rfreeimage 0.1.0

data/vendor/FreeImage/Source/LibWebP/src/dsp/dsp.upsampling_sse2.c

@@ -0,0 +1,214 @@
+// 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) {
+  UPSAMPLE_32PIXELS(r1, r2, out);
+}
+
+#define UPSAMPLE_LAST_BLOCK(tb, bb, num_pixels, out) {                         \
+  uint8_t r1[17], r2[17];                                                      \
+  memcpy(r1, (tb), (num_pixels));                                              \
+  memcpy(r2, (bb), (num_pixels));                                              \
+  /* replicate last byte */                                                    \
+  memset(r1 + (num_pixels), r1[(num_pixels) - 1], 17 - (num_pixels));          \
+  memset(r2 + (num_pixels), r2[(num_pixels) - 1], 17 - (num_pixels));          \
+  /* using the shared function instead of the macro saves ~3k code size */     \
+  Upsample32Pixels(r1, r2, out);                                               \
+}
+
+#define CONVERT2RGB(FUNC, XSTEP, top_y, bottom_y,                              \
+                    top_dst, bottom_dst, cur_x, num_pixels) {                  \
+  int n;                                                                       \
+  for (n = 0; n < (num_pixels); ++n) {                                         \
+    FUNC(top_y[(cur_x) + n], r_u[n], r_v[n],                                   \
+         top_dst + ((cur_x) + n) * XSTEP);                                     \
+  }                                                                            \
+  if (bottom_y != NULL) {                                                      \
+    for (n = 0; n < (num_pixels); ++n) {                                       \
+      FUNC(bottom_y[(cur_x) + n], r_u[64 + n], r_v[64 + n],                    \
+           bottom_dst + ((cur_x) + n) * XSTEP);                                \
+    }                                                                          \
+  }                                                                            \
+}
+
+#define CONVERT2RGB_32(FUNC, XSTEP, top_y, bottom_y,                           \
+                       top_dst, bottom_dst, cur_x) do {                        \
+  FUNC##32(top_y + (cur_x), r_u, r_v, top_dst + (cur_x) * XSTEP);              \
+  if (bottom_y != NULL) {                                                      \
+    FUNC##32(bottom_y + (cur_x), r_u + 64, r_v + 64,                           \
+             bottom_dst + (cur_x) * XSTEP);                                    \
+  }                                                                            \
+} while (0)
+
+#define SSE2_UPSAMPLE_FUNC(FUNC_NAME, FUNC, XSTEP)                             \
+static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bottom_y,           \
+                      const uint8_t* top_u, const uint8_t* top_v,              \
+                      const uint8_t* cur_u, const uint8_t* cur_v,              \
+                      uint8_t* top_dst, uint8_t* bottom_dst, int len) {        \
+  int uv_pos, pos;                                                             \
+  /* 16byte-aligned array to cache reconstructed u and v */                    \
+  uint8_t uv_buf[4 * 32 + 15];                                                 \
+  uint8_t* const r_u = (uint8_t*)((uintptr_t)(uv_buf + 15) & ~15);             \
+  uint8_t* const r_v = r_u + 32;                                               \
+                                                                               \
+  assert(top_y != NULL);                                                       \
