// Copyright 2023 The Wuffs Authors. // // Licensed under the Apache License, Version 2.0 or the MIT license // , at your // option. This file may not be copied, modified, or distributed // except according to those terms. // // SPDX-License-Identifier: Apache-2.0 OR MIT // -------- // ‼ WUFFS MULTI-FILE SECTION +x86_avx2 #if defined(WUFFS_PRIVATE_IMPL__CPU_ARCH__X86_64_V3) WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET("pclmul,popcnt,sse4.2,avx2") static void // wuffs_private_impl__swizzle_ycc_bt601fr__convert_3_bgrx_x86_avx2( wuffs_base__pixel_buffer* dst, uint32_t x, uint32_t x_end, uint32_t y, const uint8_t* up0, const uint8_t* up1, const uint8_t* up2) { if ((x + 32u) > x_end) { wuffs_private_impl__swizzle_ycc_bt601fr__convert_3_bgrx( // dst, x, x_end, y, up0, up1, up2); return; } size_t dst_stride = dst->private_impl.planes[0].stride; uint8_t* dst_iter = dst->private_impl.planes[0].ptr + (dst_stride * ((size_t)y)) + (4u * ((size_t)x)); // u0001 = u16x16 [0x0001 .. 0x0001] // u00FF = u16x16 [0x00FF .. 0x00FF] // uFF80 = u16x16 [0xFF80 .. 0xFF80] // uFFFF = u16x16 [0xFFFF .. 0xFFFF] const __m256i u0001 = _mm256_set1_epi16(+0x0001); const __m256i u00FF = _mm256_set1_epi16(+0x00FF); const __m256i uFF80 = _mm256_set1_epi16(-0x0080); const __m256i uFFFF = _mm256_set1_epi16(-0x0001); // p8000_p0000 = u16x16 [0x8000 0x0000 .. 0x8000 0x0000] const __m256i p8000_p0000 = _mm256_set_epi16( // +0x0000, -0x8000, +0x0000, -0x8000, // +0x0000, -0x8000, +0x0000, -0x8000, // +0x0000, -0x8000, +0x0000, -0x8000, // +0x0000, -0x8000, +0x0000, -0x8000); // Per wuffs_base__color_ycc_bt601fr__as__color_u32, the formulae: // // R = Y + 1.40200 * Cr // G = Y - 0.34414 * Cb - 0.71414 * Cr // B = Y + 1.77200 * Cb // // When scaled by 1<<16: // // 0.34414 becomes 0x0581A = 22554. // 0.71414 becomes 0x0B6D2 = 46802. // 1.40200 becomes 0x166E9 = 91881. // 1.77200 becomes 0x1C5A2 = 116130. // // Separate the integer and fractional parts, since we work with signed // 16-bit SIMD lanes. The fractional parts range from -0.5 .. +0.5 (as // floating-point) which is from -0x8000 .. +0x8000 (as fixed-point). // // -0x3A5E = -0x20000 + 0x1C5A2 The B:Cb factor. // +0x66E9 = -0x10000 + 0x166E9 The R:Cr factor. // -0x581A = +0x00000 - 0x0581A The G:Cb factor. // +0x492E = +0x10000 - 0x0B6D2 The G:Cr factor. const __m256i m3A5E = _mm256_set1_epi16(-0x3A5E); const __m256i p66E9 = _mm256_set1_epi16(+0x66E9); const __m256i m581A_p492E = _mm256_set_epi16( // +0x492E, -0x581A, +0x492E, -0x581A, // +0x492E, -0x581A, +0x492E, -0x581A, // +0x492E, -0x581A, +0x492E, -0x581A, // +0x492E, -0x581A, +0x492E, -0x581A); while (x < x_end) { // Load chroma values in even and odd columns (the high 8 bits of each // u16x16 element are zero) and then subtract 0x0080. // // cb_all = u8x32 [cb.00 cb.01 cb.02 cb.03 .. cb.1C cb.1D cb.1E cb.1F] // cb_eve = i16x16 [cb.00-0x80 cb.02-0x80 .. cb.1C-0x80 cb.1E-0x80 ] // cb_odd = i16x16 [cb.01-0x80 cb.03-0x80 .. cb.1D-0x80 cb.1F-0x80 ] // // Ditto for the cr_xxx Chroma-Red values. __m256i cb_all = _mm256_lddqu_si256((const __m256i*)(const void*)up1); __m256i cr_all = _mm256_lddqu_si256((const __m256i*)(const void*)up2); __m256i cb_eve = _mm256_add_epi16(uFF80, _mm256_and_si256(cb_all, u00FF)); __m256i cr_eve = _mm256_add_epi16(uFF80, _mm256_and_si256(cr_all, u00FF)); __m256i cb_odd = _mm256_add_epi16(uFF80, _mm256_srli_epi16(cb_all, 8)); __m256i cr_odd = _mm256_add_epi16(uFF80, _mm256_srli_epi16(cr_all, 8)); // ---- // Calculate: // // B-Y = (+1.77200 * Cb) as floating-point // R-Y = (+1.40200 * Cr) as floating-point // // B-Y = ((0x2_0000 - 0x3A5E) * Cb) as fixed-point // R-Y = ((0x1_0000 + 0x66E9) * Cr) as fixed-point // // B-Y = ((-0x3A5E * Cb) + ("2.0" * Cb)) // R-Y = ((+0x66E9 * Cr) + ("1.0" * Cr)) // Multiply by m3A5E or p66E9, taking the high 16 bits. There's also a // doubling (add x to itself), adding-of-1 and halving (shift right by 1). // That makes multiply-and-take-high round to nearest (instead of down). __m256i tmp_by_eve = _mm256_srai_epi16( _mm256_add_epi16( _mm256_mulhi_epi16(_mm256_add_epi16(cb_eve, cb_eve), m3A5E), u0001), 1); __m256i tmp_by_odd = _mm256_srai_epi16( _mm256_add_epi16( _mm256_mulhi_epi16(_mm256_add_epi16(cb_odd, cb_odd), m3A5E), u0001), 1); __m256i tmp_ry_eve = _mm256_srai_epi16( _mm256_add_epi16( _mm256_mulhi_epi16(_mm256_add_epi16(cr_eve, cr_eve), p66E9), u0001), 1); __m256i tmp_ry_odd = _mm256_srai_epi16( _mm256_add_epi16( _mm256_mulhi_epi16(_mm256_add_epi16(cr_odd, cr_odd), p66E9), u0001), 1); // Add (2 * Cb) and (1 * Cr). __m256i by_eve = _mm256_add_epi16(tmp_by_eve, _mm256_add_epi16(cb_eve, cb_eve)); __m256i by_odd = _mm256_add_epi16(tmp_by_odd, _mm256_add_epi16(cb_odd, cb_odd)); __m256i ry_eve = _mm256_add_epi16(tmp_ry_eve, cr_eve); __m256i ry_odd = _mm256_add_epi16(tmp_ry_odd, cr_odd); // ---- // Calculate: // // G-Y = (-0.34414 * Cb) + // (-0.71414 * Cr) as floating-point // // G-Y = ((+0x0_0000 - 0x581A) * Cb) + // ((-0x1_0000 + 0x492E) * Cr) as fixed-point // // G-Y = (-0x581A * Cb) + // (+0x492E * Cr) - ("1.0" * Cr) // Multiply-add to get ((-0x581A * Cb) + (+0x492E * Cr)). __m256i tmp0_gy_eve_lo = _mm256_madd_epi16( // _mm256_unpacklo_epi16(cb_eve, cr_eve), m581A_p492E); __m256i tmp0_gy_eve_hi = _mm256_madd_epi16( // _mm256_unpackhi_epi16(cb_eve, cr_eve), m581A_p492E); __m256i tmp0_gy_odd_lo = _mm256_madd_epi16( // _mm256_unpacklo_epi16(cb_odd, cr_odd), m581A_p492E); __m256i tmp0_gy_odd_hi = _mm256_madd_epi16( // _mm256_unpackhi_epi16(cb_odd, cr_odd), m581A_p492E); // Divide the i32x8 vectors by (1 << 16), rounding to nearest. __m256i tmp1_gy_eve_lo = _mm256_srai_epi32(_mm256_add_epi32(tmp0_gy_eve_lo, p8000_p0000), 16); __m256i tmp1_gy_eve_hi = _mm256_srai_epi32(_mm256_add_epi32(tmp0_gy_eve_hi, p8000_p0000), 16); __m256i tmp1_gy_odd_lo = _mm256_srai_epi32(_mm256_add_epi32(tmp0_gy_odd_lo, p8000_p0000), 16); __m256i tmp1_gy_odd_hi = _mm256_srai_epi32(_mm256_add_epi32(tmp0_gy_odd_hi, p8000_p0000), 16); // Pack the ((-0x581A * Cb) + (+0x492E * Cr)) as i16x16 and subtract Cr. __m256i gy_eve = _mm256_sub_epi16( _mm256_packs_epi32(tmp1_gy_eve_lo, tmp1_gy_eve_hi), cr_eve); __m256i gy_odd = _mm256_sub_epi16( _mm256_packs_epi32(tmp1_gy_odd_lo, tmp1_gy_odd_hi), cr_odd); // ---- // Add Y to (B-Y), (G-Y) and (R-Y) to produce B, G and R. // // For the resultant packed_x_xxx vectors, only elements 0 ..= 7 and 16 ..