4_t vi2x8ACE = vreinterpret_f16_u16(vand_u16(vmask_even, vreinterpret_u16_f16(vi2x8ACE9BDF.val[0]))); const float16x4_t vi2x9BDF = vreinterpret_f16_u16(vand_u16(vmask_odd, vreinterpret_u16_f16(vi2x8ACE9BDF.val[1]))); const float16x4_t vi3x8ACE = vreinterpret_f16_u16(vand_u16(vmask_even, vreinterpret_u16_f16(vi3x8ACE9BDF.val[0]))); const float16x4_t vi3x9BDF = vreinterpret_f16_u16(vand_u16(vmask_odd, vreinterpret_u16_f16(vi3x8ACE9BDF.val[1]))); const float16x4_t vi4x8ACE = vreinterpret_f16_u16(vand_u16(vmask_even, vreinterpret_u16_f16(vi4x8ACE9BDF.val[0]))); const float16x4_t vi4x9BDF = vreinterpret_f16_u16(vand_u16(vmask_odd, vreinterpret_u16_f16(vi4x8ACE9BDF.val[1]))); vo0p0 = vfma_laneq_f16(vo0p0, vi0x8ACE, vw01234567, 2); vo1p0 = vfma_laneq_f16(vo1p0, vi2x8ACE, vw01234567, 2); vo0p0 = vfma_laneq_f16(vo0p0, vi1x8ACE, vw01234567, 5); vo1p0 = vfma_laneq_f16(vo1p0, vi3x8ACE, vw01234567, 5); vo0p0 = vfma_lane_f16(vo0p0, vi2x8ACE, vw89, 0); vo1p0 = vfma_lane_f16(vo1p0, vi4x8ACE, vw89, 0); const float16x4_t vi0x7BDF = vext_f16(vi0x1357, vi0x9BDF, 3); const float16x4_t vi1x7BDF = vext_f16(vi1x1357, vi1x9BDF, 3); const float16x4_t vi2x7BDF = vext_f16(vi2x1357, vi2x9BDF, 3); const float16x4_t vi3x7BDF = vext_f16(vi3x1357, vi3x9BDF, 3); const float16x4_t vi4x7BDF = vext_f16(vi4x1357, vi4x9BDF, 3); vo0p0 = vfma_laneq_f16(vo0p0, vi0x7BDF, vw01234567, 1); vo1p0 = vfma_laneq_f16(vo1p0, vi2x7BDF, vw01234567, 1); vo0p0 = vfma_laneq_f16(vo0p0, vi1x7BDF, vw01234567, 4); vo1p0 = vfma_laneq_f16(vo1p0, vi3x7BDF, vw01234567, 4); vo0p0 = vfma_laneq_f16(vo0p0, vi2x7BDF, vw01234567, 7); vo1p0 = vfma_laneq_f16(vo1p0, vi4x7BDF, vw01234567, 7); vo0p0 = vfma_laneq_f16(vo0p0, vi0x9BDF, vw01234567, 3); vo1p0 = vfma_laneq_f16(vo1p0, vi2x9BDF, vw01234567, 3); vo0p0 = vfma_laneq_f16(vo0p0, vi1x9BDF, vw01234567, 6); vo1p0 = vfma_laneq_f16(vo1p0, vi3x9BDF, vw01234567, 6); vo0p0 = vfma_lane_f16(vo0p0, vi2x9BDF, vw89, 1); vo1p0 = vfma_lane_f16(vo1p0, vi4x9BDF, vw89, 1); float16x4_t vo0 = vmax_f16(vo0p0, vmin); float16x4_t vo1 = vmax_f16(vo1p0, vmin); vo0 = vmin_f16(vo0, vmax); vo1 = vmin_f16(vo1, vmax); w += 1 * sizeof(__fp16); if XNN_LIKELY(w == 8 * sizeof(__fp16)) { vst1_f16(o1, vo1); o1 += 4; vst1_f16(o0, vo0); o0 += 4; } else { if (w & (4 * sizeof(__fp16))) { vst1_lane_u32((void*) o1, vreinterpret_u32_f16(vo1), 0); o1 += 2; vst1_lane_u32((void*) o0, vreinterpret_u32_f16(vo0), 0); o0 += 2; vo0 = vext_f16(vo0, vo0, 2); vo1 = vext_f16(vo1, vo1, 2); } if (w & (2 * sizeof(__fp16))) { vst1_lane_f16(o1, vo1, 0); o1 += 1; vst1_lane_f16(o0, vo0, 0); o0 += 1; } } } i0 = (const __fp16*) ((uintptr_t) i4 - input_decrement); i1 = (const __fp16*) ((uintptr_t) i0 + input_width); i2 = (const __fp16*) ((uintptr_t) i1 + input_width); i3 = (const __fp16*) ((uintptr_t) i2 + input_width); i4 = (const __fp16*) ((uintptr_t) i3 + input_width); o0 = o1; o1 = (__fp16*) ((uintptr_t) o0 + output_width); output_height = doz(output_height, 2); padded_input_height = doz(padded_input_height, 4); } while (output_height != 0); }