+  {   /* Treat the first pixel in regular way */                               \
+    const int u_diag = ((top_u[0] + cur_u[0]) >> 1) + 1;                       \
+    const int v_diag = ((top_v[0] + cur_v[0]) >> 1) + 1;                       \
+    const int u0_t = (top_u[0] + u_diag) >> 1;                                 \
+    const int v0_t = (top_v[0] + v_diag) >> 1;                                 \
+    FUNC(top_y[0], u0_t, v0_t, top_dst);                                       \
+    if (bottom_y != NULL) {                                                    \
+      const int u0_b = (cur_u[0] + u_diag) >> 1;                               \
+      const int v0_b = (cur_v[0] + v_diag) >> 1;                               \
+      FUNC(bottom_y[0], u0_b, v0_b, bottom_dst);                               \
+    }                                                                          \
+  }                                                                            \
+  /* For UPSAMPLE_32PIXELS, 17 u/v values must be read-able for each block */  \
+  for (pos = 1, uv_pos = 0; pos + 32 + 1 <= len; pos += 32, uv_pos += 16) {    \
+    UPSAMPLE_32PIXELS(top_u + uv_pos, cur_u + uv_pos, r_u);                    \
+    UPSAMPLE_32PIXELS(top_v + uv_pos, cur_v + uv_pos, r_v);                    \
+    CONVERT2RGB_32(FUNC, XSTEP, top_y, bottom_y, top_dst, bottom_dst, pos);    \
+  }                                                                            \
+  if (len > 1) {                                                               \
+    const int left_over = ((len + 1) >> 1) - (pos >> 1);                       \
+    assert(left_over > 0);                                                     \
+    UPSAMPLE_LAST_BLOCK(top_u + uv_pos, cur_u + uv_pos, left_over, r_u);       \
+    UPSAMPLE_LAST_BLOCK(top_v + uv_pos, cur_v + uv_pos, left_over, r_v);       \
+    CONVERT2RGB(FUNC, XSTEP, top_y, bottom_y, top_dst, bottom_dst,             \
+                pos, len - pos);                                               \
+  }                                                                            \
+}
+
+// SSE2 variants of the fancy upsampler.
+SSE2_UPSAMPLE_FUNC(UpsampleRgbLinePair,  VP8YuvToRgb,  3)
+SSE2_UPSAMPLE_FUNC(UpsampleBgrLinePair,  VP8YuvToBgr,  3)
+SSE2_UPSAMPLE_FUNC(UpsampleRgbaLinePair, VP8YuvToRgba, 4)
+SSE2_UPSAMPLE_FUNC(UpsampleBgraLinePair, VP8YuvToBgra, 4)
+
+#undef GET_M
+#undef PACK_AND_STORE
+#undef UPSAMPLE_32PIXELS
+#undef UPSAMPLE_LAST_BLOCK
+#undef CONVERT2RGB
+#undef CONVERT2RGB_32
+#undef SSE2_UPSAMPLE_FUNC
+
+#endif  // FANCY_UPSAMPLING
+
+#endif   // WEBP_USE_SSE2
+
+//------------------------------------------------------------------------------
+
+extern void WebPInitUpsamplersSSE2(void);
+
+#ifdef FANCY_UPSAMPLING
+
+extern WebPUpsampleLinePairFunc WebPUpsamplers[/* MODE_LAST */];
+
+WEBP_TSAN_IGNORE_FUNCTION void WebPInitUpsamplersSSE2(void) {
+#if defined(WEBP_USE_SSE2)
+  VP8YUVInitSSE2();
+  WebPUpsamplers[MODE_RGB]  = UpsampleRgbLinePair;
+  WebPUpsamplers[MODE_RGBA] = UpsampleRgbaLinePair;
+  WebPUpsamplers[MODE_BGR]  = UpsampleBgrLinePair;
+  WebPUpsamplers[MODE_BGRA] = UpsampleBgraLinePair;
+  WebPUpsamplers[MODE_rgbA] = UpsampleRgbaLinePair;
+  WebPUpsamplers[MODE_bgrA] = UpsampleBgraLinePair;
+#endif   // WEBP_USE_SSE2
+}
+
+#else
+
+// this empty function is to avoid an empty .o
+WEBP_TSAN_IGNORE_FUNCTION void WebPInitUpsamplersSSE2(void) {}
+
+#endif  // FANCY_UPSAMPLING

data/vendor/FreeImage/Source/LibWebP/src/dsp/dsp.yuv.c

@@ -0,0 +1,166 @@
+// Copyright 2010 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.