= // 23 of the 32-element vectors matter (since we'll unpacklo but not // unpackhi them). Let … denote 8 ignored consecutive u8 values and let % // denote 0xFF. We'll end this section with: // // packed_b_eve = u8x32 [b00 b02 .. b0C b0E … b10 b12 .. b1C b1E …] // packed_b_odd = u8x32 [b01 b03 .. b0D b0F … b11 b13 .. b1D b1F …] // packed_g_eve = u8x32 [g00 g02 .. g0C g0E … g10 g12 .. g1C g1E …] // packed_g_odd = u8x32 [g01 g03 .. g0D g0F … g11 g13 .. g1D g1F …] // packed_r_eve = u8x32 [r00 r02 .. r0C r0E … r10 r12 .. r1C r1E …] // packed_r_odd = u8x32 [r01 r03 .. r0D r0F … r11 r13 .. r1D r1F …] // uFFFF = u8x32 [ % % .. % % … % % .. % % …] __m256i yy_all = _mm256_lddqu_si256((const __m256i*)(const void*)up0); __m256i yy_eve = _mm256_and_si256(yy_all, u00FF); __m256i yy_odd = _mm256_srli_epi16(yy_all, 8); __m256i loose_b_eve = _mm256_add_epi16(by_eve, yy_eve); __m256i loose_b_odd = _mm256_add_epi16(by_odd, yy_odd); __m256i packed_b_eve = _mm256_packus_epi16(loose_b_eve, loose_b_eve); __m256i packed_b_odd = _mm256_packus_epi16(loose_b_odd, loose_b_odd); __m256i loose_g_eve = _mm256_add_epi16(gy_eve, yy_eve); __m256i loose_g_odd = _mm256_add_epi16(gy_odd, yy_odd); __m256i packed_g_eve = _mm256_packus_epi16(loose_g_eve, loose_g_eve); __m256i packed_g_odd = _mm256_packus_epi16(loose_g_odd, loose_g_odd); __m256i loose_r_eve = _mm256_add_epi16(ry_eve, yy_eve); __m256i loose_r_odd = _mm256_add_epi16(ry_odd, yy_odd); __m256i packed_r_eve = _mm256_packus_epi16(loose_r_eve, loose_r_eve); __m256i packed_r_odd = _mm256_packus_epi16(loose_r_odd, loose_r_odd); // ---- // Mix those values (unpacking in 8, 16 and then 32 bit units) to get the // desired BGRX/RGBX order. // // From here onwards, all of our __m256i registers are u8x32. // mix00 = [b00 g00 b02 g02 .. b0E g0E b10 g10 .. b1C g1C b1E g1E] // mix01 = [b01 g01 b03 g03 .. b0F g0F b11 g11 .. b1D g1D b1F g1F] // mix02 = [r00 % r02 % .. r0E % r10 % .. r1C % r1E %] // mix03 = [r01 % r03 % .. r0F % r11 % .. r1D % r1F %] // // See also § below. __m256i mix00 = _mm256_unpacklo_epi8(packed_b_eve, packed_g_eve); __m256i mix01 = _mm256_unpacklo_epi8(packed_b_odd, packed_g_odd); __m256i mix02 = _mm256_unpacklo_epi8(packed_r_eve, uFFFF); __m256i mix03 = _mm256_unpacklo_epi8(packed_r_odd, uFFFF); // mix10 = [b00 g00 r00 % b02 g02 r02 % b04 g04 r04 % b06 g06 r06 % // b10 g10 r10 % b12 g12 r12 % b14 g14 r14 % b16 g16 r16 %] // mix11 = [b01 g01 r01 % b03 g03 r03 % b05 g05 r05 % b07 g07 r07 % // b11 g11 r11 % b13 g13 r13 % b15 g15 r15 % b17 g17 r17 %] // mix12 = [b08 g08 r08 % b0A g0A r0A % b0C g0C r0C % b0E g0E r0E % // b18 g18 r18 % b1A g1A r1A % b1C g1C r1C % b1E g1E r1E %] // mix13 = [b09 g09 r09 % b0B g0B r0B % b0D g0D r0D % b0F g0F r0F % // b19 g19 r19 % b1B g1B r1B % b1D g1D r1D % b1F g1F r1F %] __m256i mix10 = _mm256_unpacklo_epi16(mix00, mix02); __m256i mix11 = _mm256_unpacklo_epi16(mix01, mix03); __m256i mix12 = _mm256_unpackhi_epi16(mix00, mix02); __m256i mix13 = _mm256_unpackhi_epi16(mix01, mix03); // mix20 = [b00 g00 r00 % b01 g01 r01 % b02 g02 r02 % b03 g03 r03 % // b10 g10 r10 % b11 g11 r11 % b12 g12 r12 % b13 g13 r13 %] // mix21 = [b04 g04 r04 % b05 g05 r05 % b06 g06 r06 % b07 g07 r07 % // b14 g14 r14 % b15 