diff options
Diffstat (limited to 'gpr/source/lib/dng_sdk/dng_reference.cpp')
-rw-r--r-- | gpr/source/lib/dng_sdk/dng_reference.cpp | 2783 |
1 files changed, 2783 insertions, 0 deletions
diff --git a/gpr/source/lib/dng_sdk/dng_reference.cpp b/gpr/source/lib/dng_sdk/dng_reference.cpp new file mode 100644 index 0000000..5d35378 --- /dev/null +++ b/gpr/source/lib/dng_sdk/dng_reference.cpp @@ -0,0 +1,2783 @@ +/*****************************************************************************/
+// Copyright 2006-2009 Adobe Systems Incorporated
+// All Rights Reserved.
+//
+// NOTICE: Adobe permits you to use, modify, and distribute this file in
+// accordance with the terms of the Adobe license agreement accompanying it.
+/*****************************************************************************/
+
+/* $Id: //mondo/dng_sdk_1_4/dng_sdk/source/dng_reference.cpp#1 $ */
+/* $DateTime: 2012/05/30 13:28:51 $ */
+/* $Change: 832332 $ */
+/* $Author: tknoll $ */
+
+/*****************************************************************************/
+
+#include "dng_reference.h"
+
+#include "dng_1d_table.h"
+#include "dng_hue_sat_map.h"
+#include "dng_matrix.h"
+#include "dng_resample.h"
+#include "dng_utils.h"
+ +#include "stdc_includes.h" +
+/*****************************************************************************/
+
+// This module contains routines that should be as fast as possible, even
+// at the expense of slight code size increases.
+
+#include "dng_fast_module.h"
+
+/*****************************************************************************/
+
+void RefZeroBytes (void *dPtr,
+ uint32 count)
+ {
+
+ memset (dPtr, 0, count);
+
+ }
+
+/*****************************************************************************/
+
+void RefCopyBytes (const void *sPtr,
+ void *dPtr,
+ uint32 count)
+ {
+
+ memcpy (dPtr, sPtr, count);
+
+ }
+
+/*****************************************************************************/
+
+void RefSwapBytes16 (uint16 *dPtr,
+ uint32 count)
+ {
+
+ for (uint32 j = 0; j < count; j++)
+ {
+
+ dPtr [j] = SwapBytes16 (dPtr [j]);
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefSwapBytes32 (uint32 *dPtr,
+ uint32 count)
+ {
+
+ for (uint32 j = 0; j < count; j++)
+ {
+
+ dPtr [j] = SwapBytes32 (dPtr [j]);
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefSetArea8 (uint8 *dPtr,
+ uint8 value,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 rowStep,
+ int32 colStep,
+ int32 planeStep)
+ {
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ uint8 *dPtr1 = dPtr;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ uint8 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ *dPtr2 = value;
+
+ dPtr2 += planeStep;
+
+ }
+
+ dPtr1 += colStep;
+
+ }
+
+ dPtr += rowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefSetArea16 (uint16 *dPtr,
+ uint16 value,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 rowStep,
+ int32 colStep,
+ int32 planeStep)
+ {
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ uint16 *dPtr1 = dPtr;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ uint16 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ *dPtr2 = value;
+
+ dPtr2 += planeStep;
+
+ }
+
+ dPtr1 += colStep;
+
+ }
+
+ dPtr += rowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefSetArea32 (uint32 *dPtr,
+ uint32 value,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 rowStep,
+ int32 colStep,
+ int32 planeStep)
+ {
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ uint32 *dPtr1 = dPtr;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ uint32 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ *dPtr2 = value;
+
+ dPtr2 += planeStep;
+
+ }
+
+ dPtr1 += colStep;
+
+ }
+
+ dPtr += rowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefCopyArea8 (const uint8 *sPtr,
+ uint8 *dPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 sRowStep,
+ int32 sColStep,
+ int32 sPlaneStep,
+ int32 dRowStep,
+ int32 dColStep,
+ int32 dPlaneStep)
+ {
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ const uint8 *sPtr1 = sPtr;
+ uint8 *dPtr1 = dPtr;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ const uint8 *sPtr2 = sPtr1;
+ uint8 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ *dPtr2 = *sPtr2;
+
+ sPtr2 += sPlaneStep;
+ dPtr2 += dPlaneStep;
+
+ }
+
+ sPtr1 += sColStep;
+ dPtr1 += dColStep;
+
+ }
+
+ sPtr += sRowStep;
+ dPtr += dRowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefCopyArea16 (const uint16 *sPtr,
+ uint16 *dPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 sRowStep,
+ int32 sColStep,
+ int32 sPlaneStep,
+ int32 dRowStep,
+ int32 dColStep,
+ int32 dPlaneStep)
+ {
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ const uint16 *sPtr1 = sPtr;
+ uint16 *dPtr1 = dPtr;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ const uint16 *sPtr2 = sPtr1;
+ uint16 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ *dPtr2 = *sPtr2;
+
+ sPtr2 += sPlaneStep;
+ dPtr2 += dPlaneStep;
+
+ }
+
+ sPtr1 += sColStep;
+ dPtr1 += dColStep;
+
+ }
+
+ sPtr += sRowStep;
+ dPtr += dRowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefCopyArea32 (const uint32 *sPtr,
+ uint32 *dPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 sRowStep,
+ int32 sColStep,
+ int32 sPlaneStep,
+ int32 dRowStep,
+ int32 dColStep,
+ int32 dPlaneStep)
+ {
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ const uint32 *sPtr1 = sPtr;
+ uint32 *dPtr1 = dPtr;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ const uint32 *sPtr2 = sPtr1;
+ uint32 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ *dPtr2 = *sPtr2;
+
+ sPtr2 += sPlaneStep;
+ dPtr2 += dPlaneStep;
+
+ }
+
+ sPtr1 += sColStep;
+ dPtr1 += dColStep;
+
+ }
+
+ sPtr += sRowStep;
+ dPtr += dRowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefCopyArea8_16 (const uint8 *sPtr,
+ uint16 *dPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 sRowStep,
+ int32 sColStep,