+// -----------------------------------------------------------------------------
+//
+// YUV->RGB conversion functions
+//
+// Author: Skal (pascal.massimino@gmail.com)
+
+#include "./yuv.h"
+
+#if defined(WEBP_YUV_USE_TABLE)
+
+static int done = 0;
+
+static WEBP_INLINE uint8_t clip(int v, int max_value) {
+  return v < 0 ? 0 : v > max_value ? max_value : v;
+}
+
+int16_t VP8kVToR[256], VP8kUToB[256];
+int32_t VP8kVToG[256], VP8kUToG[256];
+uint8_t VP8kClip[YUV_RANGE_MAX - YUV_RANGE_MIN];
+uint8_t VP8kClip4Bits[YUV_RANGE_MAX - YUV_RANGE_MIN];
+
+WEBP_TSAN_IGNORE_FUNCTION void VP8YUVInit(void) {
+  int i;
+  if (done) {
+    return;
+  }
+#ifndef USE_YUVj
+  for (i = 0; i < 256; ++i) {
+    VP8kVToR[i] = (89858 * (i - 128) + YUV_HALF) >> YUV_FIX;
+    VP8kUToG[i] = -22014 * (i - 128) + YUV_HALF;
+    VP8kVToG[i] = -45773 * (i - 128);
+    VP8kUToB[i] = (113618 * (i - 128) + YUV_HALF) >> YUV_FIX;
+  }
+  for (i = YUV_RANGE_MIN; i < YUV_RANGE_MAX; ++i) {
+    const int k = ((i - 16) * 76283 + YUV_HALF) >> YUV_FIX;
+    VP8kClip[i - YUV_RANGE_MIN] = clip(k, 255);
+    VP8kClip4Bits[i - YUV_RANGE_MIN] = clip((k + 8) >> 4, 15);
+  }
+#else
+  for (i = 0; i < 256; ++i) {
+    VP8kVToR[i] = (91881 * (i - 128) + YUV_HALF) >> YUV_FIX;
+    VP8kUToG[i] = -22554 * (i - 128) + YUV_HALF;
+    VP8kVToG[i] = -46802 * (i - 128);
+    VP8kUToB[i] = (116130 * (i - 128) + YUV_HALF) >> YUV_FIX;
+  }
+  for (i = YUV_RANGE_MIN; i < YUV_RANGE_MAX; ++i) {
+    const int k = i;
+    VP8kClip[i - YUV_RANGE_MIN] = clip(k, 255);
+    VP8kClip4Bits[i - YUV_RANGE_MIN] = clip((k + 8) >> 4, 15);
+  }
+#endif
+
+  done = 1;
+}
+
+#else
+
+WEBP_TSAN_IGNORE_FUNCTION void VP8YUVInit(void) {}
+
+#endif  // WEBP_YUV_USE_TABLE
+
+//-----------------------------------------------------------------------------
+// Plain-C version
+
+#define ROW_FUNC(FUNC_NAME, FUNC, XSTEP)                                       \
+static void FUNC_NAME(const uint8_t* y,                                        \
+                      const uint8_t* u, const uint8_t* v,                      \
+                      uint8_t* dst, int len) {                                 \
+  const uint8_t* const end = dst + (len & ~1) * XSTEP;                         \
+  while (dst != end) {                                                         \
+    FUNC(y[0], u[0], v[0], dst);                                               \
+    FUNC(y[1], u[0], v[0], dst + XSTEP);                                       \
+    y += 2;                                                                    \
+    ++u;                                                                       \
+    ++v;                                                                       \
+    dst += 2 * XSTEP;                                                          \
+  }                                                                            \
+  if (len & 1) {                                                               \
+    FUNC(y[0], u[0], v[0], dst);                                               \
+  }                                                                            \
+}                                                                              \
+
+// All variants implemented.
