avcodec/x86: allow future 8-bit simple idct to have "DC only hack"

[ffmpeg] / libavcodec / exr.c

diff --git a/libavcodec/exr.c b/libavcodec/exr.c
index 71946400c53d8afdd5ecce7f304d9a4e06f47257..759880756dae949c0c45e1da235542685d80af6a 100644 (file)
--- a/libavcodec/exr.c
+++ b/libavcodec/exr.c
@@ -37,6 +37,7 @@
 #include <float.h>
 #include <zlib.h>
 
+#include "libavutil/avassert.h"
 #include "libavutil/common.h"
 #include "libavutil/imgutils.h"
 #include "libavutil/intfloat.h"
@@ -45,6 +46,11 @@
 
 #include "avcodec.h"
 #include "bytestream.h"
+
+#if HAVE_BIGENDIAN
+#include "bswapdsp.h"
+#endif
+
 #include "get_bits.h"
 #include "internal.h"
 #include "mathops.h"
@@ -116,6 +122,10 @@ typedef struct EXRContext {
     AVFrame *picture;
     AVCodecContext *avctx;
 
+#if HAVE_BIGENDIAN
+    BswapDSPContext bbdsp;
+#endif
+
     enum ExrCompr compression;
     enum ExrPixelType pixel_type;
     int channel_offsets[4]; // 0 = red, 1 = green, 2 = blue and 3 = alpha
@@ -220,9 +230,9 @@ static union av_intfloat32 exr_half2float(uint16_t hf)
  *
  * @return normalized 16-bit unsigned int
  */
-static inline uint16_t exr_flt2uint(uint32_t v)
+static inline uint16_t exr_flt2uint(int32_t v)
 {
-    unsigned int exp = v >> 23;
+    int32_t exp = v >> 23;
     // "HACK": negative values result in exp<  0, so clipping them to 0
     // is also handled by this condition, avoids explicit check for sign bit.
     if (exp <= 127 + 7 - 24) // we would shift out all bits anyway
@@ -267,21 +277,17 @@ static void predictor(uint8_t *src, int size)
 
 static void reorder_pixels(uint8_t *src, uint8_t *dst, int size)
 {
-    const int8_t *t1 = src;
-    const int8_t *t2 = src + (size + 1) / 2;
-    int8_t *s        = dst;
-    int8_t *stop     = s + size;
-
-    while (1) {
-        if (s < stop)
-            *(s++) = *(t1++);
-        else
-            break;
+    const uint8_t *t1 = src;
+    int half_size     = size / 2;
+    const uint8_t *t2 = src + half_size;
+    uint8_t *s        = dst;
+    int i;
 
-        if (s < stop)
-            *(s++) = *(t2++);
-        else
-            break;
+    av_assert1(size % 2 == 0);
+
+    for (i = 0; i < half_size; i++) {
+        *(s++) = *(t1++);
+        *(s++) = *(t2++);
     }
 }
 
@@ -751,7 +757,8 @@ static int piz_uncompress(EXRContext *s, const uint8_t *src, int ssize,
     uint16_t maxval, min_non_zero, max_non_zero;
     uint16_t *ptr;
     uint16_t *tmp = (uint16_t *)td->tmp;
-    uint8_t *out;
+    uint16_t *out;
+    uint16_t *in;
     int ret, i, j;
     int pixel_half_size;/* 1 for half, 2 for float and uint32 */
     EXRChannel *channel;
@@ -803,12 +810,11 @@ static int piz_uncompress(EXRContext *s, const uint8_t *src, int ssize,
 
     apply_lut(td->lut, tmp, dsize / sizeof(uint16_t));
 