+ int32 sPlaneStep,
+ int32 dRowStep,
+ int32 dColStep,
+ int32 dPlaneStep)
+ {
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ const uint8 *sPtr1 = sPtr;
+ uint16 *dPtr1 = dPtr;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ const uint8 *sPtr2 = sPtr1;
+ uint16 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ *dPtr2 = *sPtr2;
+
+ sPtr2 += sPlaneStep;
+ dPtr2 += dPlaneStep;
+
+ }
+
+ sPtr1 += sColStep;
+ dPtr1 += dColStep;
+
+ }
+
+ sPtr += sRowStep;
+ dPtr += dRowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefCopyArea8_S16 (const uint8 *sPtr,
+ int16 *dPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 sRowStep,
+ int32 sColStep,
+ int32 sPlaneStep,
+ int32 dRowStep,
+ int32 dColStep,
+ int32 dPlaneStep)
+ {
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ const uint8 *sPtr1 = sPtr;
+ int16 *dPtr1 = dPtr;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ const uint8 *sPtr2 = sPtr1;
+ int16 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ int16 x = *sPtr;
+
+ *dPtr2 = x ^ 0x8000;
+
+ sPtr2 += sPlaneStep;
+ dPtr2 += dPlaneStep;
+
+ }
+
+ sPtr1 += sColStep;
+ dPtr1 += dColStep;
+
+ }
+
+ sPtr += sRowStep;
+ dPtr += dRowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefCopyArea8_32 (const uint8 *sPtr,
+ uint32 *dPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 sRowStep,
+ int32 sColStep,
+ int32 sPlaneStep,
+ int32 dRowStep,
+ int32 dColStep,
+ int32 dPlaneStep)
+ {
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ const uint8 *sPtr1 = sPtr;
+ uint32 *dPtr1 = dPtr;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ const uint8 *sPtr2 = sPtr1;
+ uint32 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ *dPtr2 = *sPtr2;
+
+ sPtr2 += sPlaneStep;
+ dPtr2 += dPlaneStep;
+
+ }
+
+ sPtr1 += sColStep;
+ dPtr1 += dColStep;
+
+ }
+
+ sPtr += sRowStep;
+ dPtr += dRowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefCopyArea16_S16 (const uint16 *sPtr,
+ int16 *dPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 sRowStep,
+ int32 sColStep,
+ int32 sPlaneStep,
+ int32 dRowStep,
+ int32 dColStep,
+ int32 dPlaneStep)
+ {
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ const uint16 *sPtr1 = sPtr;
+ int16 *dPtr1 = dPtr;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ const uint16 *sPtr2 = sPtr1;
+ int16 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ *dPtr2 = *sPtr2 ^ 0x8000;
+
+ sPtr2 += sPlaneStep;
+ dPtr2 += dPlaneStep;
+
+ }
+
+ sPtr1 += sColStep;
+ dPtr1 += dColStep;
+
+ }
+
+ sPtr += sRowStep;
+ dPtr += dRowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefCopyArea16_32 (const uint16 *sPtr,
+ uint32 *dPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 sRowStep,
+ int32 sColStep,
+ int32 sPlaneStep,
+ int32 dRowStep,
+ int32 dColStep,
+ int32 dPlaneStep)
+ {
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ const uint16 *sPtr1 = sPtr;
+ uint32 *dPtr1 = dPtr;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ const uint16 *sPtr2 = sPtr1;
+ uint32 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ *dPtr2 = *sPtr2;
+
+ sPtr2 += sPlaneStep;
+ dPtr2 += dPlaneStep;
+
+ }
+
+ sPtr1 += sColStep;
+ dPtr1 += dColStep;
+
+ }
+
+ sPtr += sRowStep;
+ dPtr += dRowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefCopyArea8_R32 (const uint8 *sPtr,
+ real32 *dPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 sRowStep,
+ int32 sColStep,
+ int32 sPlaneStep,
+ int32 dRowStep,
+ int32 dColStep,
+ int32 dPlaneStep,
+ uint32 pixelRange)
+ {
+
+ real32 scale = 1.0f / (real32) pixelRange;
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ const uint8 *sPtr1 = sPtr;
+ real32 *dPtr1 = dPtr;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ const uint8 *sPtr2 = sPtr1;
+ real32 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ *dPtr2 = scale * (real32) *sPtr2;
+
+ sPtr2 += sPlaneStep;
+ dPtr2 += dPlaneStep;
+
+ }
+
+ sPtr1 += sColStep;
+ dPtr1 += dColStep;
+
+ }
+
+ sPtr += sRowStep;
+ dPtr += dRowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefCopyArea16_R32 (const uint16 *sPtr,
+ real32 *dPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 sRowStep,
+ int32 sColStep,
+ int32 sPlaneStep,
+ int32 dRowStep,
+ int32 dColStep,
+ int32 dPlaneStep,
+ uint32 pixelRange)
+ {
+
+ real32 scale = 1.0f / (real32) pixelRange;
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ const uint16 *sPtr1 = sPtr;
+ real32 *dPtr1 = dPtr;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ const uint16 *sPtr2 = sPtr1;
+ real32 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ *dPtr2 = scale * (real32) *sPtr2;
+
+ sPtr2 += sPlaneStep;
+ dPtr2 += dPlaneStep;
+
+ }
+
+ sPtr1 += sColStep;
+ dPtr1 += dColStep;
+
+ }
+
+ sPtr += sRowStep;
+ dPtr += dRowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefCopyAreaS16_R32 (const int16 *sPtr,
+ real32 *dPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 sRowStep,
+ int32 sColStep,
+ int32 sPlaneStep,
+ int32 dRowStep,
+ int32 dColStep,
+ int32 dPlaneStep,
+ uint32 pixelRange)
+ {
+
+ real32 scale = 1.0f / (real32) pixelRange;
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ const int16 *sPtr1 = sPtr;
+ real32 *dPtr1 = dPtr;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ const int16 *sPtr2 = sPtr1;
+ real32 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ int32 x = (*sPtr ^ 0x8000);
+
+ *dPtr2 = scale * (real32) x;
+
+ sPtr2 += sPlaneStep;
+ dPtr2 += dPlaneStep;
+
+ }
+
+ sPtr1 += sColStep;
+ dPtr1 += dColStep;
+
+ }
+
+ sPtr += sRowStep;
+ dPtr += dRowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefCopyAreaR32_8 (const real32 *sPtr,
+ uint8 *dPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 sRowStep,
+ int32 sColStep,
+ int32 sPlaneStep,
+ int32 dRowStep,
+ int32 dColStep,
+ int32 dPlaneStep,
+ uint32 pixelRange)
+ {
+
+ real32 scale = (real32) pixelRange;
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ const real32 *sPtr1 = sPtr;
