src/s8-ibilinear/sse.c.in - platform/external/XNNPACK - Git at Google

 // 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.

 $assert SSE in [2, 4]
 $assert not XOP or AVX
 $assert not AVX or SSE == 4
 $assert DATATYPE in ["S8", "U8"]
 $assert CHANNEL_TILE % 8 == 0
 $assert CHANNEL_TILE >= 8
 $assert PIXEL_TILE == 1
 $ABC = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ"
 #include <assert.h>

 $if XOP:
   #if defined(__GNUC__) || defined(__clang__)
     #include <x86intrin.h>
   #else
     #include <immintrin.h>
     #include <ammintrin.h>
   #endif
 $else:
   $SSE_HEADER = {2: "emmintrin.h", 4: "smmintrin.h"}[SSE]
   #include <${SSE_HEADER}>

 #include <xnnpack/common.h>
 #include <xnnpack/ibilinear.h>
 #include <xnnpack/unaligned.h>


 $XINT8_T = {"S8": "int8_t", "U8": "uint8_t"}[DATATYPE]
 $_MM_CVTEPX8_EPI16 = {"S8": "_mm_cvtepi8_epi16", "U8": "_mm_cvtepu8_epi16"}[DATATYPE]
 $_MM_SRXI_EPI32 = {"S8": "_mm_srai_epi32", "U8": "_mm_srli_epi32"}[DATATYPE]
 $_MM_SRXI_EPI16 = {"S8": "_mm_srai_epi16", "U8": "_mm_srli_epi16"}[DATATYPE]
 $_MM_PACKXS_EPI16 = {"S8": "_mm_packs_epi16", "U8": "_mm_packus_epi16"}[DATATYPE]
 $ISA = "xop" if XOP else "avx" if AVX else {2: "sse2", 3: "ssse3", 4: "sse41"}[SSE]
 void xnn_${DATATYPE.lower()}_ibilinear_ukernel__${ISA}_c${CHANNEL_TILE}${"" if PIXEL_TILE == 1 else "x%d" % PIXEL_TILE}(
     size_t output_pixels,
     size_t channels,
     const ${XINT8_T}**restrict input,
     size_t input_offset,
     const int16_t*restrict weights,
     ${XINT8_T}*restrict output,
     size_t output_increment) XNN_OOB_READS
 {
   assert(output_pixels != 0);
   assert(channels != 0);

   do {
     const ${XINT8_T}* i0 = (const ${XINT8_T}*) ((uintptr_t) input[0] + input_offset);
     const ${XINT8_T}* i1 = (const ${XINT8_T}*) ((uintptr_t) input[1] + input_offset);
     const ${XINT8_T}* i2 = (const ${XINT8_T}*) ((uintptr_t) input[2] + input_offset);
     const ${XINT8_T}* i3 = (const ${XINT8_T}*) ((uintptr_t) input[3] + input_offset);
     input += 4;

     const __m128i valpha = _mm_cvtsi32_si128(*((const int*) weights));
     weights += 2;
     __m128i valphah = _mm_shufflelo_epi16(valpha, _MM_SHUFFLE(0, 0, 0, 0));
     valphah = _mm_unpacklo_epi64(valphah, valphah);
     $if SSE == 2:
       __m128i valphav = _mm_shufflelo_epi16(valpha, _MM_SHUFFLE(1, 1, 1, 1));
       valphav = _mm_unpacklo_epi64(valphav, valphav);
     $else:
       __m128i valphav = _mm_srli_epi32(valpha, 16);
       valphav = _mm_shuffle_epi32(valphav, _MM_SHUFFLE(0, 0, 0, 0));

     $if SSE == 4:
       valphah = _mm_blend_epi16(valphah, _mm_sub_epi16(_mm_set1_epi32(0x08000000), valphah), 0xAA);
     $else:
       valphah = _mm_xor_si128(valphah, _mm_set1_epi32(0xFFFF0000));
       valphah = _mm_add_epi16(valphah, _mm_set1_epi32(0x08010000));

     const __m128i vrounding = _mm_set1_epi32(0x00200000);

     size_t c = channels;
     $if CHANNEL_TILE > 8:
       for (; c >= ${CHANNEL_TILE} * sizeof(${XINT8_T}); c -= ${CHANNEL_TILE} * sizeof(${XINT8_T})) {
         $if SSE == 4:
           const __m128i vtl${ABC[0:8]} = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i0));
           const __m128i vtr${ABC[0:8]} = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i1));
           const __m128i vbl${ABC[0:8]} = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i2));
           const __m128i vbr${ABC[0:8]} = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i3));
           $for C in range(8, CHANNEL_TILE, 8):
             const __m128i vtl${ABC[C:C+8]} = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) (i0 + ${C})));
             const __m128i vtr${ABC[C:C+8]} = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) (i1 + ${C})));
             const __m128i vbl${ABC[C:C+8]} = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) (i2 + ${C})));
             const __m128i vbr${ABC[C:C+8]} = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) (i3 + ${C})));
         $else:
           __m128i vtl${ABC[0:8]} = _mm_loadl_epi64((const __m128i*) i0);
           __m128i vtr${ABC[0:8]} = _mm_loadl_epi64((const __m128i*) i1);
           __m128i vbl${ABC[0:8]} = _mm_loadl_epi64((const __m128i*) i2);
           __m128i vbr${ABC[0:8]} = _mm_loadl_epi64((const __m128i*) i3);
           $for C in range(8, CHANNEL_TILE, 8):
             __m128i vtl${ABC[C:C+8]} = _mm_loadl_epi64((const __m128i*) (i0 + ${C}));
             __m128i vtr${ABC[C:C+8]} = _mm_loadl_epi64((const __m128i*) (i1 + ${C}));
             __m128i vbl${ABC[C:C+8]} = _mm_loadl_epi64((const __m128i*) (i2 + ${C}));
             __m128i vbr${ABC[C:C+8]} = _mm_loadl_epi64((const __m128i*) (i3 + ${C}));
         i0 += ${CHANNEL_TILE};
         i1 += ${CHANNEL_TILE};
         i2 += ${CHANNEL_TILE};
         i3 += ${CHANNEL_TILE};