g15 r15 % b16 g16 r16 % b17 g17 r17 %] // mix22 = [b08 g08 r08 % b09 g09 r09 % b0A g0A r0A % b0B g0B r0B % // b18 g18 r18 % b19 g19 r19 % b1A g1A r1A % b1B g1B r1B %] // mix23 = [b0C g0C r0C % b0D g0D r0D % b0E g0E r0E % b0F g0F r0F % // b1C g1C r1C % b1D g1D r1D % b1E g1E r1E % b1F g1F r1F %] __m256i mix20 = _mm256_unpacklo_epi32(mix10, mix11); __m256i mix21 = _mm256_unpackhi_epi32(mix10, mix11); __m256i mix22 = _mm256_unpacklo_epi32(mix12, mix13); __m256i mix23 = _mm256_unpackhi_epi32(mix12, mix13); // mix30 = [b00 g00 r00 % b01 g01 r01 % b02 g02 r02 % b03 g03 r03 % // b04 g04 r04 % b05 g05 r05 % b06 g06 r06 % b07 g07 r07 %] // mix31 = [b08 g08 r08 % b09 g09 r09 % b0A g0A r0A % b0B g0B r0B % // b0C g0C r0C % b0D g0D r0D % b0E g0E r0E % b0F g0F r0F %] // mix32 = [b10 g10 r10 % b11 g11 r11 % b12 g12 r12 % b13 g13 r13 % // b14 g14 r14 % b15 g15 r15 % b16 g16 r16 % b17 g17 r17 %] // mix33 = [b18 g18 r18 % b19 g19 r19 % b1A g1A r1A % b1B g1B r1B % // b1C g1C r1C % b1D g1D r1D % b1E g1E r1E % b1F g1F r1F %] __m256i mix30 = _mm256_permute2x128_si256(mix20, mix21, 0x20); __m256i mix31 = _mm256_permute2x128_si256(mix22, mix23, 0x20); __m256i mix32 = _mm256_permute2x128_si256(mix20, mix21, 0x31); __m256i mix33 = _mm256_permute2x128_si256(mix22, mix23, 0x31); // Write out four u8x32 SIMD registers (128 bytes, 32 BGRX/RGBX pixels). _mm256_storeu_si256((__m256i*)(void*)(dst_iter + 0x00), mix30); _mm256_storeu_si256((__m256i*)(void*)(dst_iter + 0x20), mix31); _mm256_storeu_si256((__m256i*)(void*)(dst_iter + 0x40), mix32); _mm256_storeu_si256((__m256i*)(void*)(dst_iter + 0x60), mix33); // Advance by up to 32 pixels. The first iteration might be smaller than 32 // so that all of the remaining steps are exactly 32. uint32_t n = 32u - (31u & (x - x_end)); dst_iter += 4u * n; up0 += n; up1 += n; up2 += n; x += n; } } // The rgbx flavor (below) is exactly the same as the bgrx flavor (above) // except for the lines marked with a § and that comments were stripped. WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET("pclmul,popcnt,sse4.2,avx2") static void // wuffs_private_impl__swizzle_ycc_bt601fr__convert_3_rgbx_x86_avx2( wuffs_base__pixel_buffer* dst, uint32_t x, uint32_t x_end, uint32_t y, const uint8_t* up0, const uint8_t* up1, const uint8_t* up2) { if ((x + 32u) > x_end) { wuffs_private_impl__swizzle_ycc_bt601fr__convert_3_rgbx( // dst, x, x_end, y, up0, up1, up2); return; } size_t dst_stride = dst->private_impl.planes[0].stride; uint8_t* dst_iter = dst->private_impl.planes[0].ptr + (dst_stride * ((size_t)y)) + (4u * ((size_t)x)); const __m256i u0001 = _mm256_set1_epi16(+0x0001); const __m256i u00FF = _mm256_set1_epi16(+0x00FF); const __m256i uFF80 = _mm256_set1_epi16(-0x0080); const __m256i uFFFF = _mm256_set1_epi16(-0x0001); const __m256i p8000_p0000 = _mm256_set_epi16( // +0x0000, -0x8000, +0x0000, -0x8000, // +0x0000, -0x8000, +0x0000, -0x8000, // +0x0000, -0x8000, +0x0000, -0x8000, // +0x0000, -0x8000, +0x0000, -0x8000); const __m256i m3A5E = _mm256_set1_epi16(-0x3A5E); const __m256i p66E9 = _mm256_set1_epi16(+0x66E9); const __m256i m581A_p492E = _mm256_set_epi16( // +0x492E, -0x581A, +0x492E, -0x581A, // +0x492E, -0x581A, +0x492E, -0x581A, // +0x492E, -0x581A, +0x492E, -0x581A, // +0x492E, -0x581A, +0x492E, -0x581A); while (x < x_end) { __m256i cb_all = _mm256_lddqu_si256((const __m256i*)(const void*)up1); __m256i cr_all = _mm256_lddqu_si256((const __m256i*)(const void*)up2); __m256i cb_eve = _mm256_add_epi16(uFF80, _mm256_and_si256(cb_all, u00FF)); __m256i cr_eve = _mm256_add_epi16(uFF80, _mm256_and_si256(cr_all, u00FF)); __m256i cb_odd = _mm256_add_epi16(uFF80, _mm256_srli_epi16(cb_all, 8)); __m256i cr_odd = _mm256_add_epi16(uFF80, _mm256_srli_epi16(cr_all, 8)); __m256i tmp_by_eve = _mm256_srai_epi16( _mm256_add_epi16( _mm256_mulhi_epi16(_mm256_add_epi16(cb_eve, cb_eve), m3A5E), u0001), 1); __m256i tmp_by_odd = _mm256_srai_epi16( _mm256_add_epi16( _mm256_mulhi_epi16(_mm256_add_epi16(cb_odd, cb_odd), m3A5E), u0001), 1); __m256i tmp_ry_eve = _mm256_srai_epi16( _mm256_add_epi16( _mm256_mulhi_epi16(_mm256_add_epi16(cr_eve, cr_eve), p66E9), u0001), 1); __m256i tmp_ry_odd = _mm256_srai_epi16( _mm256_add_epi16( _mm256_mulhi_epi16(_mm256_add_epi16(cr_odd, cr_odd), p66E9), u0001), 1); __m256i by_eve = _mm256_add_epi16(tmp_by_eve, _mm256_add_epi16(cb_eve, cb_eve)); __m256i by_odd = _mm256_add_epi16(tmp_by_odd, _mm256_add_epi16(cb_odd, cb_odd)); __m256i ry_eve = _mm256_add_epi16(tmp_ry_eve, cr_eve); __m256i ry_odd = _mm256_add_epi16(tmp_ry_odd, cr_odd); __m256i tmp0_gy_eve_lo = _mm256_madd_epi16( // _mm256_unpacklo_epi16(cb_eve, cr_eve), m581A_p492E); __m256i tmp0_gy_eve_hi = _mm256_madd_epi16( // _mm256_unpackhi_epi16(cb_eve, cr_eve), m581A_p492E); __m256i tmp0_gy_odd_lo = _mm256_madd_epi16( // _mm256_unpacklo_epi16(cb_odd, cr_odd), m581A_p492E); __m256i tmp0_gy_odd_hi = _mm256_madd_epi16( // _mm256_unpackhi_epi16(cb_odd, cr_odd), m581A_p492E); __m256i tmp1_gy_eve_lo = _mm256_srai_epi32(_mm256_add_epi32(tmp0_gy_eve_lo, p8000_p0000), 16); __m256i tmp1_gy_eve_hi = _mm256_srai_epi32(_mm256_add_epi32(tmp0_gy_eve_hi, p8000_p0000), 16); __m256i tmp1_gy_odd_lo = _mm256_srai_epi32(_mm256_add_epi32(tmp0_gy_odd_lo, p8000_p0000), 16); __m256i tmp1_gy_odd_hi = _mm256_srai_epi32(_mm256_add_epi32(tmp0_gy_odd_hi, p8000_p0000), 16); __m256i gy_eve = _mm256_sub_epi16( _mm256_packs_epi32(tmp1_gy_eve_lo, tmp1_gy_eve_hi), cr_eve); __m256i gy_odd = _mm256_sub_epi16( _mm256_packs_epi32(tmp1_gy_odd_lo, tmp1_gy_odd_hi), cr_odd); __m256i yy_all = _mm256_lddqu_si256((const __m256i*)(const void*)up0); __m256i yy_eve = _mm256_and_si256(yy_all, u00FF); __m256i yy_odd = _mm256_srli_epi16(yy_all, 8); __m256i loose_b_eve = _mm256_add_epi16(by_eve, yy_eve); __m256i loose_b_odd = _mm256_add_epi16(by_odd, yy_odd); __m256i packed_b_eve = _mm256_packus_epi16(loose_b_eve, loose_b_eve); __m256i packed_b_odd = _mm256_packus_epi16(loose_b_odd, loose_b_odd); __m256i loose_g_eve = _mm256_add_epi16(gy_eve, yy_eve); __m256i loose_g_odd = _mm256_add_epi16(gy_odd, yy_odd); __m256i packed_g_eve = _mm256_packus_epi16(loose_g_eve, loose_g_eve); __m256i packed_g_odd = _mm256_packus_epi16(loose_g_odd, loose_g_odd); __m256i loose_r_eve = _mm256_add_epi16(ry_eve, yy_eve); __m256i loose_r_odd = _mm256_add_epi16(ry_odd, yy_odd); __m256i packed_r_eve = _mm256_packus_epi16(loose_r_eve, loose_r_eve); __m256i packed_r_odd = _mm256_packus_epi16(loose_r_odd, loose_r_odd); // § Note the swapped B and R channels. __m256i mix00 = _mm256_unpacklo_epi8(packed_r_eve, packed_g_eve); __m256i mix01 = _mm256_unpacklo_epi8(packed_r_odd, packed_g_odd); __m256i mix02 = _mm256_unpacklo_epi8(packed_b_eve, uFFFF); __m256i mix03 = _mm256_unpacklo_epi8(packed_b_odd, uFFFF); __m256i mix10 = _mm256_unpacklo_epi16(mix00, mix02); __m256i mix11 = _mm256_unpacklo_epi16(mix01, mix03); __m256i mix12 = _mm256_unpackhi_epi16(mix00, mix02); __m256i mix13 = _mm256_unpackhi_epi16(mix01, mix03); __m256i mix20 = _mm256_unpacklo_epi32(mix10, mix11); __m256i mix21 = _mm256_unpackhi_epi32(mix10, mix11); __m256i mix22 = _mm256_unpacklo_epi32(mix12, mix13); __m256i mix23 = _mm256_unpackhi_epi32(mix12, mix13); __m256i mix30 = _mm256_permute2x128_si256(mix20, mix21, 0x20); __m256i mix31 = _mm256_permute2x128_si256(mix22, mix23, 0x20); __m256i mix32 = _mm256_permute2x128_si256(mix20, mix21, 0x31); __m256i mix33 = _mm256_permute2x128_si256(mix22, mix23, 0x31); _mm256_storeu_si256((__m256i*)(void*)(dst_iter + 0x00), mix30); _mm256_storeu_si256((__m256i*)(void*)(dst_iter + 0x20), mix31); _mm256_storeu_si256((__m256i*)(void*)(dst_iter + 0x40), mix32); _mm256_storeu_si256((__m256i*)(void*)(dst_iter + 0x60), mix33); uint32_t n = 32u - (31u & (x - x_end)); dst_iter += 4u * n; up0 += n; up1 += n; up2 += n; x += n; } } #if defined(__GNUC__) && !defined(__clang__) // No-op. #else WUFFS_BASE__MAYBE_ATTRIBUTE_TARGET("pclmul,popcnt,sse4.2,avx2") static const uint8_t* // wuffs_private_impl__swizzle_ycc__upsample_inv_h2v2_triangle_x86_avx2( uint8_t* dst_ptr, const uint8_t* src_ptr_major, const uint8_t* src_ptr_minor, size_t src_len, uint32_t h1v2_bias_ignored, bool first_column, bool last_column, uint8_t odd_column_rounding) { uint8_t* dp = dst_ptr; const uint8_t* sp_major = src_ptr_major; const uint8_t* sp_minor = src_ptr_minor; const uint8_t oc = odd_column_rounding; if (first_column) { src_len--; if ((src_len <= 0u) && last_column) { uint32_t sv = (12u * ((uint32_t)(*sp_major++))) + // (4u * ((uint32_t)(*sp_minor++))); *dp++ = (uint8_t)((sv + 8u) >> 4u); *dp++ = (uint8_t)((sv + oc) >> 4u); return dst_ptr; } uint32_t sv_major_m1 = sp_major[-0]; // Clamp offset to zero. uint32_t sv_minor_m1 = sp_minor[-0]; // Clamp offset to zero. uint32_t sv_major_p1 = sp_major[+1]; uint32_t sv_minor_p1 = sp_minor[+1]; uint32_t sv = (9u * ((uint32_t)(*sp_major++))) + // (3u * ((uint32_t)(*sp_minor++))); *dp++ = (uint8_t)((sv + (3u * sv_major_m1) + (sv_minor_m1) + 8u) >> 4u); *dp++ = (uint8_t)((sv + (3u * sv_major_p1) + (sv_minor_p1) + oc) >> 4u); if (src_len <= 0u) { return dst_ptr; } } if (last_column) { src_len--; } if (src_len < 32) { // This fallback is the same as the non-SIMD-capable code path. for (; src_len > 0u; src_len--) { uint32_t sv_major_m1 = sp_major[-1]; uint32_t sv_minor_m1 = sp_minor[-1]; uint32_t sv_major_p1 = sp_major[+1]; uint32_t sv_minor_p1 = sp_minor[+1]; uint32_t sv = (9u * ((uint32_t)(*sp_major++))) + // (3u * ((uint32_t)(*sp_minor++))); *dp++ = (uint8_t)((sv + (3u * sv_major_m1) + (sv_minor_m1) + 8u) >> 4u); *dp++ = (uint8_t)((sv + (3u * sv_major_p1) + (sv_minor_p1) + oc) >> 4u); } } else { while (src_len > 0u) { // Load 1+32+1 samples (six u8x32 vectors) from the major (jxx) and minor // (nxx) rows. // // major_p0 = [j00 j01 j02 j03 .. j28 j29 j30 j31] // p0 = "plus 0" // minor_p0 = [n00 n01 n02 n03 .. n28 n29 n30 n31] // p0 = "plus 0" // major_m1 = [jm1 j00 j01 j02 .. j27 j28 j29 j30] // m1 = "minus 1" // minor_m1 = [nm1 n00 n01 n02 .. n27 n28 n29 n30] // m1 = "minus 1" // major_p1 = [j01 j02 j03 j04 .. j29 j30 j31 j32] // p1 = "plus 1" // minor_p1 = [n01 n02 n03 n04 .. n29 n30 n31 n32] // p1 = "plus 1" __m256i major_p0 = _mm256_lddqu_si256((const __m256i*)(const void*)(sp_major + 0)); __m256i minor_p0 = _mm256_lddqu_si256((const __m256i*)(const void*)(sp_minor + 0)); __m256i major_m1 = _mm256_lddqu_si256((const __m256i*)(const void*)(sp_major - 1)); __m256i minor_m1 = _mm256_lddqu_si256((const __m256i*)(const void*)(sp_minor - 1)); __m256i major_p1 = _mm256_lddqu_si256((const __m256i*)(const void*)(sp_major + 1)); __m256i minor_p1 = _mm256_lddqu_si256((const __m256i*)(const void*)(sp_minor + 1)); // Unpack, staying with u8x32 vectors. // // step1_p0_lo = [j00 n00 j01 n01 .. j07 n07 j16 n16 j17 n17 .. j23 n23] // step1_p0_hi = [j08 n08 j09 n09 .. j15 n15 j24 n24 j25 n25 .. j31 n31] // step1_m1_lo = [jm1 nm1 j00 n00 .. j06 n06 j15 n15 j16 n16 .. j22 n22] // step1_m1_hi = [j07 n07 j08 n08 .. j14 n14 j23 n23 j24 n24 .. j30 n30] // step1_p1_lo = [j01 n01 j02 n02 .. j08 n08 j17 n17 j18 n18 .. j24 n24] // step1_p1_hi = [j09 n09 j10 n10 .. j16 n16 j25 n25 j26 n26 .. j32 n32] __m256i step1_p0_lo = _mm256_unpacklo_epi8(major_p0, minor_p0); __m256i step1_p0_hi = _mm256_unpackhi_epi8(major_p0, minor_p0); __m256i step1_m1_lo = _mm256_unpacklo_epi8(major_m1, minor_m1); __m256i step1_m1_hi = _mm256_unpackhi_epi8(major_m1, minor_m1); __m256i step1_p1_lo = _mm256_unpacklo_epi8(major_p1, minor_p1); __m256i step1_p1_hi = _mm256_unpackhi_epi8(major_p1, minor_p1); // Multiply-add to get u16x16 vectors. // // step2_p0_lo = [9*j00+3*n00 9*j01+3*n01 .. 9*j23+3*n23] // step2_p0_hi = [9*j08+3*n08 9*j09+3*n09 .. 9*j31+3*n31] // step2_m1_lo = [3*jm1+1*nm1 3*j00+1*n00 .. 3*j22+1*n22] // step2_m1_hi = [3*j07+1*n07 3*j08+1*n08 .. 3*j30+1*n30] // step2_p1_lo = [3*j01+1*n01 3*j02+1*n02 .. 3*j24+1*n24] // step2_p1_hi = [3*j09+1*n09 3*j10+1*n10 .. 3*j32+1*n32] const __m256i k0309 = _mm256_set1_epi16(0x0309); const __m256i k0103 = _mm256_set1_epi16(0x0103); __m256i step2_p0_lo = _mm256_maddubs_epi16(step1_p0_lo, k0309); __m256i step2_p0_hi = _mm256_maddubs_epi16(step1_p0_hi, k0309); __m256i step2_m1_lo = _mm256_maddubs_epi16(step1_m1_lo, k0103); __m256i step2_m1_hi = _mm256_maddubs_epi16(step1_m1_hi, k0103); __m256i step2_p1_lo = _mm256_maddubs_epi16(step1_p1_lo, k0103); __m256i step2_p1_hi = _mm256_maddubs_epi16(step1_p1_hi, k0103); // Compute the weighted sums of (p0, m1) and (p0, p1). For example: // // step3_m1_lo[00] = ((9*j00) + (3*n00) + (3*jm1) + (1*nm1)) as u16 // step3_p1_hi[15] = ((9*j31) + (3*n31) + (3*j32) + (1*n32)) as u16 __m256i step3_m1_lo = _mm256_add_epi16(step2_p0_lo, step2_m1_lo); __m256i step3_m1_hi = _mm256_add_epi16(step2_p0_hi, step2_m1_hi); __m256i step3_p1_lo = _mm256_add_epi16(step2_p0_lo, step2_p1_lo); __m256i step3_p1_hi = _mm256_add_epi16(step2_p0_hi, step2_p1_hi); // Bias by 8 (on the left) or 7 (on the right) and then divide by 16 // (which is 9+3+3+1) to get a weighted average. On the left (m1), shift // the u16 right value by 4. On the right (p1), shift right by 4 and then // shift left by 8 so that, when still in the u16x16 little-endian // interpretation, we have: // - m1_element = (etcetera + 8) >> 4 // - p1_element = ((etcetera + oc) >> 4) << 8 // // step4_m1_lo = [0x00?? 