+ uint8 *dPtr1 = dPtr;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ const real32 *sPtr2 = sPtr1;
+ uint8 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ *dPtr2 = (uint8) (Pin_Overrange (*sPtr2) * scale + 0.5f);
+
+ sPtr2 += sPlaneStep;
+ dPtr2 += dPlaneStep;
+
+ }
+
+ sPtr1 += sColStep;
+ dPtr1 += dColStep;
+
+ }
+
+ sPtr += sRowStep;
+ dPtr += dRowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefCopyAreaR32_16 (const real32 *sPtr,
+ uint16 *dPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 sRowStep,
+ int32 sColStep,
+ int32 sPlaneStep,
+ int32 dRowStep,
+ int32 dColStep,
+ int32 dPlaneStep,
+ uint32 pixelRange)
+ {
+
+ real32 scale = (real32) pixelRange;
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ const real32 *sPtr1 = sPtr;
+ uint16 *dPtr1 = dPtr;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ const real32 *sPtr2 = sPtr1;
+ uint16 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ *dPtr2 = (uint16) (Pin_Overrange (*sPtr2) * scale + 0.5f);
+
+ sPtr2 += sPlaneStep;
+ dPtr2 += dPlaneStep;
+
+ }
+
+ sPtr1 += sColStep;
+ dPtr1 += dColStep;
+
+ }
+
+ sPtr += sRowStep;
+ dPtr += dRowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefCopyAreaR32_S16 (const real32 *sPtr,
+ int16 *dPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 sRowStep,
+ int32 sColStep,
+ int32 sPlaneStep,
+ int32 dRowStep,
+ int32 dColStep,
+ int32 dPlaneStep,
+ uint32 pixelRange)
+ {
+
+ real32 scale = (real32) pixelRange;
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ const real32 *sPtr1 = sPtr;
+ int16 *dPtr1 = dPtr;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ const real32 *sPtr2 = sPtr1;
+ int16 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ int32 x = (int32) (Pin_Overrange (*sPtr2) * scale + 0.5f);
+
+ *dPtr2 = (int16) (x ^ 0x8000);
+
+ sPtr2 += sPlaneStep;
+ dPtr2 += dPlaneStep;
+
+ }
+
+ sPtr1 += sColStep;
+ dPtr1 += dColStep;
+
+ }
+
+ sPtr += sRowStep;
+ dPtr += dRowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefRepeatArea8 (const uint8 *sPtr,
+ uint8 *dPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 rowStep,
+ int32 colStep,
+ int32 planeStep,
+ uint32 repeatV,
+ uint32 repeatH,
+ uint32 phaseV,
+ uint32 phaseH)
+ {
+
+ const uint8 *sPtr0 = sPtr + phaseV * rowStep +
+ phaseH * colStep;
+
+ int32 backStepV = (repeatV - 1) * rowStep;
+ int32 backStepH = (repeatH - 1) * colStep;
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ const uint8 *sPtr1 = sPtr0;
+ uint8 *dPtr1 = dPtr;
+
+ uint32 colPhase = phaseH;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ const uint8 *sPtr2 = sPtr1;
+ uint8 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ *dPtr2 = *sPtr2;
+
+ sPtr2 += planeStep;
+ dPtr2 += planeStep;
+
+ }
+
+ if (++colPhase == repeatH)
+ {
+ colPhase = 0;
+ sPtr1 -= backStepH;
+ }
+ else
+ {
+ sPtr1 += colStep;
+ }
+
+ dPtr1 += colStep;
+
+ }
+
+ if (++phaseV == repeatV)
+ {
+ phaseV = 0;
+ sPtr0 -= backStepV;
+ }
+ else
+ {
+ sPtr0 += rowStep;
+ }
+
+ dPtr += rowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefRepeatArea16 (const uint16 *sPtr,
+ uint16 *dPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 rowStep,
+ int32 colStep,
+ int32 planeStep,
+ uint32 repeatV,
+ uint32 repeatH,
+ uint32 phaseV,
+ uint32 phaseH)
+ {
+
+ const uint16 *sPtr0 = sPtr + phaseV * rowStep +
+ phaseH * colStep;
+
+ int32 backStepV = (repeatV - 1) * rowStep;
+ int32 backStepH = (repeatH - 1) * colStep;
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ const uint16 *sPtr1 = sPtr0;
+ uint16 *dPtr1 = dPtr;
+
+ uint32 colPhase = phaseH;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ const uint16 *sPtr2 = sPtr1;
+ uint16 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ *dPtr2 = *sPtr2;
+
+ sPtr2 += planeStep;
+ dPtr2 += planeStep;
+
+ }
+
+ if (++colPhase == repeatH)
+ {
+ colPhase = 0;
+ sPtr1 -= backStepH;
+ }
+ else
+ {
+ sPtr1 += colStep;
+ }
+
+ dPtr1 += colStep;
+
+ }
+
+ if (++phaseV == repeatV)
+ {
+ phaseV = 0;
+ sPtr0 -= backStepV;
+ }
+ else
+ {
+ sPtr0 += rowStep;
+ }
+
+ dPtr += rowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefRepeatArea32 (const uint32 *sPtr,
+ uint32 *dPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 rowStep,
+ int32 colStep,
+ int32 planeStep,
+ uint32 repeatV,
+ uint32 repeatH,
+ uint32 phaseV,
+ uint32 phaseH)
+ {
+
+ const uint32 *sPtr0 = sPtr + phaseV * rowStep +
+ phaseH * colStep;
+
+ int32 backStepV = (repeatV - 1) * rowStep;
+ int32 backStepH = (repeatH - 1) * colStep;
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ const uint32 *sPtr1 = sPtr0;
+ uint32 *dPtr1 = dPtr;
+
+ uint32 colPhase = phaseH;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ const uint32 *sPtr2 = sPtr1;
+ uint32 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ *dPtr2 = *sPtr2;
+
+ sPtr2 += planeStep;
+ dPtr2 += planeStep;
+
+ }
+
+ if (++colPhase == repeatH)
+ {
+ colPhase = 0;
+ sPtr1 -= backStepH;
+ }
+ else
+ {
+ sPtr1 += colStep;
+ }
+
+ dPtr1 += colStep;
+
+ }
+
+ if (++phaseV == repeatV)
+ {
+ phaseV = 0;
+ sPtr0 -= backStepV;
+ }
+ else
+ {
+ sPtr0 += rowStep;
+ }
+
+ dPtr += rowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefShiftRight16 (uint16 *dPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 rowStep,
+ int32 colStep,
+ int32 planeStep,
+ uint32 shift)
+ {
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ uint16 *dPtr1 = dPtr;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ uint16 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ *dPtr2 >>= shift;
+
+ dPtr2 += planeStep;
+
+ }
+
+ dPtr1 += colStep;
+
+ }
+
+ dPtr += rowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefBilinearRow16 (const uint16 *sPtr,
+ uint16 *dPtr,
+ uint32 cols,
+ uint32 patPhase,
+ uint32 patCount,
+ const uint32 * kernCounts,
+ const int32 * const * kernOffsets,
+ const uint16 * const * kernWeights,
