view release on metacpan or  search on metacpan

src/Source/LibWebP/src/dsp/dsp.upsampling_sse2.c  view on Meta::CPAN

// Copyright 2011 Google Inc. All Rights Reserved.
//
// Use of this source code is governed by a BSD-style license
// that can be found in the COPYING file in the root of the source
// tree. An additional intellectual property rights grant can be found
// in the file PATENTS. All contributing project authors may
// be found in the AUTHORS file in the root of the source tree.
// -----------------------------------------------------------------------------
//
// SSE2 version of YUV to RGB upsampling functions.
//
// Author: somnath@google.com (Somnath Banerjee)

#include "./dsp.h"

#if defined(WEBP_USE_SSE2)

#include <assert.h>
#include <emmintrin.h>
#include <string.h>
#include "./yuv.h"

#ifdef FANCY_UPSAMPLING

// We compute (9*a + 3*b + 3*c + d + 8) / 16 as follows
// u = (9*a + 3*b + 3*c + d + 8) / 16
//   = (a + (a + 3*b + 3*c + d) / 8 + 1) / 2
//   = (a + m + 1) / 2
// where m = (a + 3*b + 3*c + d) / 8
//         = ((a + b + c + d) / 2 + b + c) / 4
//
// Let's say  k = (a + b + c + d) / 4.
// We can compute k as
// k = (s + t + 1) / 2 - ((a^d) | (b^c) | (s^t)) & 1
// where s = (a + d + 1) / 2 and t = (b + c + 1) / 2
//
// Then m can be written as
// m = (k + t + 1) / 2 - (((b^c) & (s^t)) | (k^t)) & 1

// Computes out = (k + in + 1) / 2 - ((ij & (s^t)) | (k^in)) & 1
#define GET_M(ij, in, out) do {                                                \
  const __m128i tmp0 = _mm_avg_epu8(k, (in));     /* (k + in + 1) / 2 */       \
  const __m128i tmp1 = _mm_and_si128((ij), st);   /* (ij) & (s^t) */           \
  const __m128i tmp2 = _mm_xor_si128(k, (in));    /* (k^in) */                 \
  const __m128i tmp3 = _mm_or_si128(tmp1, tmp2);  /* ((ij) & (s^t)) | (k^in) */\
  const __m128i tmp4 = _mm_and_si128(tmp3, one);  /* & 1 -> lsb_correction */  \
  (out) = _mm_sub_epi8(tmp0, tmp4);    /* (k + in + 1) / 2 - lsb_correction */ \
} while (0)

// pack and store two alternating pixel rows
#define PACK_AND_STORE(a, b, da, db, out) do {                                 \
  const __m128i t_a = _mm_avg_epu8(a, da);  /* (9a + 3b + 3c +  d + 8) / 16 */ \
  const __m128i t_b = _mm_avg_epu8(b, db);  /* (3a + 9b +  c + 3d + 8) / 16 */ \
  const __m128i t_1 = _mm_unpacklo_epi8(t_a, t_b);                             \
  const __m128i t_2 = _mm_unpackhi_epi8(t_a, t_b);                             \
  _mm_store_si128(((__m128i*)(out)) + 0, t_1);                                 \
  _mm_store_si128(((__m128i*)(out)) + 1, t_2);                                 \
} while (0)

// Loads 17 pixels each from rows r1 and r2 and generates 32 pixels.
#define UPSAMPLE_32PIXELS(r1, r2, out) {                                       \
  const __m128i one = _mm_set1_epi8(1);                                        \
  const __m128i a = _mm_loadu_si128((const __m128i*)&(r1)[0]);                 \
  const __m128i b = _mm_loadu_si128((const __m128i*)&(r1)[1]);                 \
  const __m128i c = _mm_loadu_si128((const __m128i*)&(r2)[0]);                 \
  const __m128i d = _mm_loadu_si128((const __m128i*)&(r2)[1]);                 \
                                                                               \
  const __m128i s = _mm_avg_epu8(a, d);        /* s = (a + d + 1) / 2 */       \
  const __m128i t = _mm_avg_epu8(b, c);        /* t = (b + c + 1) / 2 */       \
  const __m128i st = _mm_xor_si128(s, t);      /* st = s^t */                  \
                                                                               \
  const __m128i ad = _mm_xor_si128(a, d);      /* ad = a^d */                  \
  const __m128i bc = _mm_xor_si128(b, c);      /* bc = b^c */                  \
                                                                               \
  const __m128i t1 = _mm_or_si128(ad, bc);     /* (a^d) | (b^c) */             \
  const __m128i t2 = _mm_or_si128(t1, st);     /* (a^d) | (b^c) | (s^t) */     \
  const __m128i t3 = _mm_and_si128(t2, one);   /* (a^d) | (b^c) | (s^t) & 1 */ \
  const __m128i t4 = _mm_avg_epu8(s, t);                                       \
  const __m128i k = _mm_sub_epi8(t4, t3);      /* k = (a + b + c + d) / 4 */   \
  __m128i diag1, diag2;                                                        \
                                                                               \
  GET_M(bc, t, diag1);                  /* diag1 = (a + 3b + 3c + d) / 8 */    \
  GET_M(ad, s, diag2);                  /* diag2 = (3a + b + c + 3d) / 8 */    \
                                                                               \
  /* pack the alternate pixels */                                              \
  PACK_AND_STORE(a, b, diag1, diag2, out +      0);  /* store top */           \
  PACK_AND_STORE(c, d, diag2, diag1, out + 2 * 32);  /* store bottom */        \
}

// Turn the macro into a function for reducing code-size when non-critical
static void Upsample32Pixels(const uint8_t r1[], const uint8_t r2[],
                             uint8_t* const out) {