         $if SSE != 4:
           $if DATATYPE == "U8":
             __m128i vzero = _mm_setzero_si128();
             $for C in range(0, CHANNEL_TILE, 8):
               vtl${ABC[C:C+8]} = _mm_unpacklo_epi8(vtl${ABC[C:C+8]}, vzero);
               vtr${ABC[C:C+8]} = _mm_unpacklo_epi8(vtr${ABC[C:C+8]}, vzero);
               vbl${ABC[C:C+8]} = _mm_unpacklo_epi8(vbl${ABC[C:C+8]}, vzero);
               vbr${ABC[C:C+8]} = _mm_unpacklo_epi8(vbr${ABC[C:C+8]}, vzero);
           $else:
             $for C in range(0, CHANNEL_TILE, 8):
               vtl${ABC[C:C+8]} = _mm_srai_epi16(_mm_unpacklo_epi8(vtl${ABC[C:C+8]}, vtl${ABC[C:C+8]}), 8);
               vtr${ABC[C:C+8]} = _mm_srai_epi16(_mm_unpacklo_epi8(vtr${ABC[C:C+8]}, vtr${ABC[C:C+8]}), 8);
               vbl${ABC[C:C+8]} = _mm_srai_epi16(_mm_unpacklo_epi8(vbl${ABC[C:C+8]}, vbl${ABC[C:C+8]}), 8);
               vbr${ABC[C:C+8]} = _mm_srai_epi16(_mm_unpacklo_epi8(vbr${ABC[C:C+8]}, vbr${ABC[C:C+8]}), 8);

         $for C in range(0, CHANNEL_TILE, 8):
           const __m128i vdr${ABC[C:C+8]} = _mm_sub_epi16(vbr${ABC[C:C+8]}, vtr${ABC[C:C+8]});
           const __m128i vt${ABC[C:C+4]} = _mm_madd_epi16(_mm_unpacklo_epi16(vtr${ABC[C:C+8]}, vtl${ABC[C:C+8]}), valphah);
           const __m128i vdl${ABC[C:C+8]} = _mm_sub_epi16(vbl${ABC[C:C+8]}, vtl${ABC[C:C+8]});
           const __m128i vt${ABC[C+4:C+8]} = _mm_madd_epi16(_mm_unpackhi_epi16(vtr${ABC[C:C+8]}, vtl${ABC[C:C+8]}), valphah);

         $for C in range(0, CHANNEL_TILE, 8):
           const __m128i vd${ABC[C:C+4]} = _mm_madd_epi16(_mm_unpacklo_epi16(vdr${ABC[C:C+8]}, vdl${ABC[C:C+8]}), valphah);
           const __m128i vd${ABC[C+4:C+8]} = _mm_madd_epi16(_mm_unpackhi_epi16(vdr${ABC[C:C+8]}, vdl${ABC[C:C+8]}), valphah);

         $if SSE == 4:
           $for C in range(0, CHANNEL_TILE, 4):
             __m128i vacc${ABC[C:C+4]} = _mm_mullo_epi32(vd${ABC[C:C+4]}, valphav);
         $else:
           $for C in range(0, CHANNEL_TILE, 4):
             __m128i vacc${ABC[C:C+4]} = _mm_slli_epi32(_mm_mulhi_epu16(vd${ABC[C:C+4]}, valphav), 16);

           $for C in range(0, CHANNEL_TILE, 4):
             vacc${ABC[C:C+4]} = _mm_add_epi16(_mm_mullo_epi16(vd${ABC[C:C+4]}, valphav), vacc${ABC[C:C+4]});

         $for C in range(0, CHANNEL_TILE, 4):
           vacc${ABC[C:C+4]} = _mm_add_epi32(_mm_slli_epi32(vt${ABC[C:C+4]}, 11), vacc${ABC[C:C+4]});

         $for C in range(0, CHANNEL_TILE, 4):
           vacc${ABC[C:C+4]} = ${_MM_SRXI_EPI32}(_mm_add_epi16(vacc${ABC[C:C+4]}, vrounding), 22);