+ uint32 sShift)
+ {
+
+ for (uint32 j = 0; j < cols; j++)
+ {
+
+ const uint16 *p = sPtr + (j >> sShift);
+
+ uint32 count = kernCounts [patPhase];
+
+ const int32 *offsets = kernOffsets [patPhase];
+ const uint16 *weights = kernWeights [patPhase];
+
+ if (++patPhase == patCount)
+ {
+ patPhase = 0;
+ }
+
+ uint32 total = 128;
+
+ for (uint32 k = 0; k < count; k++)
+ {
+
+ int32 offset = offsets [k];
+ uint32 weight = weights [k];
+
+ uint32 pixel = p [offset];
+
+ total += pixel * weight;
+
+ }
+
+ dPtr [j] = (uint16) (total >> 8);
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefBilinearRow32 (const real32 *sPtr,
+ real32 *dPtr,
+ uint32 cols,
+ uint32 patPhase,
+ uint32 patCount,
+ const uint32 * kernCounts,
+ const int32 * const * kernOffsets,
+ const real32 * const * kernWeights,
+ uint32 sShift)
+ {
+
+ for (uint32 j = 0; j < cols; j++)
+ {
+
+ const real32 *p = sPtr + (j >> sShift);
+
+ uint32 count = kernCounts [patPhase];
+
+ const int32 *offsets = kernOffsets [patPhase];
+ const real32 *weights = kernWeights [patPhase];
+
+ if (++patPhase == patCount)
+ {
+ patPhase = 0;
+ }
+
+ real32 total = 0.0f;
+
+ for (uint32 k = 0; k < count; k++)
+ {
+
+ int32 offset = offsets [k];
+ real32 weight = weights [k];
+
+ real32 pixel = p [offset];
+
+ total += pixel * weight;
+
+ }
+
+ dPtr [j] = total;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefBaselineABCtoRGB (const real32 *sPtrA,
+ const real32 *sPtrB,
+ const real32 *sPtrC,
+ real32 *dPtrR,
+ real32 *dPtrG,
+ real32 *dPtrB,
+ uint32 count,
+ const dng_vector &cameraWhite,
+ const dng_matrix &cameraToRGB)
+ {
+
+ real32 clipA = (real32) cameraWhite [0];
+ real32 clipB = (real32) cameraWhite [1];
+ real32 clipC = (real32) cameraWhite [2];
+
+ real32 m00 = (real32) cameraToRGB [0] [0];
+ real32 m01 = (real32) cameraToRGB [0] [1];
+ real32 m02 = (real32) cameraToRGB [0] [2];
+
+ real32 m10 = (real32) cameraToRGB [1] [0];
+ real32 m11 = (real32) cameraToRGB [1] [1];
+ real32 m12 = (real32) cameraToRGB [1] [2];
+
+ real32 m20 = (real32) cameraToRGB [2] [0];
+ real32 m21 = (real32) cameraToRGB [2] [1];
+ real32 m22 = (real32) cameraToRGB [2] [2];
+
+ for (uint32 col = 0; col < count; col++)
+ {
+
+ real32 A = sPtrA [col];
+ real32 B = sPtrB [col];
+ real32 C = sPtrC [col];
+
+ A = Min_real32 (A, clipA);
+ B = Min_real32 (B, clipB);
+ C = Min_real32 (C, clipC);
+
+ real32 r = m00 * A + m01 * B + m02 * C;
+ real32 g = m10 * A + m11 * B + m12 * C;
+ real32 b = m20 * A + m21 * B + m22 * C;
+
+ r = Pin_real32 (0.0f, r, 1.0f);
+ g = Pin_real32 (0.0f, g, 1.0f);
+ b = Pin_real32 (0.0f, b, 1.0f);
+
+ dPtrR [col] = r;
+ dPtrG [col] = g;
+ dPtrB [col] = b;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefBaselineABCDtoRGB (const real32 *sPtrA,
+ const real32 *sPtrB,
+ const real32 *sPtrC,
+ const real32 *sPtrD,
+ real32 *dPtrR,
+ real32 *dPtrG,
+ real32 *dPtrB,
+ uint32 count,
+ const dng_vector &cameraWhite,
+ const dng_matrix &cameraToRGB)
+ {
+
+ real32 clipA = (real32) cameraWhite [0];
+ real32 clipB = (real32) cameraWhite [1];
+ real32 clipC = (real32) cameraWhite [2];
+ real32 clipD = (real32) cameraWhite [3];
+
+ real32 m00 = (real32) cameraToRGB [0] [0];
+ real32 m01 = (real32) cameraToRGB [0] [1];
+ real32 m02 = (real32) cameraToRGB [0] [2];
+ real32 m03 = (real32) cameraToRGB [0] [3];
+
+ real32 m10 = (real32) cameraToRGB [1] [0];
+ real32 m11 = (real32) cameraToRGB [1] [1];
+ real32 m12 = (real32) cameraToRGB [1] [2];
+ real32 m13 = (real32) cameraToRGB [1] [3];
+
+ real32 m20 = (real32) cameraToRGB [2] [0];
+ real32 m21 = (real32) cameraToRGB [2] [1];
+ real32 m22 = (real32) cameraToRGB [2] [2];
+ real32 m23 = (real32) cameraToRGB [2] [3];
+
+ for (uint32 col = 0; col < count; col++)
+ {
+
+ real32 A = sPtrA [col];
+ real32 B = sPtrB [col];
+ real32 C = sPtrC [col];
+ real32 D = sPtrD [col];
+
+ A = Min_real32 (A, clipA);
+ B = Min_real32 (B, clipB);
+ C = Min_real32 (C, clipC);
+ D = Min_real32 (D, clipD);
+
+ real32 r = m00 * A + m01 * B + m02 * C + m03 * D;
+ real32 g = m10 * A + m11 * B + m12 * C + m13 * D;
+ real32 b = m20 * A + m21 * B + m22 * C + m23 * D;
+
+ r = Pin_real32 (0.0f, r, 1.0f);
+ g = Pin_real32 (0.0f, g, 1.0f);
+ b = Pin_real32 (0.0f, b, 1.0f);
+
+ dPtrR [col] = r;
+ dPtrG [col] = g;
+ dPtrB [col] = b;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefBaselineHueSatMap (const real32 *sPtrR,
+ const real32 *sPtrG,
+ const real32 *sPtrB,
+ real32 *dPtrR,
+ real32 *dPtrG,
+ real32 *dPtrB,
+ uint32 count,
+ const dng_hue_sat_map &lut,
+ const dng_1d_table *encodeTable,
+ const dng_1d_table *decodeTable)
+ {
+
+ uint32 hueDivisions;
+ uint32 satDivisions;
+ uint32 valDivisions;
+
+ lut.GetDivisions (hueDivisions,
+ satDivisions,
+ valDivisions);
+
+ real32 hScale = (hueDivisions < 2) ? 0.0f : (hueDivisions * (1.0f / 6.0f));
+ real32 sScale = (real32) (satDivisions - 1);
+ real32 vScale = (real32) (valDivisions - 1);
+
+ int32 maxHueIndex0 = hueDivisions - 1;
+ int32 maxSatIndex0 = satDivisions - 2;
+ int32 maxValIndex0 = valDivisions - 2;
+
+ const bool hasEncodeTable = ((encodeTable != NULL) && (encodeTable->Table () != NULL));
+ const bool hasDecodeTable = ((decodeTable != NULL) && (decodeTable->Table () != NULL));
+
+ const bool hasTable = hasEncodeTable && hasDecodeTable;
+
+ const dng_hue_sat_map::HSBModify *tableBase = lut.GetConstDeltas ();
+
+ int32 hueStep = satDivisions;
+ int32 valStep = hueDivisions * hueStep;
+
+ #if 0 // Not required with "2.5D" table optimization.
+
+ if (valDivisions < 2)
+ {
+ valStep = 0;
+ maxValIndex0 = 0;
+ }
+
+ #endif
+
+ for (uint32 j = 0; j < count; j++)
+ {
+
+ real32 r = sPtrR [j];
+ real32 g = sPtrG [j];
+ real32 b = sPtrB [j];
+
+ real32 h, s, v;
+
+ DNG_RGBtoHSV (r, g, b, h, s, v);
+
+ real32 vEncoded = v;
+
+ real32 hueShift;
+ real32 satScale;
+ real32 valScale;
+
+ if (valDivisions < 2) // Optimize most common case of "2.5D" table.
