Add support for decoding to 565 to libjpeg_turbo

jdmerge.c

 /*
  * jdmerge.c
  *
  * This file was part of the Independent JPEG Group's software:
  * Copyright (C) 1994-1996, Thomas G. Lane.
  * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
  * libjpeg-turbo Modifications:
- * Copyright (C) 2009, 2011, D. R. Commander.
+ * Copyright (C) 2009, 2011, 2014 D. R. Commander.
+ * Copyright (C) 2013, Linaro Limited.
  * For conditions of distribution and use, see the accompanying README file.
  *
  * This file contains code for merged upsampling/color conversion.
  *
  * This file combines functions from jdsample.c and jdcolor.c;
  * read those files first to understand what's going on.
  *
  * When the chroma components are to be upsampled by simple replication
  * (ie, box filtering), we can save some work in color conversion by
  * calculating all the output pixels corresponding to a pair of chroma
@@ skipping 234 matching lines @@
                     JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
                     JDIMENSION out_rows_avail)
 /* 2:1 vertical sampling case: may need a spare row. */
 {
   my_upsample_ptr upsample = (my_upsample_ptr) cinfo->upsample;
   JSAMPROW work_ptrs[2];
   JDIMENSION num_rows;          /* number of rows returned to caller */
 
   if (upsample->spare_full) {
     /* If we have a spare row saved from a previous cycle, just return it. */
+    JDIMENSION size = upsample->out_row_width;
+    if (cinfo->out_color_space == JCS_RGB565)
+      size = cinfo->output_width * 2;
     jcopy_sample_rows(& upsample->spare_row, 0, output_buf + *out_row_ctr, 0,
-                      1, upsample->out_row_width);
+                      1, size);
     num_rows = 1;
     upsample->spare_full = FALSE;
   } else {
     /* Figure number of rows to return to caller. */
     num_rows = 2;
     /* Not more than the distance to the end of the image. */
     if (num_rows > upsample->rows_to_go)
       num_rows = upsample->rows_to_go;
     /* And not more than what the client can accept: */
     out_rows_avail -= *out_row_ctr;
@@ skipping 135 matching lines @@
       break;
     default:
       h2v2_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr,
                                     output_buf);
       break;
   }
 }
 