         $for C in range(0, CHANNEL_TILE, 8):
           const __m128i vacc${ABC[C:C+8]} = _mm_packs_epi32(vacc${ABC[C:C+4]}, vacc${ABC[C+4:C+8]});

         $for C in range(0, CHANNEL_TILE, 16):
           $if C + 8 < CHANNEL_TILE:
             const __m128i vo${ABC[C:C+16]} = ${_MM_PACKXS_EPI16}(vacc${ABC[C:C+8]}, vacc${ABC[C+8:C+16]});
           $else:
             const __m128i vo${ABC[C:C+8]} = ${_MM_PACKXS_EPI16}(vacc${ABC[C:C+8]}, vacc${ABC[C:C+8]});

         _mm_storeu_si128((__m128i*) output, vo${ABC[0:16]});
         $for C in range(16, CHANNEL_TILE, 16):
           $if C + 8 < CHANNEL_TILE:
             _mm_storeu_si128((__m128i*) (output + ${C}), vo${ABC[C:C+16]});
           $else:
             _mm_storel_epi64((__m128i*) (output + ${C}), vo${ABC[C:C+8]});
         output += ${CHANNEL_TILE};
       }
     for (; c >= 8 * sizeof(${XINT8_T}); c -= 8 * sizeof(${XINT8_T})) {
       $if SSE == 4:
         const __m128i vtl01234567 = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i0));
         i0 += 8;
         const __m128i vtr01234567 = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i1));
         i1 += 8;
         const __m128i vbl01234567 = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i2));
         i2 += 8;
         const __m128i vbr01234567 = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i3));
         i3 += 8;
       $else:
         __m128i vtl01234567 = _mm_loadl_epi64((const __m128i*) i0);
         i0 += 8;
         __m128i vtr01234567 = _mm_loadl_epi64((const __m128i*) i1);
         i1 += 8;
         __m128i vbl01234567 = _mm_loadl_epi64((const __m128i*) i2);
         i2 += 8;
         __m128i vbr01234567 = _mm_loadl_epi64((const __m128i*) i3);
         i3 += 8;

       $if SSE != 4:
         $if DATATYPE == "U8":
           __m128i vzero = _mm_setzero_si128();
           vtl01234567 = _mm_unpacklo_epi8(vtl01234567, vzero);
           vtr01234567 = _mm_unpacklo_epi8(vtr01234567, vzero);
           vbl01234567 = _mm_unpacklo_epi8(vbl01234567, vzero);
           vbr01234567 = _mm_unpacklo_epi8(vbr01234567, vzero);
         $else:
           vtl01234567 = _mm_srai_epi16(_mm_unpacklo_epi8(vtl01234567, vtl01234567), 8);
           vtr01234567 = _mm_srai_epi16(_mm_unpacklo_epi8(vtr01234567, vtr01234567), 8);
           vbl01234567 = _mm_srai_epi16(_mm_unpacklo_epi8(vbl01234567, vbl01234567), 8);
           vbr01234567 = _mm_srai_epi16(_mm_unpacklo_epi8(vbr01234567, vbr01234567), 8);

       const __m128i vdr01234567 = _mm_sub_epi16(vbr01234567, vtr01234567);
       const __m128i vt0123 = _mm_madd_epi16(_mm_unpacklo_epi16(vtr01234567, vtl01234567), valphah);
       const __m128i vdl01234567 = _mm_sub_epi16(vbl01234567, vtl01234567);
       const __m128i vt4567 = _mm_madd_epi16(_mm_unpackhi_epi16(vtr01234567, vtl01234567), valphah);

       const __m128i vd0123 = _mm_madd_epi16(_mm_unpacklo_epi16(vdr01234567, vdl01234567), valphah);
       const __m128i vd4567 = _mm_madd_epi16(_mm_unpackhi_epi16(vdr01234567, vdl01234567), valphah);

       $if SSE == 4:
         __m128i vacc0123 = _mm_mullo_epi32(vd0123, valphav);
         __m128i vacc4567 = _mm_mullo_epi32(vd4567, valphav);
       $else:
         __m128i vacc0123 = _mm_slli_epi32(_mm_mulhi_epu16(vd0123, valphav), 16);
         __m128i vacc4567 = _mm_slli_epi32(_mm_mulhi_epu16(vd4567, valphav), 16);

         vacc0123 = _mm_add_epi16(_mm_mullo_epi16(vd0123, valphav), vacc0123);
         vacc4567 = _mm_add_epi16(_mm_mullo_epi16(vd4567, valphav), vacc4567);

       vacc0123 = _mm_add_epi32(_mm_slli_epi32(vt0123, 11), vacc0123);
       vacc4567 = _mm_add_epi32(_mm_slli_epi32(vt4567, 11), vacc4567);

       vacc0123 = ${_MM_SRXI_EPI32}(_mm_add_epi16(vacc0123, vrounding), 22);
       vacc4567 = ${_MM_SRXI_EPI32}(_mm_add_epi16(vacc4567, vrounding), 22);

       const __m128i vacc01234567 = _mm_packs_epi32(vacc0123, vacc4567);

       const __m128i vo01234567 = ${_MM_PACKXS_EPI16}(vacc01234567, vacc01234567);