+ {
+
+ real32 hScaled = h * hScale;
+ real32 sScaled = s * sScale;
+
+ int32 hIndex0 = (int32) hScaled;
+ int32 sIndex0 = (int32) sScaled;
+
+ sIndex0 = Min_int32 (sIndex0, maxSatIndex0);
+
+ int32 hIndex1 = hIndex0 + 1;
+
+ if (hIndex0 >= maxHueIndex0)
+ {
+ hIndex0 = maxHueIndex0;
+ hIndex1 = 0;
+ }
+
+ real32 hFract1 = hScaled - (real32) hIndex0;
+ real32 sFract1 = sScaled - (real32) sIndex0;
+
+ real32 hFract0 = 1.0f - hFract1;
+ real32 sFract0 = 1.0f - sFract1;
+
+ const dng_hue_sat_map::HSBModify *entry00 = tableBase + hIndex0 * hueStep +
+ sIndex0;
+
+ const dng_hue_sat_map::HSBModify *entry01 = entry00 + (hIndex1 - hIndex0) * hueStep;
+
+ real32 hueShift0 = hFract0 * entry00->fHueShift +
+ hFract1 * entry01->fHueShift;
+
+ real32 satScale0 = hFract0 * entry00->fSatScale +
+ hFract1 * entry01->fSatScale;
+
+ real32 valScale0 = hFract0 * entry00->fValScale +
+ hFract1 * entry01->fValScale;
+
+ entry00++;
+ entry01++;
+
+ real32 hueShift1 = hFract0 * entry00->fHueShift +
+ hFract1 * entry01->fHueShift;
+
+ real32 satScale1 = hFract0 * entry00->fSatScale +
+ hFract1 * entry01->fSatScale;
+
+ real32 valScale1 = hFract0 * entry00->fValScale +
+ hFract1 * entry01->fValScale;
+
+ hueShift = sFract0 * hueShift0 + sFract1 * hueShift1;
+ satScale = sFract0 * satScale0 + sFract1 * satScale1;
+ valScale = sFract0 * valScale0 + sFract1 * valScale1;
+
+ }
+
+ else
+ {
+
+ if (hasTable)
+ {
+ vEncoded = encodeTable->Interpolate (Pin_real32 (v));
+ }
+
+ real32 hScaled = h * hScale;
+ real32 sScaled = s * sScale;
+ real32 vScaled = vEncoded * vScale;
+
+ int32 hIndex0 = (int32) hScaled;
+ int32 sIndex0 = (int32) sScaled;
+ int32 vIndex0 = (int32) vScaled;
+
+ sIndex0 = Min_int32 (sIndex0, maxSatIndex0);
+ vIndex0 = Min_int32 (vIndex0, maxValIndex0);
+
+ int32 hIndex1 = hIndex0 + 1;
+
+ if (hIndex0 >= maxHueIndex0)
+ {
+ hIndex0 = maxHueIndex0;
+ hIndex1 = 0;
+ }
+
+ real32 hFract1 = hScaled - (real32) hIndex0;
+ real32 sFract1 = sScaled - (real32) sIndex0;
+ real32 vFract1 = vScaled - (real32) vIndex0;
+
+ real32 hFract0 = 1.0f - hFract1;
+ real32 sFract0 = 1.0f - sFract1;
+ real32 vFract0 = 1.0f - vFract1;
+
+ const dng_hue_sat_map::HSBModify *entry00 = tableBase + vIndex0 * valStep +
+ hIndex0 * hueStep +
+ sIndex0;
+
+ const dng_hue_sat_map::HSBModify *entry01 = entry00 + (hIndex1 - hIndex0) * hueStep;
+
+ const dng_hue_sat_map::HSBModify *entry10 = entry00 + valStep;
+ const dng_hue_sat_map::HSBModify *entry11 = entry01 + valStep;
+
+ real32 hueShift0 = vFract0 * (hFract0 * entry00->fHueShift +
+ hFract1 * entry01->fHueShift) +
+ vFract1 * (hFract0 * entry10->fHueShift +
+ hFract1 * entry11->fHueShift);
+
+ real32 satScale0 = vFract0 * (hFract0 * entry00->fSatScale +
+ hFract1 * entry01->fSatScale) +
+ vFract1 * (hFract0 * entry10->fSatScale +
+ hFract1 * entry11->fSatScale);
+
+ real32 valScale0 = vFract0 * (hFract0 * entry00->fValScale +
+ hFract1 * entry01->fValScale) +
+ vFract1 * (hFract0 * entry10->fValScale +
+ hFract1 * entry11->fValScale);
+
+ entry00++;
+ entry01++;
+ entry10++;
+ entry11++;
+
+ real32 hueShift1 = vFract0 * (hFract0 * entry00->fHueShift +
+ hFract1 * entry01->fHueShift) +
+ vFract1 * (hFract0 * entry10->fHueShift +
+ hFract1 * entry11->fHueShift);
+
+ real32 satScale1 = vFract0 * (hFract0 * entry00->fSatScale +
+ hFract1 * entry01->fSatScale) +
+ vFract1 * (hFract0 * entry10->fSatScale +
+ hFract1 * entry11->fSatScale);
+
+ real32 valScale1 = vFract0 * (hFract0 * entry00->fValScale +
+ hFract1 * entry01->fValScale) +
+ vFract1 * (hFract0 * entry10->fValScale +
+ hFract1 * entry11->fValScale);
+
+ hueShift = sFract0 * hueShift0 + sFract1 * hueShift1;
+ satScale = sFract0 * satScale0 + sFract1 * satScale1;
+ valScale = sFract0 * valScale0 + sFract1 * valScale1;
+
+ }
+
+ hueShift *= (6.0f / 360.0f); // Convert to internal hue range.
+
+ h += hueShift;
+
+ s = Min_real32 (s * satScale, 1.0f);
+
+ vEncoded = Pin_real32 (vEncoded * valScale);
+
+ v = hasTable ? decodeTable->Interpolate (vEncoded) : vEncoded;
+
+ DNG_HSVtoRGB (h, s, v, r, g, b);
+
+ dPtrR [j] = r;
+ dPtrG [j] = g;
+ dPtrB [j] = b;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefBaselineRGBtoGray (const real32 *sPtrR,
+ const real32 *sPtrG,
+ const real32 *sPtrB,
+ real32 *dPtrG,
+ uint32 count,
+ const dng_matrix &matrix)
+ {
+
+ real32 m00 = (real32) matrix [0] [0];
+ real32 m01 = (real32) matrix [0] [1];
+ real32 m02 = (real32) matrix [0] [2];
+
+ for (uint32 col = 0; col < count; col++)
+ {
+
+ real32 R = sPtrR [col];
+ real32 G = sPtrG [col];
+ real32 B = sPtrB [col];
+
+ real32 g = m00 * R + m01 * G + m02 * B;
+
+ g = Pin_real32 (0.0f, g, 1.0f);
+
+ dPtrG [col] = g;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefBaselineRGBtoRGB (const real32 *sPtrR,
+ const real32 *sPtrG,
+ const real32 *sPtrB,
+ real32 *dPtrR,
+ real32 *dPtrG,
+ real32 *dPtrB,
+ uint32 count,
+ const dng_matrix &matrix)
+ {
+
+ real32 m00 = (real32) matrix [0] [0];
+ real32 m01 = (real32) matrix [0] [1];
+ real32 m02 = (real32) matrix [0] [2];
+
+ real32 m10 = (real32) matrix [1] [0];
+ real32 m11 = (real32) matrix [1] [1];
+ real32 m12 = (real32) matrix [1] [2];
+
+ real32 m20 = (real32) matrix [2] [0];
+ real32 m21 = (real32) matrix [2] [1];