 
 /*
+ * RGB565 conversion
+ */
+
+#define PACK_SHORT_565_LE(r, g, b)   ((((r) << 8) & 0xF800) |  \
+                                      (((g) << 3) & 0x7E0) | ((b) >> 3))
+#define PACK_SHORT_565_BE(r, g, b)   (((r) & 0xF8) | ((g) >> 5) |  \
+                                      (((g) << 11) & 0xE000) |  \
+                                      (((b) << 5) & 0x1F00))
+
+#define PACK_TWO_PIXELS_LE(l, r)     ((r << 16) | l)
+#define PACK_TWO_PIXELS_BE(l, r)     ((l << 16) | r)
+
+#define PACK_NEED_ALIGNMENT(ptr)  (((size_t)(ptr)) & 3)
+
+#define WRITE_TWO_PIXELS_LE(addr, pixels) {  \
+  ((INT16*)(addr))[0] = (pixels);  \
+  ((INT16*)(addr))[1] = (pixels) >> 16;  \
+}
+#define WRITE_TWO_PIXELS_BE(addr, pixels) {  \
+  ((INT16*)(addr))[1] = (pixels);  \
+  ((INT16*)(addr))[0] = (pixels) >> 16;  \
+}
+
+#define DITHER_565_R(r, dither)  ((r) + ((dither) & 0xFF))
+#define DITHER_565_G(g, dither)  ((g) + (((dither) & 0xFF) >> 1))
+#define DITHER_565_B(b, dither)  ((b) + ((dither) & 0xFF))
+
+
+/* Declarations for ordered dithering
+ *
+ * We use a 4x4 ordered dither array packed into 32 bits.  This array is
+ * sufficent for dithering RGB888 to RGB565.
+ */
+
+#define DITHER_MASK       0x3
+#define DITHER_ROTATE(x)  (((x) << 24) | (((x) >> 8) & 0x00FFFFFF))
+static const INT32 dither_matrix[4] = {
+  0x0008020A,
+  0x0C040E06,
+  0x030B0109,
+  0x0F070D05
+};
+
+
+/* Include inline routines for RGB565 conversion */
+
+#define PACK_SHORT_565 PACK_SHORT_565_LE
+#define PACK_TWO_PIXELS PACK_TWO_PIXELS_LE
+#define WRITE_TWO_PIXELS WRITE_TWO_PIXELS_LE
+#define h2v1_merged_upsample_565_internal h2v1_merged_upsample_565_le
+#define h2v1_merged_upsample_565D_internal h2v1_merged_upsample_565D_le
+#define h2v2_merged_upsample_565_internal h2v2_merged_upsample_565_le
+#define h2v2_merged_upsample_565D_internal h2v2_merged_upsample_565D_le
+#include "jdmrg565.c"
+#undef PACK_SHORT_565
+#undef PACK_TWO_PIXELS
+#undef WRITE_TWO_PIXELS
+#undef h2v1_merged_upsample_565_internal
+#undef h2v1_merged_upsample_565D_internal
+#undef h2v2_merged_upsample_565_internal
+#undef h2v2_merged_upsample_565D_internal
+
+#define PACK_SHORT_565 PACK_SHORT_565_BE
+#define PACK_TWO_PIXELS PACK_TWO_PIXELS_BE
+#define WRITE_TWO_PIXELS WRITE_TWO_PIXELS_BE
+#define h2v1_merged_upsample_565_internal h2v1_merged_upsample_565_be
+#define h2v1_merged_upsample_565D_internal h2v1_merged_upsample_565D_be
+#define h2v2_merged_upsample_565_internal h2v2_merged_upsample_565_be
+#define h2v2_merged_upsample_565D_internal h2v2_merged_upsample_565D_be
+#include "jdmrg565.c"
+#undef PACK_SHORT_565
+#undef PACK_TWO_PIXELS
+#undef WRITE_TWO_PIXELS
+#undef h2v1_merged_upsample_565_internal
+#undef h2v1_merged_upsample_565D_internal
+#undef h2v2_merged_upsample_565_internal
+#undef h2v2_merged_upsample_565D_internal
+
+
+static INLINE boolean is_big_endian(void)
+{
+  int test_value = 1;
+  if(*(char *)&test_value != 1)
+    return TRUE;
+  return FALSE;
+}
+
+
+METHODDEF(void)
+h2v1_merged_upsample_565 (j_decompress_ptr cinfo,
+                          JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr,
+                          JSAMPARRAY output_buf)
+{
+  if (is_big_endian())
+    h2v1_merged_upsample_565_be(cinfo, input_buf, in_row_group_ctr,
+                                output_buf);
+  else
+    h2v1_merged_upsample_565_le(cinfo, input_buf, in_row_group_ctr,
+                                output_buf);
+}
+
+
+METHODDEF(void)
+h2v1_merged_upsample_565D (j_decompress_ptr cinfo,
+                           JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr,
+                           JSAMPARRAY output_buf)
+{
+  if (is_big_endian())
+    h2v1_merged_upsample_565D_be(cinfo, input_buf, in_row_group_ctr,
+                                 output_buf);
+  else
+    h2v1_merged_upsample_565D_le(cinfo, input_buf, in_row_group_ctr,
+                                 output_buf);
+}
+
+
+METHODDEF(void)
+h2v2_merged_upsample_565 (j_decompress_ptr cinfo,
+                          JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr,
+                          JSAMPARRAY output_buf)
+{
+  if (is_big_endian())
+    h2v2_merged_upsample_565_be(cinfo, input_buf, in_row_group_ctr,
+                                output_buf);
+  else
+    h2v2_merged_upsample_565_le(cinfo, input_buf, in_row_group_ctr,
+                                output_buf);
+}
+
+
+METHODDEF(void)
+h2v2_merged_upsample_565D (j_decompress_ptr cinfo,
+                           JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr,
+                           JSAMPARRAY output_buf)
+{
+  if (is_big_endian())
+    h2v2_merged_upsample_565D_be(cinfo, input_buf, in_row_group_ctr,
+                                 output_buf);
+  else
+    h2v2_merged_upsample_565D_le(cinfo, input_buf, in_row_group_ctr,
+                                 output_buf);
+}
+
+
+/*
  * Module initialization routine for merged upsampling/color conversion.
  *
  * NB: this is called under the conditions determined by use_merged_upsample()
  * in jdmaster.c.  That routine MUST correspond to the actual capabilities
  * of this module; no safety checks are made here.
  */
 
 GLOBAL(void)
 jinit_merged_upsampler (j_decompress_ptr cinfo)
 {
   my_upsample_ptr upsample;
 
   upsample = (my_upsample_ptr)
     (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                                 SIZEOF(my_upsampler));
   cinfo->upsample = (struct jpeg_upsampler *) upsample;
   upsample->pub.start_pass = start_pass_merged_upsample;
   upsample->pub.need_context_rows = FALSE;
 
   upsample->out_row_width = cinfo->output_width * cinfo->out_color_components;
 
   if (cinfo->max_v_samp_factor == 2) {
     upsample->pub.upsample = merged_2v_upsample;
     if (jsimd_can_h2v2_merged_upsample())
       upsample->upmethod = jsimd_h2v2_merged_upsample;
     else
       upsample->upmethod = h2v2_merged_upsample;
+    if (cinfo->out_color_space == JCS_RGB565) {
+      if (cinfo->dither_mode != JDITHER_NONE) {
+        upsample->upmethod = h2v2_merged_upsample_565D;
+      } else {
+        upsample->upmethod = h2v2_merged_upsample_565;
+      }
+    }
     /* Allocate a spare row buffer */
     upsample->spare_row = (JSAMPROW)
       (*cinfo->mem->alloc_large) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                 (size_t) (upsample->out_row_width * SIZEOF(JSAMPLE)));
   } else {
     upsample->pub.upsample = merged_1v_upsample;
     if (jsimd_can_h2v1_merged_upsample())
       upsample->upmethod = jsimd_h2v1_merged_upsample;
     else
       upsample->upmethod = h2v1_merged_upsample;
+    if (cinfo->out_color_space == JCS_RGB565) {
+      if (cinfo->dither_mode != JDITHER_NONE) {
+        upsample->upmethod = h2v1_merged_upsample_565D;
+      } else {
+        upsample->upmethod = h2v1_merged_upsample_565;
+      }
+    }
     /* No spare row needed */
     upsample->spare_row = NULL;
   }
 
   build_ycc_rgb_table(cinfo);
 }
 
 #endif /* UPSAMPLE_MERGING_SUPPORTED */