+ROW_FUNC(YuvToRgbRow,      VP8YuvToRgb,  3)
+ROW_FUNC(YuvToBgrRow,      VP8YuvToBgr,  3)
+ROW_FUNC(YuvToRgbaRow,     VP8YuvToRgba, 4)
+ROW_FUNC(YuvToBgraRow,     VP8YuvToBgra, 4)
+ROW_FUNC(YuvToArgbRow,     VP8YuvToArgb, 4)
+ROW_FUNC(YuvToRgba4444Row, VP8YuvToRgba4444, 2)
+ROW_FUNC(YuvToRgb565Row,   VP8YuvToRgb565, 2)
+
+#undef ROW_FUNC
+
+// Main call for processing a plane with a WebPSamplerRowFunc function:
+void WebPSamplerProcessPlane(const uint8_t* y, int y_stride,
+                             const uint8_t* u, const uint8_t* v, int uv_stride,
+                             uint8_t* dst, int dst_stride,
+                             int width, int height, WebPSamplerRowFunc func) {
+  int j;
+  for (j = 0; j < height; ++j) {
+    func(y, u, v, dst, width);
+    y += y_stride;
+    if (j & 1) {
+      u += uv_stride;
+      v += uv_stride;
+    }
+    dst += dst_stride;
+  }
+}
+
+//-----------------------------------------------------------------------------
+// Main call
+
+WebPSamplerRowFunc WebPSamplers[MODE_LAST];
+
+extern void WebPInitSamplersSSE2(void);
+extern void WebPInitSamplersMIPS32(void);
+extern void WebPInitSamplersMIPSdspR2(void);
+
+static volatile VP8CPUInfo yuv_last_cpuinfo_used =
+    (VP8CPUInfo)&yuv_last_cpuinfo_used;
+
+WEBP_TSAN_IGNORE_FUNCTION void WebPInitSamplers(void) {
+  if (yuv_last_cpuinfo_used == VP8GetCPUInfo) return;
+
+  WebPSamplers[MODE_RGB]       = YuvToRgbRow;
+  WebPSamplers[MODE_RGBA]      = YuvToRgbaRow;
+  WebPSamplers[MODE_BGR]       = YuvToBgrRow;
+  WebPSamplers[MODE_BGRA]      = YuvToBgraRow;
+  WebPSamplers[MODE_ARGB]      = YuvToArgbRow;
+  WebPSamplers[MODE_RGBA_4444] = YuvToRgba4444Row;
+  WebPSamplers[MODE_RGB_565]   = YuvToRgb565Row;
+  WebPSamplers[MODE_rgbA]      = YuvToRgbaRow;
+  WebPSamplers[MODE_bgrA]      = YuvToBgraRow;
+  WebPSamplers[MODE_Argb]      = YuvToArgbRow;
+  WebPSamplers[MODE_rgbA_4444] = YuvToRgba4444Row;
+
+  // If defined, use CPUInfo() to overwrite some pointers with faster versions.
+  if (VP8GetCPUInfo != NULL) {
+#if defined(WEBP_USE_SSE2)
+    if (VP8GetCPUInfo(kSSE2)) {
+      WebPInitSamplersSSE2();
+    }
+#endif  // WEBP_USE_SSE2
+#if defined(WEBP_USE_MIPS32)
+    if (VP8GetCPUInfo(kMIPS32)) {
+      WebPInitSamplersMIPS32();
+    }
+#endif  // WEBP_USE_MIPS32
+#if defined(WEBP_USE_MIPS_DSP_R2)
+    if (VP8GetCPUInfo(kMIPSdspR2)) {
+      WebPInitSamplersMIPSdspR2();
+    }
+#endif  // WEBP_USE_MIPS_DSP_R2
+  }
+  yuv_last_cpuinfo_used = VP8GetCPUInfo;
+}
+
+//-----------------------------------------------------------------------------

data/vendor/FreeImage/Source/LibWebP/src/dsp/dsp.yuv_mips32.c

@@ -0,0 +1,100 @@
+// Copyright 2014 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.
+// -----------------------------------------------------------------------------
+//
+// MIPS version of YUV to RGB upsampling functions.