-    out = td->uncompressed_data;
+    out = (uint16_t *)td->uncompressed_data;
     for (i = 0; i < td->ysize; i++) {
         tmp_offset = 0;
         for (j = 0; j < s->nb_channels; j++) {
-            uint16_t *in;
-            EXRChannel *channel = &s->channels[j];
+            channel = &s->channels[j];
             if (channel->pixel_type == EXR_HALF)
                 pixel_half_size = 1;
             else
@@ -816,8 +822,13 @@ static int piz_uncompress(EXRContext *s, const uint8_t *src, int ssize,
 
             in = tmp + tmp_offset * td->xsize * td->ysize + i * td->xsize * pixel_half_size;
             tmp_offset += pixel_half_size;
+
+#if HAVE_BIGENDIAN
+            s->bbdsp.bswap16_buf(out, in, td->xsize * pixel_half_size);
+#else
             memcpy(out, in, td->xsize * 2 * pixel_half_size);
-            out += td->xsize * 2 * pixel_half_size;
+#endif
+            out += td->xsize * pixel_half_size;
         }
     }
 
@@ -1049,7 +1060,7 @@ static int decode_block(AVCodecContext *avctx, void *tdata,
     uint8_t *ptr;
     uint32_t data_size;
     uint64_t line, col = 0;
-    uint64_t tileX, tileY, tileLevelX, tileLevelY;
+    uint64_t tile_x, tile_y, tile_level_x, tile_level_y;
     const uint8_t *src;
     int axmax = (avctx->width - (s->xmax + 1)) * 2 * s->desc->nb_components; /* nb pixel to add at the right of the datawindow */
     int bxmin = s->xmin * 2 * s->desc->nb_components; /* nb pixel to add at the left of the datawindow */
@@ -1067,16 +1078,16 @@ static int decode_block(AVCodecContext *avctx, void *tdata,
 
         src  = buf + line_offset + 20;
 
-        tileX = AV_RL32(src - 20);
-        tileY = AV_RL32(src - 16);
-        tileLevelX = AV_RL32(src - 12);
-        tileLevelY = AV_RL32(src - 8);
+        tile_x = AV_RL32(src - 20);
+        tile_y = AV_RL32(src - 16);
+        tile_level_x = AV_RL32(src - 12);
+        tile_level_y = AV_RL32(src - 8);
 
         data_size = AV_RL32(src - 4);
         if (data_size <= 0 || data_size > buf_size)
             return AVERROR_INVALIDDATA;
 
-        if (tileLevelX || tileLevelY) { /* tile level, is not the full res level */
+        if (tile_level_x || tile_level_y) { /* tile level, is not the full res level */
             avpriv_report_missing_feature(s->avctx, "Subres tile before full res tile");
             return AVERROR_PATCHWELCOME;
         }
@@ -1086,15 +1097,15 @@ static int decode_block(AVCodecContext *avctx, void *tdata,
             return AVERROR_PATCHWELCOME;
         }
 
-        line = s->tile_attr.ySize * tileY;
-        col = s->tile_attr.xSize * tileX;
+        line = s->tile_attr.ySize * tile_y;
+        col = s->tile_attr.xSize * tile_x;
 
         if (line < s->ymin || line > s->ymax ||
             col  < s->xmin || col  > s->xmax)
             return AVERROR_INVALIDDATA;
 
-        td->ysize = FFMIN(s->tile_attr.ySize, s->ydelta - tileY * s->tile_attr.ySize);
-        td->xsize = FFMIN(s->tile_attr.xSize, s->xdelta - tileX * s->tile_attr.xSize);
+        td->ysize = FFMIN(s->tile_attr.ySize, s->ydelta - tile_y * s->tile_attr.ySize);
+        td->xsize = FFMIN(s->tile_attr.xSize, s->xdelta - tile_x * s->tile_attr.xSize);
 
         if (col) { /* not the first tile of the line */
             bxmin = 0; /* doesn't add pixel at the left of the datawindow */
@@ -1793,6 +1804,10 @@ static av_cold int decode_init(AVCodecContext *avctx)
 
     s->avctx              = avctx;
 
+#if HAVE_BIGENDIAN
+    ff_bswapdsp_init(&s->bbdsp);
+#endif
+
     trc_func = avpriv_get_trc_function_from_trc(s->apply_trc_type);
     if (trc_func) {
         for (i = 0; i < 65536; ++i) {