       _mm_storel_epi64((__m128i*) output, vo01234567);
       output += 8;
     }
     if XNN_UNLIKELY(c != 0) {
       $if SSE == 4:
         const __m128i vtl01234567 = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i0));
         const __m128i vtr01234567 = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i1));
         const __m128i vbl01234567 = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i2));
         const __m128i vbr01234567 = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i3));
       $else:
         __m128i vtl01234567 = _mm_loadl_epi64((const __m128i*) i0);
         __m128i vtr01234567 = _mm_loadl_epi64((const __m128i*) i1);
         __m128i vbl01234567 = _mm_loadl_epi64((const __m128i*) i2);
         __m128i vbr01234567 = _mm_loadl_epi64((const __m128i*) i3);

       $if SSE != 4:
         $if DATATYPE == "U8":
           __m128i vzero = _mm_setzero_si128();
           vtl01234567 = _mm_unpacklo_epi8(vtl01234567, vzero);
           vtr01234567 = _mm_unpacklo_epi8(vtr01234567, vzero);
           vbl01234567 = _mm_unpacklo_epi8(vbl01234567, vzero);
           vbr01234567 = _mm_unpacklo_epi8(vbr01234567, vzero);
         $else:
           vtl01234567 = _mm_srai_epi16(_mm_unpacklo_epi8(vtl01234567, vtl01234567), 8);
           vtr01234567 = _mm_srai_epi16(_mm_unpacklo_epi8(vtr01234567, vtr01234567), 8);
           vbl01234567 = _mm_srai_epi16(_mm_unpacklo_epi8(vbl01234567, vbl01234567), 8);
           vbr01234567 = _mm_srai_epi16(_mm_unpacklo_epi8(vbr01234567, vbr01234567), 8);

       const __m128i vdr01234567 = _mm_sub_epi16(vbr01234567, vtr01234567);
       const __m128i vt0123 = _mm_madd_epi16(_mm_unpacklo_epi16(vtr01234567, vtl01234567), valphah);
       const __m128i vdl01234567 = _mm_sub_epi16(vbl01234567, vtl01234567);
       const __m128i vt4567 = _mm_madd_epi16(_mm_unpackhi_epi16(vtr01234567, vtl01234567), valphah);

       const __m128i vd0123 = _mm_madd_epi16(_mm_unpacklo_epi16(vdr01234567, vdl01234567), valphah);
       const __m128i vd4567 = _mm_madd_epi16(_mm_unpackhi_epi16(vdr01234567, vdl01234567), valphah);

       $if SSE == 4:
         __m128i vacc0123 = _mm_mullo_epi32(vd0123, valphav);
         __m128i vacc4567 = _mm_mullo_epi32(vd4567, valphav);
       $else:
         __m128i vacc0123 = _mm_slli_epi32(_mm_mulhi_epu16(vd0123, valphav), 16);
         __m128i vacc4567 = _mm_slli_epi32(_mm_mulhi_epu16(vd4567, valphav), 16);

         vacc0123 = _mm_add_epi16(_mm_mullo_epi16(vd0123, valphav), vacc0123);
         vacc4567 = _mm_add_epi16(_mm_mullo_epi16(vd4567, valphav), vacc4567);

       vacc0123 = _mm_add_epi32(_mm_slli_epi32(vt0123, 11), vacc0123);
       vacc4567 = _mm_add_epi32(_mm_slli_epi32(vt4567, 11), vacc4567);

       vacc0123 = ${_MM_SRXI_EPI32}(_mm_add_epi16(vacc0123, vrounding), 22);
       vacc4567 = ${_MM_SRXI_EPI32}(_mm_add_epi16(vacc4567, vrounding), 22);

       const __m128i vacc01234567 = _mm_packs_epi32(vacc0123, vacc4567);

       __m128i vo01234567 = ${_MM_PACKXS_EPI16}(vacc01234567, vacc01234567);

       if (c & (4 * sizeof(${XINT8_T}))) {
         unaligned_store_u32(output, (uint32_t) _mm_cvtsi128_si32(vo01234567));
         output += 4;
         vo01234567 = _mm_srli_epi64(vo01234567, 32);
       }
       $if SSE == 4:
         if (c & (2 * sizeof(${XINT8_T}))) {
           unaligned_store_u16(output, (uint16_t) _mm_extract_epi16(vo01234567, 0));
           output += 2;
           vo01234567 = _mm_srli_epi32(vo01234567, 16);
         }
         if (c & (1 * sizeof(${XINT8_T}))) {
           *output++ = (uint8_t) _mm_extract_epi8(vo01234567, 0);
         }
       $else:
         uint32_t vo0123 = (uint32_t) _mm_cvtsi128_si32(vo01234567);
         if (c & (2 * sizeof(${XINT8_T}))) {
           unaligned_store_u16(output, (uint16_t) vo0123);
           output += 2;
           vo0123 >>= 16;
         }
         if (c & (1 * sizeof(${XINT8_T}))) {
           *output++ = (uint8_t) vo0123;
         }
     }

     output = (${XINT8_T}*) ((uintptr_t) output + output_increment);
   } while (--output_pixels != 0);
 }
	// 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.