+//
+// Author(s):  Djordje Pesut    (djordje.pesut@imgtec.com)
+//             Jovan Zelincevic (jovan.zelincevic@imgtec.com)
+
+#include "./dsp.h"
+
+#if defined(WEBP_USE_MIPS32)
+
+#include "./yuv.h"
+
+//------------------------------------------------------------------------------
+// simple point-sampling
+
+#define ROW_FUNC(FUNC_NAME, XSTEP, R, G, B, A)                                 \
+static void FUNC_NAME(const uint8_t* y,                                        \
+                      const uint8_t* u, const uint8_t* v,                      \
+                      uint8_t* dst, int len) {                                 \
+  int i, r, g, b;                                                              \
+  int temp0, temp1, temp2, temp3, temp4;                                       \
+  for (i = 0; i < (len >> 1); i++) {                                           \
+    temp1 = kVToR * v[0];                                                      \
+    temp3 = kVToG * v[0];                                                      \
+    temp2 = kUToG * u[0];                                                      \
+    temp4 = kUToB * u[0];                                                      \
+    temp0 = kYScale * y[0];                                                    \
+    temp1 += kRCst;                                                            \
+    temp3 -= kGCst;                                                            \
+    temp2 += temp3;                                                            \
+    temp4 += kBCst;                                                            \
+    r = VP8Clip8(temp0 + temp1);                                               \
+    g = VP8Clip8(temp0 - temp2);                                               \
+    b = VP8Clip8(temp0 + temp4);                                               \
+    temp0 = kYScale * y[1];                                                    \
+    dst[R] = r;                                                                \
+    dst[G] = g;                                                                \
+    dst[B] = b;                                                                \
+    if (A) dst[A] = 0xff;                                                      \
+    r = VP8Clip8(temp0 + temp1);                                               \
+    g = VP8Clip8(temp0 - temp2);                                               \
+    b = VP8Clip8(temp0 + temp4);                                               \
+    dst[R + XSTEP] = r;                                                        \
+    dst[G + XSTEP] = g;                                                        \
+    dst[B + XSTEP] = b;                                                        \
+    if (A) dst[A + XSTEP] = 0xff;                                              \
+    y += 2;                                                                    \
+    ++u;                                                                       \
+    ++v;                                                                       \
+    dst += 2 * XSTEP;                                                          \
+  }                                                                            \
+  if (len & 1) {                                                               \
+    temp1 = kVToR * v[0];                                                      \
+    temp3 = kVToG * v[0];                                                      \
+    temp2 = kUToG * u[0];                                                      \
+    temp4 = kUToB * u[0];                                                      \
+    temp0 = kYScale * y[0];                                                    \
+    temp1 += kRCst;                                                            \
+    temp3 -= kGCst;                                                            \
+    temp2 += temp3;                                                            \
+    temp4 += kBCst;                                                            \
+    r = VP8Clip8(temp0 + temp1);                                               \
+    g = VP8Clip8(temp0 - temp2);                                               \
+    b = VP8Clip8(temp0 + temp4);                                               \
+    dst[R] = r;                                                                \
+    dst[G] = g;                                                                \
+    dst[B] = b;                                                                \
+    if (A) dst[A] = 0xff;                                                      \
+  }                                                                            \
+}
+
+ROW_FUNC(YuvToRgbRow,      3, 0, 1, 2, 0)
+ROW_FUNC(YuvToRgbaRow,     4, 0, 1, 2, 3)
+ROW_FUNC(YuvToBgrRow,      3, 2, 1, 0, 0)
+ROW_FUNC(YuvToBgraRow,     4, 2, 1, 0, 3)
+
+#undef ROW_FUNC
+
+#endif   // WEBP_USE_MIPS32
+
+//------------------------------------------------------------------------------
+
+extern void WebPInitSamplersMIPS32(void);
+
+WEBP_TSAN_IGNORE_FUNCTION void WebPInitSamplersMIPS32(void) {
+#if defined(WEBP_USE_MIPS32)
+  WebPSamplers[MODE_RGB]  = YuvToRgbRow;
+  WebPSamplers[MODE_RGBA] = YuvToRgbaRow;
+  WebPSamplers[MODE_BGR]  = YuvToBgrRow;
+  WebPSamplers[MODE_BGRA] = YuvToBgraRow;
+#endif  // WEBP_USE_MIPS32
+}

data/vendor/FreeImage/Source/LibWebP/src/dsp/dsp.yuv_mips_dsp_r2.c

@@ -0,0 +1,131 @@
+// Copyright 2014 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.