+ real32 m22 = (real32) matrix [2] [2];
+
+ for (uint32 col = 0; col < count; col++)
+ {
+
+ real32 R = sPtrR [col];
+ real32 G = sPtrG [col];
+ real32 B = sPtrB [col];
+
+ real32 r = m00 * R + m01 * G + m02 * B;
+ real32 g = m10 * R + m11 * G + m12 * B;
+ real32 b = m20 * R + m21 * G + m22 * B;
+
+ r = Pin_real32 (0.0f, r, 1.0f);
+ g = Pin_real32 (0.0f, g, 1.0f);
+ b = Pin_real32 (0.0f, b, 1.0f);
+
+ dPtrR [col] = r;
+ dPtrG [col] = g;
+ dPtrB [col] = b;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefBaseline1DTable (const real32 *sPtr,
+ real32 *dPtr,
+ uint32 count,
+ const dng_1d_table &table)
+ {
+
+ for (uint32 col = 0; col < count; col++)
+ {
+
+ real32 x = sPtr [col];
+
+ real32 y = table.Interpolate (x);
+
+ dPtr [col] = y;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefBaselineRGBTone (const real32 *sPtrR,
+ const real32 *sPtrG,
+ const real32 *sPtrB,
+ real32 *dPtrR,
+ real32 *dPtrG,
+ real32 *dPtrB,
+ uint32 count,
+ const dng_1d_table &table)
+ {
+
+ for (uint32 col = 0; col < count; col++)
+ {
+
+ real32 r = sPtrR [col];
+ real32 g = sPtrG [col];
+ real32 b = sPtrB [col];
+
+ real32 rr;
+ real32 gg;
+ real32 bb;
+
+ #define RGBTone(r, g, b, rr, gg, bb)\
+ {\
+ \
+ DNG_ASSERT (r >= g && g >= b && r > b, "Logic Error RGBTone");\
+ \
+ rr = table.Interpolate (r);\
+ bb = table.Interpolate (b);\
+ \
+ gg = bb + ((rr - bb) * (g - b) / (r - b));\
+ \
+ }
+
+ if (r >= g)
+ {
+
+ if (g > b)
+ {
+
+ // Case 1: r >= g > b
+
+ RGBTone (r, g, b, rr, gg, bb);
+
+ }
+
+ else if (b > r)
+ {
+
+ // Case 2: b > r >= g
+
+ RGBTone (b, r, g, bb, rr, gg);
+
+ }
+
+ else if (b > g)
+ {
+
+ // Case 3: r >= b > g
+
+ RGBTone (r, b, g, rr, bb, gg);
+
+ }
+
+ else
+ {
+
+ // Case 4: r >= g == b
+
+ DNG_ASSERT (r >= g && g == b, "Logic Error 2");
+
+ rr = table.Interpolate (r);
+ gg = table.Interpolate (g);
+ bb = gg;
+
+ }
+
+ }
+
+ else
+ {
+
+ if (r >= b)
+ {
+
+ // Case 5: g > r >= b
+
+ RGBTone (g, r, b, gg, rr, bb);
+
+ }
+
+ else if (b > g)
+ {
+
+ // Case 6: b > g > r
+
+ RGBTone (b, g, r, bb, gg, rr);
+
+ }
+
+ else
+ {
+
+ // Case 7: g >= b > r
+
+ RGBTone (g, b, r, gg, bb, rr);
+
+ }
+
+ }
+
+ #undef RGBTone
+
+ dPtrR [col] = rr;
+ dPtrG [col] = gg;
+ dPtrB [col] = bb;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefResampleDown16 (const uint16 *sPtr,
+ uint16 *dPtr,
+ uint32 sCount,
+ int32 sRowStep,
+ const int16 *wPtr,
+ uint32 wCount,
+ uint32 pixelRange)
+ {
+
+ for (uint32 j = 0; j < sCount; j++)
+ {
+
+ int32 total = 8192;
+
+ const uint16 *s = sPtr + j;
+
+ for (uint32 k = 0; k < wCount; k++)
+ {
+
+ total += wPtr [k] * (int32) s [0];
+
+ s += sRowStep;
+
+ }
+
+ dPtr [j] = (uint16) Pin_int32 (0,
+ total >> 14,
+ pixelRange);
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefResampleDown32 (const real32 *sPtr,
+ real32 *dPtr,
+ uint32 sCount,
+ int32 sRowStep,
+ const real32 *wPtr,
+ uint32 wCount)
+ {
+
+ uint32 col;
+
+ // Process first row.
+
+ real32 w = wPtr [0];
+
+ for (col = 0; col < sCount; col++)
+ {
+
+ dPtr [col] = w * sPtr [col];
+
+ }
+
+ sPtr += sRowStep;
+
+ // Process middle rows.
+
+ for (uint32 j = 1; j < wCount - 1; j++)
+ {
+
+ w = wPtr [j];
+
+ for (col = 0; col < sCount; col++)
+ {
+
+ dPtr [col] += w * sPtr [col];
+
+ }
+
+ sPtr += sRowStep;
+
+ }
+
+ // Process last row.
+
+ w = wPtr [wCount - 1];
+
+ for (col = 0; col < sCount; col++)
+ {
+
+ dPtr [col] = Pin_real32 (0.0f,
+ dPtr [col] + w * sPtr [col],
+ 1.0f);
+
+ }
+
+ }
+
+/******************************************************************************/
+
+void RefResampleAcross16 (const uint16 *sPtr,
+ uint16 *dPtr,
+ uint32 dCount,
+ const int32 *coord,
+ const int16 *wPtr,
+ uint32 wCount,
+ uint32 wStep,
+ uint32 pixelRange)
+ {
+
+ for (uint32 j = 0; j < dCount; j++)
+ {
+
+ int32 sCoord = coord [j];
+
+ int32 sFract = sCoord & kResampleSubsampleMask;
+ int32 sPixel = sCoord >> kResampleSubsampleBits;
+
+ const int16 *w = wPtr + sFract * wStep;
+ const uint16 *s = sPtr + sPixel;
+
+ int32 total = w [0] * (int32) s [0];
+
+ for (uint32 k = 1; k < wCount; k++)
+ {
+
+ total += w [k] * (int32) s [k];
+
+ }
+
+ dPtr [j] = (uint16) Pin_int32 (0,
+ (total + 8192) >> 14,
+ pixelRange);
+
+ }
+
+ }
+
+/******************************************************************************/
+
+void RefResampleAcross32 (const real32 *sPtr,
+ real32 *dPtr,
+ uint32 dCount,
+ const int32 *coord,
+ const real32 *wPtr,
+ uint32 wCount,
+ uint32 wStep)
+ {
+
+ for (uint32 j = 0; j < dCount; j++)
+ {
+
+ int32 sCoord = coord [j];
+
+ int32 sFract = sCoord & kResampleSubsampleMask;
+ int32 sPixel = sCoord >> kResampleSubsampleBits;
+
+ const real32 *w = wPtr + sFract * wStep;
+ const real32 *s = sPtr + sPixel;
+
+ real32 total = w [0] * s [0];
+
+ for (uint32 k = 1; k < wCount; k++)
+ {
+
+ total += w [k] * s [k];
+
+ }
+
+ dPtr [j] = Pin_real32 (0.0f, total, 1.0f);
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+bool RefEqualBytes (const void *sPtr,
+ const void *dPtr,
+ uint32 count)
+ {
+
+ return memcmp (dPtr, sPtr, count) == 0;
+
+ }
+
+/*****************************************************************************/
+
+bool RefEqualArea8 (const uint8 *sPtr,
+ const uint8 *dPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 sRowStep,