	$assert SSE in [2, 4]
	$assert not XOP or AVX
	$assert not AVX or SSE == 4
	$assert DATATYPE in ["S8", "U8"]
	$assert CHANNEL_TILE % 8 == 0
	$assert CHANNEL_TILE >= 8
	$assert PIXEL_TILE == 1
	$ABC = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ"
	#include <assert.h>

	$if XOP:
	#if defined(__GNUC__) \|\| defined(__clang__)
	#include <x86intrin.h>
	#else
	#include <immintrin.h>
	#include <ammintrin.h>
	#endif
	$else:
	$SSE_HEADER = {2: "emmintrin.h", 4: "smmintrin.h"}[SSE]
	#include <${SSE_HEADER}>

	#include <xnnpack/common.h>
	#include <xnnpack/ibilinear.h>
	#include <xnnpack/unaligned.h>


	$XINT8_T = {"S8": "int8_t", "U8": "uint8_t"}[DATATYPE]
	$_MM_CVTEPX8_EPI16 = {"S8": "_mm_cvtepi8_epi16", "U8": "_mm_cvtepu8_epi16"}[DATATYPE]
	$_MM_SRXI_EPI32 = {"S8": "_mm_srai_epi32", "U8": "_mm_srli_epi32"}[DATATYPE]
	$_MM_SRXI_EPI16 = {"S8": "_mm_srai_epi16", "U8": "_mm_srli_epi16"}[DATATYPE]
	$_MM_PACKXS_EPI16 = {"S8": "_mm_packs_epi16", "U8": "_mm_packus_epi16"}[DATATYPE]
	$ISA = "xop" if XOP else "avx" if AVX else {2: "sse2", 3: "ssse3", 4: "sse41"}[SSE]
	void xnn_${DATATYPE.lower()}_ibilinear_ukernel__${ISA}_c${CHANNEL_TILE}${"" if PIXEL_TILE == 1 else "x%d" % PIXEL_TILE}(
	size_t output_pixels,
	size_t channels,
	const ${XINT8_T}**restrict input,
	size_t input_offset,
	const int16_t*restrict weights,
	${XINT8_T}*restrict output,
	size_t output_increment) XNN_OOB_READS
	{
	assert(output_pixels != 0);
	assert(channels != 0);

	do {
	const ${XINT8_T}* i0 = (const ${XINT8_T}*) ((uintptr_t) input[0] + input_offset);
	const ${XINT8_T}* i1 = (const ${XINT8_T}*) ((uintptr_t) input[1] + input_offset);
	const ${XINT8_T}* i2 = (const ${XINT8_T}*) ((uintptr_t) input[2] + input_offset);
	const ${XINT8_T}* i3 = (const ${XINT8_T}*) ((uintptr_t) input[3] + input_offset);
	input += 4;

	const __m128i valpha = _mm_cvtsi32_si128(((const int) weights));
	weights += 2;
	__m128i valphah = _mm_shufflelo_epi16(valpha, _MM_SHUFFLE(0, 0, 0, 0));
	valphah = _mm_unpacklo_epi64(valphah, valphah);
	$if SSE == 2:
	__m128i valphav = _mm_shufflelo_epi16(valpha, _MM_SHUFFLE(1, 1, 1, 1));
	valphav = _mm_unpacklo_epi64(valphav, valphav);
	$else:
	__m128i valphav = _mm_srli_epi32(valpha, 16);
	valphav = _mm_shuffle_epi32(valphav, _MM_SHUFFLE(0, 0, 0, 0));

	$if SSE == 4:
	valphah = _mm_blend_epi16(valphah, _mm_sub_epi16(_mm_set1_epi32(0x08000000), valphah), 0xAA);
	$else:
	valphah = _mm_xor_si128(valphah, _mm_set1_epi32(0xFFFF0000));
	valphah = _mm_add_epi16(valphah, _mm_set1_epi32(0x08010000));

	const __m128i vrounding = _mm_set1_epi32(0x00200000);

	size_t c = channels;
	$if CHANNEL_TILE > 8:
	for (; c >= ${CHANNEL_TILE} * sizeof(${XINT8_T}); c -= ${CHANNEL_TILE} * sizeof(${XINT8_T})) {
	$if SSE == 4:
	const __m128i vtl${ABC[0:8]} = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i0));
	const __m128i vtr${ABC[0:8]} = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i1));
	const __m128i vbl${ABC[0:8]} = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i2));
	const __m128i vbr${ABC[0:8]} = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i3));
	$for C in range(8, CHANNEL_TILE, 8):
	const __m128i vtl${ABC[C:C+8]} = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) (i0 + ${C})));
	const __m128i vtr${ABC[C:C+8]} = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) (i1 + ${C})));
	const __m128i vbl${ABC[C:C+8]} = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) (i2 + ${C})));
	const __m128i vbr${ABC[C:C+8]} = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) (i3 + ${C})));
	$else:
	__m128i vtl${ABC[0:8]} = _mm_loadl_epi64((const __m128i*) i0);
	__m128i vtr${ABC[0:8]} = _mm_loadl_epi64((const __m128i*) i1);
	__m128i vbl${ABC[0:8]} = _mm_loadl_epi64((const __m128i*) i2);
	__m128i vbr${ABC[0:8]} = _mm_loadl_epi64((const __m128i*) i3);
	$for C in range(8, CHANNEL_TILE, 8):
	__m128i vtl${ABC[C:C+8]} = _mm_loadl_epi64((const __m128i*) (i0 + ${C}));
	__m128i vtr${ABC[C:C+8]} = _mm_loadl_epi64((const __m128i*) (i1 + ${C}));
	__m128i vbl${ABC[C:C+8]} = _mm_loadl_epi64((const __m128i*) (i2 + ${C}));
	__m128i vbr${ABC[C:C+8]} = _mm_loadl_epi64((const __m128i*) (i3 + ${C}));
	i0 += ${CHANNEL_TILE};
	i1 += ${CHANNEL_TILE};
	i2 += ${CHANNEL_TILE};
	i3 += ${CHANNEL_TILE};