+// -----------------------------------------------------------------------------
+//
+// MIPS DSPr2 version of YUV to RGB upsampling functions.
+//
+// Author(s):  Branimir Vasic (branimir.vasic@imgtec.com)
+//             Djordje Pesut  (djordje.pesut@imgtec.com)
+
+#include "./dsp.h"
+
+#if defined(WEBP_USE_MIPS_DSP_R2)
+
+#include "./yuv.h"
+
+//------------------------------------------------------------------------------
+// simple point-sampling
+
+#define ROW_FUNC_PART_1()                                                      \
+  "lbu              %[temp3],   0(%[v])                         \n\t"          \
+  "lbu              %[temp4],   0(%[u])                         \n\t"          \
+  "lbu              %[temp0],   0(%[y])                         \n\t"          \
+  "mul              %[temp1],   %[t_con_1],     %[temp3]        \n\t"          \
+  "mul              %[temp3],   %[t_con_2],     %[temp3]        \n\t"          \
+  "mul              %[temp2],   %[t_con_3],     %[temp4]        \n\t"          \
+  "mul              %[temp4],   %[t_con_4],     %[temp4]        \n\t"          \
+  "mul              %[temp0],   %[t_con_5],     %[temp0]        \n\t"          \
+  "addu             %[temp1],   %[temp1],       %[t_con_6]      \n\t"          \
+  "subu             %[temp3],   %[temp3],       %[t_con_7]      \n\t"          \
+  "addu             %[temp2],   %[temp2],       %[temp3]        \n\t"          \
+  "addu             %[temp4],   %[temp4],       %[t_con_8]      \n\t"          \
+
+#define ROW_FUNC_PART_2(R, G, B, K)                                            \
+  "addu             %[temp5],   %[temp0],       %[temp1]        \n\t"          \
+  "subu             %[temp6],   %[temp0],       %[temp2]        \n\t"          \
+  "addu             %[temp7],   %[temp0],       %[temp4]        \n\t"          \
+".if "#K"                                                       \n\t"          \
+  "lbu              %[temp0],   1(%[y])                         \n\t"          \
+".endif                                                         \n\t"          \
+  "shll_s.w         %[temp5],   %[temp5],       9               \n\t"          \
+  "shll_s.w         %[temp6],   %[temp6],       9               \n\t"          \
+".if "#K"                                                       \n\t"          \
+  "mul              %[temp0],   %[t_con_5],     %[temp0]        \n\t"          \
+".endif                                                         \n\t"          \
+  "shll_s.w         %[temp7],   %[temp7],       9               \n\t"          \
+  "precrqu_s.qb.ph  %[temp5],   %[temp5],       $zero           \n\t"          \
+  "precrqu_s.qb.ph  %[temp6],   %[temp6],       $zero           \n\t"          \
+  "precrqu_s.qb.ph  %[temp7],   %[temp7],       $zero           \n\t"          \
+  "srl              %[temp5],   %[temp5],       24              \n\t"          \
+  "srl              %[temp6],   %[temp6],       24              \n\t"          \
+  "srl              %[temp7],   %[temp7],       24              \n\t"          \
+  "sb               %[temp5],   "#R"(%[dst])                    \n\t"          \
+  "sb               %[temp6],   "#G"(%[dst])                    \n\t"          \
+  "sb               %[temp7],   "#B"(%[dst])                    \n\t"          \
+
+#define ASM_CLOBBER_LIST()                                                     \
+  : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2),             \
+    [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),             \
+    [temp6]"=&r"(temp6), [temp7]"=&r"(temp7)                                   \
+  : [t_con_1]"r"(t_con_1), [t_con_2]"r"(t_con_2), [t_con_3]"r"(t_con_3),       \
+    [t_con_4]"r"(t_con_4), [t_con_5]"r"(t_con_5), [t_con_6]"r"(t_con_6),       \
