// Auto-generated file. Do not edit! // Template: src/qs8-gavgpool/unipass-sse.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 void xnn_qs8_gavgpool_minmax_ukernel_7x__sse41_c16_acc2( size_t rows, size_t channels, const int8_t* input, size_t input_stride, const int8_t* zero, int8_t* output, const union xnn_qs8_avgpool_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_DISABLE_TSAN { assert(rows != 0); assert(rows <= 7); assert(channels != 0); const int8_t* i0 = input; const int8_t* i1 = (const int8_t*) ((uintptr_t) i0 + input_stride); if XNN_UNPREDICTABLE(rows < 2) { i1 = zero; } const int8_t* i2 = (const int8_t*) ((uintptr_t) i1 + input_stride); if XNN_UNPREDICTABLE(rows <= 2) { i2 = zero; } const int8_t* i3 = (const int8_t*) ((uintptr_t) i2 + input_stride); if XNN_UNPREDICTABLE(rows < 4) { i3 = zero; } const int8_t* i4 = (const int8_t*) ((uintptr_t) i3 + input_stride); if XNN_UNPREDICTABLE(rows <= 4) { i4 = zero; } const int8_t* i5 = (const int8_t*) ((uintptr_t) i4 + input_stride); if XNN_UNPREDICTABLE(rows < 6) { i5 = zero; } const int8_t* i6 = (const int8_t*) ((uintptr_t) i5 + input_stride); if XNN_UNPREDICTABLE(rows <= 6) { i6 = zero; } const __m128i vbias = _mm_load_si128((const __m128i*) params->sse2.bias); const __m128i vmultiplier = _mm_load_si128((const __m128i*) params->sse2.multiplier); const __m128i vrounding = _mm_load_si128((const __m128i*) params->sse2.rounding); const __m128i vshift = _mm_loadl_epi64((const __m128i*) params->sse2.shift); while (channels >= 16) { const __m128i vxi0x01234567 = _mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) i0)); const __m128i vxi0x89ABCDEF = _mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) (i0 + 8))); i0 += 16; const __m128i vxi1x01234567 = _mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) i1)); const __m128i vxi1x89ABCDEF = _mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) (i1 + 8))); i1 += 16; const __m128i vxi2x01234567 = _mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) i2)); const __m128i vxi2x89ABCDEF = _mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) (i2 + 8))); i2 += 16; const __m128i vxi3x01234567 = _mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) i3)); const __m128i vxi3x89ABCDEF = _mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) (i3 + 8))); i3 += 16; const __m128i vxi4x01234567 = _mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) i4)); const __m128i vxi4x89ABCDEF = _mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) (i4 + 8))); i4 += 16; const __m128i vxi5x01234567 = _mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) i5)); const __m128i vxi5x89ABCDEF = _mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) (i5 + 8))); i5 += 16; const __m128i vxi6x01234567 = _mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) i6)); const __m128i vxi6x89ABCDEF = _mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) (i6 + 8))); i6 += 16; __m128i vacc0x01234567 = _mm_add_epi16(vxi0x01234567, vxi1x01234567); __m128i vacc0x89ABCDEF = _mm_add_epi16(vxi0x89ABCDEF, vxi1x89ABCDEF); __m128i vacc1x01234567 = _mm_add_epi16(vxi2x01234567, vxi3x01234567); __m128i vacc1x89ABCDEF = _mm_add_epi16(vxi2x89ABCDEF, vxi3x89ABCDEF); vacc0x01234567 = _mm_add_epi16(vacc0x01234567, vxi4x01234567); vacc0x89ABCDEF = _mm_add_epi16(vacc0x89ABCDEF, vxi4x89ABCDEF); vacc1x01234567 = _mm_add_epi16(vacc1x01234567, vxi5x01234567); vacc1x89ABCDEF = _mm_add_epi16(vacc1x89ABCDEF, vxi5x89ABCDEF); vacc0x01234567 = _mm_add_epi16(vacc0x01234567, vxi6x01234567); vacc0x89ABCDEF = _mm_add_epi16(vacc0x89ABCDEF, vxi6x89ABCDEF); // Add up all accumulators to vacc0x0123456789ABCDEF vacc0x01234567 = _mm_add_epi16(vacc0x01234567, vacc1x01234567); vacc0x89ABCDEF = _mm_add_epi16(vacc0x89ABCDEF, vacc1x89ABCDEF); const __m128i vacc0123 = _mm_add_epi32(vbias, _mm_cvtepi16_epi32(vacc0x01234567)); const __m128i vacc4567 = _mm_add_epi32(vbias, _mm_unpackhi_epi16(vacc0x01234567, _mm_cmpgt_epi16(_mm_setzero_si128(), vacc0x01234567))); const __m128i vacc89AB = _mm_add_epi32(vbias, _mm_cvtepi16_epi32(vacc0x89ABCDEF)); const __m128i vaccCDEF = _mm_add_epi32(vbias, _mm_unpackhi_epi16(vacc0x89ABCDEF, _mm_cmpgt_epi16(_mm_setzero_si128(), vacc0x89ABCDEF))); const __m128i vabsacc0123 = _mm_abs_epi32(vacc0123); const __m128i vabsacc4567 = _mm_abs_epi32(vacc4567); const __m128i vabsacc89AB = _mm_abs_epi32(vacc89AB); const __m128i vabsaccCDEF = _mm_abs_epi32(vaccCDEF); const __m128i vabsacc13 = _mm_shuffle_epi32(vabsacc0123, _MM_SHUFFLE(3, 3, 1, 1)); const __m128i vabsacc57 = _mm_shuffle_epi32(vabsacc4567, _MM_SHUFFLE(3, 3, 1, 1)); const __m128i vabsacc9B = _mm_shuffle_epi32(vabsacc89AB, _MM_SHUFFLE(3, 3, 1, 1)); const __m128i vabsaccDF = _mm_shuffle_epi32(vabsaccCDEF, _MM_SHUFFLE(3, 3, 1, 1)); const __m128i vabsprod02 = _mm_mul_epu32(vabsacc0123, vmultiplier); const __m128i vabsprod13 = _mm_mul_epu32(vabsacc13, vmultiplier); const __m128i vabsprod46 = _mm_mul_epu32(vabsacc4567, vmultiplier); const __m128i vabsprod57 = _mm_mul_epu32(vabsacc57, vmultiplier); const __m128i vabsprod8A = _mm_mul_epu32(vabsacc89AB, vmultiplier); const __m128i vabsprod9B = _mm_mul_epu32(vabsacc9B, vmultiplier); const __m128i vabsprodCE = _mm_mul_epu32(vabsaccCDEF, vmultiplier); const __m128i vabsprodDF = _mm_mul_epu32(vabsaccDF, vmultiplier); const __m128i vabsout02 = _mm_srl_epi64(_mm_add_epi64(vabsprod02, vrounding), vshift); const __m128i vabsout13 = _mm_srl_epi64(_mm_add_epi64(vabsprod13, vrounding), vshift); const __m128i vabsout46 = _mm_srl_epi64(_mm_add_epi64(vabsprod46, vrounding), vshift); const __m128i vabsout57 = _mm_srl_epi64(_mm_add_epi64(vabsprod57, vrounding), vshift); const __m128i vabsout8A = _mm_srl_epi64(_mm_add_epi64(vabsprod8A, vrounding), vshift); const __m128i vabsout9B = _mm_srl_epi64(_mm_add_epi64(vabsprod9B, vrounding), vshift); const __m128i vabsoutCE = _mm_srl_epi64(_mm_add_epi64(vabsprodCE, vrounding), vshift); const __m128i vabsoutDF = _mm_srl_epi64(_mm_add_epi64(vabsprodDF, vrounding), vshift); const __m128i vabsout0123 = _mm_blend_epi16(vabsout02, _mm_shuffle_epi32(vabsout13, _MM_SHUFFLE(2, 2, 0, 0)), 0xCC); const __m128i vabsout4567 = _mm_blend_epi16(vabsout46, _mm_shuffle_epi32(vabsout57, _MM_SHUFFLE(2, 2, 0, 0)), 0xCC); const __m128i vabsout89AB = _mm_blend_epi16(vabsout8A, _mm_shuffle_epi32(vabsout9B, _MM_SHUFFLE(2, 2, 0, 0)), 0xCC); const __m128i vabsoutCDEF = _mm_blend_epi16(vabsoutCE, _mm_shuffle_epi32(vabsoutDF, _MM_SHUFFLE(2, 2, 0, 0)), 0xCC); const __m128i vout0123 = _mm_sign_epi32(vabsout0123, vacc0123); const __m128i vout4567 = _mm_sign_epi32(vabsout4567, vacc4567); const __m128i vout89AB = _mm_sign_epi32(vabsout89AB, vacc89AB); const __m128i voutCDEF = _mm_sign_epi32(vabsoutCDEF, vaccCDEF); const __m128i voutput_zero_point = _mm_load_si128((const __m128i*) params->sse2.output_zero_point); __m128i vout01234567 = _mm_adds_epi16(_mm_packs_epi32(vout0123, vout4567), voutput_zero_point); __m128i vout89ABCDEF = _mm_adds_epi16(_mm_packs_epi32(vout89AB, voutCDEF), voutput_zero_point); const __m128i voutput_min = _mm_load_si128((const __m128i*) params->sse2.output_min); const __m128i voutput_max = _mm_load_si128((const __m128i*) params->sse2.output_max); vout01234567 = _mm_min_epi16(_mm_max_epi16(vout01234567, voutput_min), voutput_max); vout89ABCDEF = _mm_min_epi16(_mm_max_epi16(vout89ABCDEF, voutput_min), voutput_max); __m128i vout0123456789ABCDEF = _mm_packs_epi16(vout01234567, vout89ABCDEF); _mm_storeu_si128((__m128i*) output, vout0123456789ABCDEF); output += 16; channels -= 16; } if XNN_UNLIKELY(channels != 0) { do { const __m128i vxi0x01234567 = _mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) i0)); i0 += 8; const __m128i vxi1x01234567 = _mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) i1)); i1 += 8; const __m128i vxi2x01234567 = _mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) i2)); i2 += 8; const __m128i vxi3x01234567 = _mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) i3)); i3 += 8; const __m128i vxi4x01234567 = _mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) i4)); i4 += 8; const __m128i vxi5x01234567 = _mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) i5)); i5 += 8; const __m128i vxi6x01234567 = _mm_cvtepi8_epi16(_mm_loadl_epi64((const __m128i*) i6)); i6 += 8; __m128i vacc0x01234567 = _mm_add_epi16(vxi0x01234567, vxi1x01234567); __m128i vacc1x01234567 = _mm_add_epi16(vxi2x01234567, vxi3x01234567); vacc0x01234567 = _mm_add_epi16(vacc0x01234567, vxi4x01234567); vacc1x01234567 = _mm_add_epi16(vacc1x01234567, vxi5x01234567); vacc0x01234567 = _mm_add_epi16(vacc0x01234567, vxi6x01234567); // Add up all accumulators to vacc0x01234567 vacc0x01234567 = _mm_add_epi16(vacc0x01234567, vacc1x01234567); const __m128i vacc0123 = _mm_add_epi32(vbias, _mm_cvtepi16_epi32(vacc0x01234567)); const __m128i vacc4567 = _mm_add_epi32(vbias, _mm_unpackhi_epi16(vacc0x01234567, _mm_cmpgt_epi16(_mm_setzero_si128(), vacc0x01234567))); const __m128i vabsacc0123 = _mm_abs_epi32(vacc0123); const __m128i vabsacc4567 = _mm_abs_epi32(vacc4567); const __m128i vabsacc13 = _mm_shuffle_epi32(vabsacc0123, _MM_SHUFFLE(3, 3, 1, 1)); const __m128i vabsacc57 = _mm_shuffle_epi32(vabsacc4567, _MM_SHUFFLE(3, 3, 1, 1)); const __m128i vabsprod02 = _mm_mul_epu32(vabsacc0123, vmultiplier); const __m128i vabsprod13 = _mm_mul_epu32(vabsacc13, vmultiplier); const __m128i vabsprod46 = _mm_mul_epu32(vabsacc4567, vmultiplier); const __m128i vabsprod57 = _mm_mul_epu32(vabsacc57, vmultiplier); const __m128i vabsout02 = _mm_srl_epi64(_mm_add_epi64(vabsprod02, vrounding), vshift); const __m128i vabsout13 = _mm_srl_epi64(_mm_add_epi64(vabsprod13, vrounding), vshift); const __m128i vabsout46 = _mm_srl_epi64(_mm_add_epi64(vabsprod46, vrounding), vshift); const __m128i vabsout57 = _mm_srl_epi64(_mm_add_epi64(vabsprod57, vrounding), vshift); const __m128i vabsout0123 = _mm_blend_epi16(vabsout02, _mm_shuffle_epi32(vabsout13, _MM_SHUFFLE(2, 2, 0, 0)), 0xCC); const __m128i vabsout4567 = _mm_blend_epi16(vabsout46, _mm_shuffle_epi32(vabsout57, _MM_SHUFFLE(2, 2, 0, 0)), 0xCC); const __m128i vout0123 = _mm_sign_epi32(vabsout0123, vacc0123); const __m128i vout4567 = _mm_sign_epi32(vabsout4567, vacc4567); const __m128i voutput_zero_point = _mm_load_si128((const __m128i*) params->sse2.output_zero_point); __m128i vout01234567 = _mm_adds_epi16(_mm_packs_epi32(vout0123, vout4567), voutput_zero_point); const __m128i voutput_min = _mm_load_si128((const __m128i*) params->sse2.output_min); const __m128i voutput_max = _mm_load_si128((const __m128i*) params->sse2.output_max); vout01234567 = _mm_min_epi16(_mm_max_epi16(vout01234567, voutput_min), voutput_max); __m128i vout0123456701234567 = _mm_packs_epi16(vout01234567, vout01234567); if XNN_LIKELY(channels >= 8) { _mm_storel_epi64((__m128i*) output, vout0123456701234567); output += 8; channels -= 8; } else { if (channels & 4) { *((uint32_t*) output) = (uint32_t) _mm_cvtsi128_si32(vout0123456701234567); vout0123456701234567 = _mm_srli_epi64(vout0123456701234567, 32); output += 4; } if (channels & 2) { *((uint16_t*) output) = (uint16_t) _mm_extract_epi16(vout0123456701234567, 0); vout0123456701234567 = _mm_srli_epi32(vout0123456701234567, 16); output += 2; } if (channels & 1) { *output = (int8_t) _mm_extract_epi8(vout0123456701234567, 0); output += 1; } channels = 0; } } while (channels != 0); } }