	$if SSE != 4:
	$if DATATYPE == "U8":
	__m128i vzero = _mm_setzero_si128();
	$for C in range(0, CHANNEL_TILE, 8):
	vtl${ABC[C:C+8]} = _mm_unpacklo_epi8(vtl${ABC[C:C+8]}, vzero);
	vtr${ABC[C:C+8]} = _mm_unpacklo_epi8(vtr${ABC[C:C+8]}, vzero);
	vbl${ABC[C:C+8]} = _mm_unpacklo_epi8(vbl${ABC[C:C+8]}, vzero);
	vbr${ABC[C:C+8]} = _mm_unpacklo_epi8(vbr${ABC[C:C+8]}, vzero);
	$else:
	$for C in range(0, CHANNEL_TILE, 8):
	vtl${ABC[C:C+8]} = _mm_srai_epi16(_mm_unpacklo_epi8(vtl${ABC[C:C+8]}, vtl${ABC[C:C+8]}), 8);
	vtr${ABC[C:C+8]} = _mm_srai_epi16(_mm_unpacklo_epi8(vtr${ABC[C:C+8]}, vtr${ABC[C:C+8]}), 8);
	vbl${ABC[C:C+8]} = _mm_srai_epi16(_mm_unpacklo_epi8(vbl${ABC[C:C+8]}, vbl${ABC[C:C+8]}), 8);
	vbr${ABC[C:C+8]} = _mm_srai_epi16(_mm_unpacklo_epi8(vbr${ABC[C:C+8]}, vbr${ABC[C:C+8]}), 8);

	$for C in range(0, CHANNEL_TILE, 8):
	const __m128i vdr${ABC[C:C+8]} = _mm_sub_epi16(vbr${ABC[C:C+8]}, vtr${ABC[C:C+8]});
	const __m128i vt${ABC[C:C+4]} = _mm_madd_epi16(_mm_unpacklo_epi16(vtr${ABC[C:C+8]}, vtl${ABC[C:C+8]}), valphah);
	const __m128i vdl${ABC[C:C+8]} = _mm_sub_epi16(vbl${ABC[C:C+8]}, vtl${ABC[C:C+8]});
	const __m128i vt${ABC[C+4:C+8]} = _mm_madd_epi16(_mm_unpackhi_epi16(vtr${ABC[C:C+8]}, vtl${ABC[C:C+8]}), valphah);

	$for C in range(0, CHANNEL_TILE, 8):
	const __m128i vd${ABC[C:C+4]} = _mm_madd_epi16(_mm_unpacklo_epi16(vdr${ABC[C:C+8]}, vdl${ABC[C:C+8]}), valphah);
	const __m128i vd${ABC[C+4:C+8]} = _mm_madd_epi16(_mm_unpackhi_epi16(vdr${ABC[C:C+8]}, vdl${ABC[C:C+8]}), valphah);

	$if SSE == 4:
	$for C in range(0, CHANNEL_TILE, 4):
	__m128i vacc${ABC[C:C+4]} = _mm_mullo_epi32(vd${ABC[C:C+4]}, valphav);
	$else:
	$for C in range(0, CHANNEL_TILE, 4):
	__m128i vacc${ABC[C:C+4]} = _mm_slli_epi32(_mm_mulhi_epu16(vd${ABC[C:C+4]}, valphav), 16);

	$for C in range(0, CHANNEL_TILE, 4):
	vacc${ABC[C:C+4]} = _mm_add_epi16(_mm_mullo_epi16(vd${ABC[C:C+4]}, valphav), vacc${ABC[C:C+4]});

	$for C in range(0, CHANNEL_TILE, 4):
	vacc${ABC[C:C+4]} = _mm_add_epi32(_mm_slli_epi32(vt${ABC[C:C+4]}, 11), vacc${ABC[C:C+4]});

	$for C in range(0, CHANNEL_TILE, 4):
	vacc${ABC[C:C+4]} = ${_MM_SRXI_EPI32}(_mm_add_epi16(vacc${ABC[C:C+4]}, vrounding), 22);

	$for C in range(0, CHANNEL_TILE, 8):
	const __m128i vacc${ABC[C:C+8]} = _mm_packs_epi32(vacc${ABC[C:C+4]}, vacc${ABC[C+4:C+8]});