+    [u]"r"(u), [v]"r"(v), [y]"r"(y), [dst]"r"(dst),                            \
+    [t_con_7]"r"(t_con_7), [t_con_8]"r"(t_con_8)                               \
+  : "memory", "hi", "lo"                                                       \
+
+#define ROW_FUNC(FUNC_NAME, XSTEP, R, G, B, A)                                 \
+static void FUNC_NAME(const uint8_t* y,                                        \
+                      const uint8_t* u, const uint8_t* v,                      \
+                      uint8_t* dst, int len) {                                 \
+  int i;                                                                       \
+  uint32_t temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;             \
+  const int t_con_1 = kVToR;                                                   \
+  const int t_con_2 = kVToG;                                                   \
+  const int t_con_3 = kUToG;                                                   \
+  const int t_con_4 = kUToB;                                                   \
+  const int t_con_5 = kYScale;                                                 \
+  const int t_con_6 = kRCst;                                                   \
+  const int t_con_7 = kGCst;                                                   \
+  const int t_con_8 = kBCst;                                                   \
+  for (i = 0; i < (len >> 1); i++) {                                           \
+    __asm__ volatile (                                                         \
+      ROW_FUNC_PART_1()                                                        \
+      ROW_FUNC_PART_2(R, G, B, 1)                                              \
+      ROW_FUNC_PART_2(R + XSTEP, G + XSTEP, B + XSTEP, 0)                      \
+      ASM_CLOBBER_LIST()                                                       \
+    );                                                                         \
+    if (A) dst[A] = dst[A + XSTEP] = 0xff;                                     \
+    y += 2;                                                                    \
+    ++u;                                                                       \
+    ++v;                                                                       \
+    dst += 2 * XSTEP;                                                          \
+  }                                                                            \
+  if (len & 1) {                                                               \
+    __asm__ volatile (                                                         \
+      ROW_FUNC_PART_1()                                                        \
+      ROW_FUNC_PART_2(R, G, B, 0)                                              \
+      ASM_CLOBBER_LIST()                                                       \
+    );                                                                         \
+    if (A) dst[A] = 0xff;                                                      \
+  }                                                                            \
+}
+
+ROW_FUNC(YuvToRgbRow,      3, 0, 1, 2, 0)
+ROW_FUNC(YuvToRgbaRow,     4, 0, 1, 2, 3)
+ROW_FUNC(YuvToBgrRow,      3, 2, 1, 0, 0)
+ROW_FUNC(YuvToBgraRow,     4, 2, 1, 0, 3)
+
+#undef ROW_FUNC
+#undef ASM_CLOBBER_LIST
+#undef ROW_FUNC_PART_2
+#undef ROW_FUNC_PART_1
+
+#endif  // WEBP_USE_MIPS_DSP_R2
+
+//------------------------------------------------------------------------------
+
+extern void WebPInitSamplersMIPSdspR2(void);
+
+WEBP_TSAN_IGNORE_FUNCTION void WebPInitSamplersMIPSdspR2(void) {
+#if defined(WEBP_USE_MIPS_DSP_R2)
+  WebPSamplers[MODE_RGB]  = YuvToRgbRow;
+  WebPSamplers[MODE_RGBA] = YuvToRgbaRow;
+  WebPSamplers[MODE_BGR]  = YuvToBgrRow;
+  WebPSamplers[MODE_BGRA] = YuvToBgraRow;
+#endif  // WEBP_USE_MIPS_DSP_R2
+}