+ int32 sColStep,
+ int32 sPlaneStep,
+ int32 dRowStep,
+ int32 dColStep,
+ int32 dPlaneStep)
+ {
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ const uint8 *sPtr1 = sPtr;
+ const uint8 *dPtr1 = dPtr;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ const uint8 *sPtr2 = sPtr1;
+ const uint8 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ if (*dPtr2 != *sPtr2)
+ return false;
+
+ sPtr2 += sPlaneStep;
+ dPtr2 += dPlaneStep;
+
+ }
+
+ sPtr1 += sColStep;
+ dPtr1 += dColStep;
+
+ }
+
+ sPtr += sRowStep;
+ dPtr += dRowStep;
+
+ }
+
+ return true;
+
+ }
+
+/*****************************************************************************/
+
+bool RefEqualArea16 (const uint16 *sPtr,
+ const uint16 *dPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 sRowStep,
+ int32 sColStep,
+ int32 sPlaneStep,
+ int32 dRowStep,
+ int32 dColStep,
+ int32 dPlaneStep)
+ {
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ const uint16 *sPtr1 = sPtr;
+ const uint16 *dPtr1 = dPtr;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ const uint16 *sPtr2 = sPtr1;
+ const uint16 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ if (*dPtr2 != *sPtr2)
+ return false;
+
+ sPtr2 += sPlaneStep;
+ dPtr2 += dPlaneStep;
+
+ }
+
+ sPtr1 += sColStep;
+ dPtr1 += dColStep;
+
+ }
+
+ sPtr += sRowStep;
+ dPtr += dRowStep;
+
+ }
+
+ return true;
+
+ }
+
+/*****************************************************************************/
+
+bool RefEqualArea32 (const uint32 *sPtr,
+ const uint32 *dPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 sRowStep,
+ int32 sColStep,
+ int32 sPlaneStep,
+ int32 dRowStep,
+ int32 dColStep,
+ int32 dPlaneStep)
+ {
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ const uint32 *sPtr1 = sPtr;
+ const uint32 *dPtr1 = dPtr;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ const uint32 *sPtr2 = sPtr1;
+ const uint32 *dPtr2 = dPtr1;
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ if (*dPtr2 != *sPtr2)
+ return false;
+
+ sPtr2 += sPlaneStep;
+ dPtr2 += dPlaneStep;
+
+ }
+
+ sPtr1 += sColStep;
+ dPtr1 += dColStep;
+
+ }
+
+ sPtr += sRowStep;
+ dPtr += dRowStep;
+
+ }
+
+ return true;
+
+ }
+
+/*****************************************************************************/
+
+void RefVignetteMask16 (uint16 *mPtr,
+ uint32 rows,
+ uint32 cols,
+ int32 rowStep,
+ int64 offsetH,
+ int64 offsetV,
+ int64 stepH,
+ int64 stepV,
+ uint32 tBits,
+ const uint16 *table)
+ {
+
+ uint32 tShift = 32 - tBits;
+ uint32 tRound = (1 << (tShift - 1));
+ uint32 tLimit = 1 << tBits;
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ int64 baseDelta = (offsetV + 32768) >> 16;
+
+ baseDelta = baseDelta * baseDelta + tRound;
+
+ int64 deltaH = offsetH + 32768;
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ int64 temp = deltaH >> 16;
+
+ int64 delta = baseDelta + (temp * temp);
+
+ uint32 index = Min_uint32 ((uint32) (delta >> tShift), tLimit);
+
+ mPtr [col] = table [index];
+
+ deltaH += stepH;
+
+ }
+
+ offsetV += stepV;
+
+ mPtr += rowStep;
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefVignette16 (int16 *sPtr,
+ const uint16 *mPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 sRowStep,
+ int32 sPlaneStep,
+ int32 mRowStep,
+ uint32 mBits)
+ {
+
+ const uint32 mRound = 1 << (mBits - 1);
+
+ switch (planes)
+ {
+
+ case 1:
+ {
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ uint32 s = sPtr [col] + 32768;
+
+ uint32 m = mPtr [col];
+
+ s = (s * m + mRound) >> mBits;
+
+ s = Min_uint32 (s, 65535);
+
+ sPtr [col] = (int16) (s - 32768);
+
+ }
+
+ sPtr += sRowStep;
+
+ mPtr += mRowStep;
+
+ }
+
+ break;
+
+ }
+
+ case 3:
+ {
+
+ int16 *rPtr = sPtr;
+ int16 *gPtr = rPtr + sPlaneStep;
+ int16 *bPtr = gPtr + sPlaneStep;
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ uint32 r = rPtr [col] + 32768;
+ uint32 g = gPtr [col] + 32768;
+ uint32 b = bPtr [col] + 32768;
+
+ uint32 m = mPtr [col];
+
+ r = (r * m + mRound) >> mBits;
+ g = (g * m + mRound) >> mBits;
+ b = (b * m + mRound) >> mBits;
+
+ r = Min_uint32 (r, 65535);
+ g = Min_uint32 (g, 65535);
+ b = Min_uint32 (b, 65535);
+
+ rPtr [col] = (int16) (r - 32768);
+ gPtr [col] = (int16) (g - 32768);
+ bPtr [col] = (int16) (b - 32768);
+
+ }
+
+ rPtr += sRowStep;
+ gPtr += sRowStep;
+ bPtr += sRowStep;
+
+ mPtr += mRowStep;
+
+ }
+
+ break;
+
+ }
+
+ case 4:
+ {
+
+ int16 *aPtr = sPtr;
+ int16 *bPtr = aPtr + sPlaneStep;
+ int16 *cPtr = bPtr + sPlaneStep;
+ int16 *dPtr = cPtr + sPlaneStep;
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ uint32 a = aPtr [col] + 32768;
+ uint32 b = bPtr [col] + 32768;
+ uint32 c = cPtr [col] + 32768;
+ uint32 d = dPtr [col] + 32768;
+
+ uint32 m = mPtr [col];
+
+ a = (a * m + mRound) >> mBits;
+ b = (b * m + mRound) >> mBits;
+ c = (c * m + mRound) >> mBits;
+ d = (d * m + mRound) >> mBits;
+
+ a = Min_uint32 (a, 65535);
+ b = Min_uint32 (b, 65535);
+ c = Min_uint32 (c, 65535);
+ d = Min_uint32 (d, 65535);
+
+ aPtr [col] = (int16) (a - 32768);
+ bPtr [col] = (int16) (b - 32768);
+ cPtr [col] = (int16) (c - 32768);
+ dPtr [col] = (int16) (d - 32768);
+
+ }
+
+ aPtr += sRowStep;
+ bPtr += sRowStep;
+ cPtr += sRowStep;
+ dPtr += sRowStep;
+
+ mPtr += mRowStep;
+
+ }
+
+ break;
+
+ }
+
+ default:
+ {