	$for C in range(0, CHANNEL_TILE, 16):
	$if C + 8 < CHANNEL_TILE:
	const __m128i vo${ABC[C:C+16]} = ${_MM_PACKXS_EPI16}(vacc${ABC[C:C+8]}, vacc${ABC[C+8:C+16]});
	$else:
	const __m128i vo${ABC[C:C+8]} = ${_MM_PACKXS_EPI16}(vacc${ABC[C:C+8]}, vacc${ABC[C:C+8]});

	_mm_storeu_si128((__m128i*) output, vo${ABC[0:16]});
	$for C in range(16, CHANNEL_TILE, 16):
	$if C + 8 < CHANNEL_TILE:
	_mm_storeu_si128((__m128i*) (output + ${C}), vo${ABC[C:C+16]});
	$else:
	_mm_storel_epi64((__m128i*) (output + ${C}), vo${ABC[C:C+8]});
	output += ${CHANNEL_TILE};
	}
	for (; c >= 8 * sizeof(${XINT8_T}); c -= 8 * sizeof(${XINT8_T})) {
	$if SSE == 4:
	const __m128i vtl01234567 = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i0));
	i0 += 8;
	const __m128i vtr01234567 = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i1));
	i1 += 8;
	const __m128i vbl01234567 = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i2));
	i2 += 8;
	const __m128i vbr01234567 = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i3));
	i3 += 8;
	$else:
	__m128i vtl01234567 = _mm_loadl_epi64((const __m128i*) i0);
	i0 += 8;
	__m128i vtr01234567 = _mm_loadl_epi64((const __m128i*) i1);
	i1 += 8;
	__m128i vbl01234567 = _mm_loadl_epi64((const __m128i*) i2);
	i2 += 8;
	__m128i vbr01234567 = _mm_loadl_epi64((const __m128i*) i3);
	i3 += 8;

	$if SSE != 4:
	$if DATATYPE == "U8":
	__m128i vzero = _mm_setzero_si128();
	vtl01234567 = _mm_unpacklo_epi8(vtl01234567, vzero);
	vtr01234567 = _mm_unpacklo_epi8(vtr01234567, vzero);
	vbl01234567 = _mm_unpacklo_epi8(vbl01234567, vzero);
	vbr01234567 = _mm_unpacklo_epi8(vbr01234567, vzero);
	$else:
	vtl01234567 = _mm_srai_epi16(_mm_unpacklo_epi8(vtl01234567, vtl01234567), 8);
	vtr01234567 = _mm_srai_epi16(_mm_unpacklo_epi8(vtr01234567, vtr01234567), 8);
	vbl01234567 = _mm_srai_epi16(_mm_unpacklo_epi8(vbl01234567, vbl01234567), 8);
	vbr01234567 = _mm_srai_epi16(_mm_unpacklo_epi8(vbr01234567, vbr01234567), 8);

	const __m128i vdr01234567 = _mm_sub_epi16(vbr01234567, vtr01234567);
	const __m128i vt0123 = _mm_madd_epi16(_mm_unpacklo_epi16(vtr01234567, vtl01234567), valphah);
	const __m128i vdl01234567 = _mm_sub_epi16(vbl01234567, vtl01234567);
	const __m128i vt4567 = _mm_madd_epi16(_mm_unpackhi_epi16(vtr01234567, vtl01234567), valphah);

	const __m128i vd0123 = _mm_madd_epi16(_mm_unpacklo_epi16(vdr01234567, vdl01234567), valphah);
	const __m128i vd4567 = _mm_madd_epi16(_mm_unpackhi_epi16(vdr01234567, vdl01234567), valphah);

	$if SSE == 4:
	__m128i vacc0123 = _mm_mullo_epi32(vd0123, valphav);
	__m128i vacc4567 = _mm_mullo_epi32(vd4567, valphav);
	$else:
	__m128i vacc0123 = _mm_slli_epi32(_mm_mulhi_epu16(vd0123, valphav), 16);
	__m128i vacc4567 = _mm_slli_epi32(_mm_mulhi_epu16(vd4567, valphav), 16);

	vacc0123 = _mm_add_epi16(_mm_mullo_epi16(vd0123, valphav), vacc0123);
	vacc4567 = _mm_add_epi16(_mm_mullo_epi16(vd4567, valphav), vacc4567);

	vacc0123 = _mm_add_epi32(_mm_slli_epi32(vt0123, 11), vacc0123);
	vacc4567 = _mm_add_epi32(_mm_slli_epi32(vt4567, 11), vacc4567);

	vacc0123 = ${_MM_SRXI_EPI32}(_mm_add_epi16(vacc0123, vrounding), 22);
	vacc4567 = ${_MM_SRXI_EPI32}(_mm_add_epi16(vacc4567, vrounding), 22);

	const __m128i vacc01234567 = _mm_packs_epi32(vacc0123, vacc4567);

	const __m128i vo01234567 = ${_MM_PACKXS_EPI16}(vacc01234567, vacc01234567);