0x00?? ... 0x00?? 0x00??] // step4_p1_lo = [0x??00 0x??00 ... 0x??00 0x??00] // step4_m1_hi = [0x00?? 0x00?? ... 0x00?? 0x00??] // step4_p1_hi = [0x??00 0x??00 ... 0x??00 0x??00] __m256i step4_m1_lo = _mm256_srli_epi16( _mm256_add_epi16(step3_m1_lo, _mm256_set1_epi16(8)), 4); __m256i step4_p1_lo = _mm256_slli_epi16( _mm256_srli_epi16( _mm256_add_epi16(step3_p1_lo, _mm256_set1_epi16(oc)), 4), 8); __m256i step4_m1_hi = _mm256_srli_epi16( _mm256_add_epi16(step3_m1_hi, _mm256_set1_epi16(8)), 4); __m256i step4_p1_hi = _mm256_slli_epi16( _mm256_srli_epi16( _mm256_add_epi16(step3_p1_hi, _mm256_set1_epi16(oc)), 4), 8); // Bitwise-or two "0x00"-rich u16x16 vectors to get a u8x32 vector. Do // that twice. Once for the low columns and once for the high columns. // // In terms of jxx (major row) or nxx (minor row) source samples: // - low columns means ( 0 .. 8; 16 .. 24). // - high columns means ( 8 .. 16; 24 .. 32). // // In terms of dxx destination samples (there are twice as many): // - low columns means ( 0 .. 16; 32 .. 48). // - high columns means (16 .. 32; 48 .. 64). // // step5_lo = [d00 d01 .. d14 d15 d32 d33 .. d46 d47] // step5_hi = [d16 d17 .. d30 d31 d48 d49 .. d62 d63] // // The d00, d02 ... d62 even elements come from (p0, m1) weighted sums. // The d01, d03 ... d63 odd elements come from (p0, p1) weighted sums. __m256i step5_lo = _mm256_or_si256(step4_m1_lo, step4_p1_lo); __m256i step5_hi = _mm256_or_si256(step4_m1_hi, step4_p1_hi); // Permute and store. // // step6_00_31 = [d00 d01 .. d14 d15 d16 d17 .. d30 d31] // step6_32_63 = [d32 d33 .. d46 d47 d48 d49 .. d62 d63] __m256i step6_00_31 = _mm256_permute2x128_si256(step5_lo, step5_hi, 0x20); __m256i step6_32_63 = _mm256_permute2x128_si256(step5_lo, step5_hi, 0x31); _mm256_storeu_si256((__m256i*)(void*)(dp + 0x00), step6_00_31); _mm256_storeu_si256((__m256i*)(void*)(dp + 0x20), step6_32_63); // Advance by up to 32 source samples (64 destination samples). The first // iteration might be smaller than 32 so that all of the remaining steps // are exactly 32. size_t n = 32u - (31u & (0u - src_len)); dp += 2u * n; sp_major += n; sp_minor += n; src_len -= n; } } if (last_column) { uint32_t sv_major_m1 = sp_major[-1]; uint32_t sv_minor_m1 = sp_minor[-1]; uint32_t sv_major_p1 = sp_major[+0]; // Clamp offset to zero. uint32_t sv_minor_p1 = sp_minor[+0]; // Clamp offset to zero. uint32_t sv = (9u * ((uint32_t)(*sp_major++))) + // (3u * ((uint32_t)(*sp_minor++))); *dp++ = (uint8_t)((sv + (3u * sv_major_m1) + (sv_minor_m1) + 8u) >> 4u); *dp++ = (uint8_t)((sv + (3u * sv_major_p1) + (sv_minor_p1) + oc) >> 4u); } return dst_ptr; } #endif #endif // defined(WUFFS_PRIVATE_IMPL__CPU_ARCH__X86_64_V3) // ‼ WUFFS MULTI-FILE SECTION -x86_avx2