	_mm_storel_epi64((__m128i*) output, vo01234567);
	output += 8;
	}
	if XNN_UNLIKELY(c != 0) {
	$if SSE == 4:
	const __m128i vtl01234567 = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i0));
	const __m128i vtr01234567 = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i1));
	const __m128i vbl01234567 = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i2));
	const __m128i vbr01234567 = ${_MM_CVTEPX8_EPI16}(_mm_loadl_epi64((const __m128i*) i3));
	$else:
	__m128i vtl01234567 = _mm_loadl_epi64((const __m128i*) i0);
	__m128i vtr01234567 = _mm_loadl_epi64((const __m128i*) i1);
	__m128i vbl01234567 = _mm_loadl_epi64((const __m128i*) i2);
	__m128i vbr01234567 = _mm_loadl_epi64((const __m128i*) i3);

	$if SSE != 4:
	$if DATATYPE == "U8":
	__m128i vzero = _mm_setzero_si128();
	vtl01234567 = _mm_unpacklo_epi8(vtl01234567, vzero);
	vtr01234567 = _mm_unpacklo_epi8(vtr01234567, vzero);
	vbl01234567 = _mm_unpacklo_epi8(vbl01234567, vzero);
	vbr01234567 = _mm_unpacklo_epi8(vbr01234567, vzero);
	$else:
	vtl01234567 = _mm_srai_epi16(_mm_unpacklo_epi8(vtl01234567, vtl01234567), 8);
	vtr01234567 = _mm_srai_epi16(_mm_unpacklo_epi8(vtr01234567, vtr01234567), 8);
	vbl01234567 = _mm_srai_epi16(_mm_unpacklo_epi8(vbl01234567, vbl01234567), 8);
	vbr01234567 = _mm_srai_epi16(_mm_unpacklo_epi8(vbr01234567, vbr01234567), 8);

	const __m128i vdr01234567 = _mm_sub_epi16(vbr01234567, vtr01234567);
	const __m128i vt0123 = _mm_madd_epi16(_mm_unpacklo_epi16(vtr01234567, vtl01234567), valphah);
	const __m128i vdl01234567 = _mm_sub_epi16(vbl01234567, vtl01234567);
	const __m128i vt4567 = _mm_madd_epi16(_mm_unpackhi_epi16(vtr01234567, vtl01234567), valphah);

	const __m128i vd0123 = _mm_madd_epi16(_mm_unpacklo_epi16(vdr01234567, vdl01234567), valphah);
	const __m128i vd4567 = _mm_madd_epi16(_mm_unpackhi_epi16(vdr01234567, vdl01234567), valphah);

	$if SSE == 4:
	__m128i vacc0123 = _mm_mullo_epi32(vd0123, valphav);
	__m128i vacc4567 = _mm_mullo_epi32(vd4567, valphav);
	$else:
	__m128i vacc0123 = _mm_slli_epi32(_mm_mulhi_epu16(vd0123, valphav), 16);
	__m128i vacc4567 = _mm_slli_epi32(_mm_mulhi_epu16(vd4567, valphav), 16);

	vacc0123 = _mm_add_epi16(_mm_mullo_epi16(vd0123, valphav), vacc0123);
	vacc4567 = _mm_add_epi16(_mm_mullo_epi16(vd4567, valphav), vacc4567);

	vacc0123 = _mm_add_epi32(_mm_slli_epi32(vt0123, 11), vacc0123);
	vacc4567 = _mm_add_epi32(_mm_slli_epi32(vt4567, 11), vacc4567);

	vacc0123 = ${_MM_SRXI_EPI32}(_mm_add_epi16(vacc0123, vrounding), 22);
	vacc4567 = ${_MM_SRXI_EPI32}(_mm_add_epi16(vacc4567, vrounding), 22);

	const __m128i vacc01234567 = _mm_packs_epi32(vacc0123, vacc4567);

	__m128i vo01234567 = ${_MM_PACKXS_EPI16}(vacc01234567, vacc01234567);

	if (c & (4 * sizeof(${XINT8_T}))) {
	unaligned_store_u32(output, (uint32_t) _mm_cvtsi128_si32(vo01234567));
	output += 4;
	vo01234567 = _mm_srli_epi64(vo01234567, 32);
	}
	$if SSE == 4:
	if (c & (2 * sizeof(${XINT8_T}))) {
	unaligned_store_u16(output, (uint16_t) _mm_extract_epi16(vo01234567, 0));
	output += 2;
	vo01234567 = _mm_srli_epi32(vo01234567, 16);
	}
	if (c & (1 * sizeof(${XINT8_T}))) {
	*output++ = (uint8_t) _mm_extract_epi8(vo01234567, 0);
	}
	$else:
	uint32_t vo0123 = (uint32_t) _mm_cvtsi128_si32(vo01234567);
	if (c & (2 * sizeof(${XINT8_T}))) {
	unaligned_store_u16(output, (uint16_t) vo0123);
	output += 2;
	vo0123 >>= 16;
	}
	if (c & (1 * sizeof(${XINT8_T}))) {
	*output++ = (uint8_t) vo0123;
	}
	}

	output = (${XINT8_T}*) ((uintptr_t) output + output_increment);
	} while (--output_pixels != 0);
	}