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ int16 *planePtr = sPtr;
+
+ const uint16 *maskPtr = mPtr;
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ uint32 s = planePtr [col] + 32768;
+
+ uint32 m = maskPtr [col];
+
+ s = (s * m + mRound) >> mBits;
+
+ s = Min_uint32 (s, 65535);
+
+ planePtr [col] = (int16) (s - 32768);
+
+ }
+
+ planePtr += sRowStep;
+
+ maskPtr += mRowStep;
+
+ }
+
+ sPtr += sPlaneStep;
+
+ }
+
+ break;
+
+ }
+
+ }
+
+ }
+
+/*****************************************************************************/
+
+void RefVignette32 (real32 *sPtr,
+ const uint16 *mPtr,
+ uint32 rows,
+ uint32 cols,
+ uint32 planes,
+ int32 sRowStep,
+ int32 sPlaneStep,
+ int32 mRowStep,
+ uint32 mBits)
+ {
+
+ const real32 kNorm = 1.0f / (1 << mBits);
+
+ switch (planes)
+ {
+
+ case 1:
+ {
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ real32 s = sPtr [col];
+
+ uint16 m = mPtr [col];
+
+ real32 scale = m * kNorm;
+
+ s = Min_real32 (s * scale, 1.0f);
+
+ sPtr [col] = s;
+
+ }
+
+ sPtr += sRowStep;
+
+ mPtr += mRowStep;
+
+ }
+
+ break;
+
+ }
+
+ case 3:
+ {
+
+ real32 *rPtr = sPtr;
+ real32 *gPtr = rPtr + sPlaneStep;
+ real32 *bPtr = gPtr + sPlaneStep;
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ real32 r = rPtr [col];
+ real32 g = gPtr [col];
+ real32 b = bPtr [col];
+
+ uint16 m = mPtr [col];
+
+ real32 scale = m * kNorm;
+
+ r = Min_real32 (r * scale, 1.0f);
+ g = Min_real32 (g * scale, 1.0f);
+ b = Min_real32 (b * scale, 1.0f);
+
+ rPtr [col] = r;
+ gPtr [col] = g;
+ bPtr [col] = b;
+
+ }
+
+ rPtr += sRowStep;
+ gPtr += sRowStep;
+ bPtr += sRowStep;
+
+ mPtr += mRowStep;
+
+ }
+
+ break;
+
+ }
+
+ case 4:
+ {
+
+ real32 *aPtr = sPtr;
+ real32 *bPtr = aPtr + sPlaneStep;
+ real32 *cPtr = bPtr + sPlaneStep;
+ real32 *dPtr = cPtr + sPlaneStep;
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ real32 a = aPtr [col];
+ real32 b = bPtr [col];
+ real32 c = cPtr [col];
+ real32 d = dPtr [col];
+
+ uint16 m = mPtr [col];
+
+ real32 scale = m * kNorm;
+
+ a = Min_real32 (a * scale, 1.0f);
+ b = Min_real32 (b * scale, 1.0f);
+ c = Min_real32 (c * scale, 1.0f);
+ d = Min_real32 (d * scale, 1.0f);
+
+ aPtr [col] = a;
+ bPtr [col] = b;
+ cPtr [col] = c;
+ dPtr [col] = d;
+
+ }
+
+ aPtr += sRowStep;
+ bPtr += sRowStep;
+ cPtr += sRowStep;
+ dPtr += sRowStep;
+
+ mPtr += mRowStep;
+
+ }
+
+ break;
+
+ }
+
+ default:
+ {
+
+ for (uint32 plane = 0; plane < planes; plane++)
+ {
+
+ real32 *planePtr = sPtr;
+
+ const uint16 *maskPtr = mPtr;
+
+ for (uint32 row = 0; row < rows; row++)
+ {
+
+ for (uint32 col = 0; col < cols; col++)
+ {
+
+ real32 s = planePtr [col];
+
+ uint16 m = maskPtr [col];
+
+ real32 scale = m * kNorm;
+
+ s = Min_real32 (s * scale, 1.0f);
+
+ planePtr [col] = s;
+
+ }
+
+ planePtr += sRowStep;
+
+ maskPtr += mRowStep;
+
+ }
+
+ sPtr += sPlaneStep;
+
+ }
+
+ break;
+
+ }
+
+ }
+
+ }
+
+/******************************************************************************/
+
+void RefMapArea16 (uint16 *dPtr,
+ uint32 count0,
+ uint32 count1,
+ uint32 count2,
+ int32 step0,
+ int32 step1,
+ int32 step2,
+ const uint16 *map)
+ {
+
+ if (step2 == 1 && count2 >= 32)
+ {
+
+ for (uint32 index0 = 0; index0 < count0; index0++)
+ {
+
+ uint16 *d1 = dPtr;
+
+ for (uint32 index1 = 0; index1 < count1; index1++)
+ {
+
+ uint16 *d2 = d1;
+
+ uint32 count = count2;
+
+ // Get the data 32-bit aligned if it is not.
+
+ if (!IsAligned32 (dPtr))
+ {
+
+ d2 [0] = map [d2 [0]];
+
+ count--;
+
+ d2++;
+
+ }
+
+ // Use 32-bit reads and writes for bulk processing.
+
+ uint32 *dPtr32 = (uint32 *) d2;
+
+ // Process in blocks of 16 pixels.
+
+ uint32 blocks = count >> 4;
+
+ count -= blocks << 4;
+ d2 += blocks << 4;
+
+ while (blocks--)
+ {
+
+ uint32 x0, x1, x2, x3, x4, x5, x6, x7;
+ uint32 p0, p1, p2, p3, p4, p5, p6, p7;
+
+ // Use 32 bit reads & writes, and pack and unpack the 16-bit values.
+ // This results in slightly higher performance.
+
+ // Note that this code runs on both little-endian and big-endian systems,
+ // since the pixels are either never swapped or double swapped.
+
+ x0 = dPtr32 [0];
+ x1 = dPtr32 [1];
+ x2 = dPtr32 [2];
+ x3 = dPtr32 [3];
+
+ p0 = map [x0 >> 16 ];
+ p1 = map [x0 & 0x0FFFF];
+ p2 = map [x1 >> 16 ];
+ p3 = map [x1 & 0x0FFFF];
+ p4 = map [x2 >> 16 ];
+ p5 = map [x2 & 0x0FFFF];
+ p6 = map [x3 >> 16 ];
+ p7 = map [x3 & 0x0FFFF];
+
+ x0 = (p0 << 16) | p1;
+ x1 = (p2 << 16) | p3;
+ x2 = (p4 << 16) | p5;
+ x3 = (p6 << 16) | p7;
+
+ x4 = dPtr32 [4];
+ x5 = dPtr32 [5];
+ x6 = dPtr32 [6];
+ x7 = dPtr32 [7];
+
+ dPtr32 [0] = x0;
+ dPtr32 [1] = x1;
+ dPtr32 [2] = x2;
+ dPtr32 [3] = x3;
+
+ p0 = map [x4 >> 16 ];
+ p1 = map [x4 & 0x0FFFF];
+ p2 = map [x5 >> 16 ];
+ p3 = map [x5 & 0x0FFFF];
+ p4 = map [x6 >> 16 ];
+ p5 = map [x6 & 0x0FFFF];
+ p6 = map [x7 >> 16 ];
+ p7 = map [x7 & 0x0FFFF];
+
+ x4 = (p0 << 16) | p1;
+ x5 = (p2 << 16) | p3;
+ x6 = (p4 << 16) | p5;
+ x7 = (p6 << 16) | p7;
+
+ dPtr32 [4] = x4;
+ dPtr32 [5] = x5;
+ dPtr32 [6] = x6;
+ dPtr32 [7] = x7;
+
+ dPtr32 += 8;
+
+ }
+
+ // Process remaining columns.
+
+ for (uint32 j = 0; j < count; j++)
+ {
+
+ d2 [j] = map [d2 [j]];
+
+ }
+
+ d1 += step1;
+
+ }
+
+ dPtr += step0;
+
+ }
+
+ }
+
+ else
+ {
+
+ for (uint32 index0 = 0; index0 < count0; index0++)
+ {
+
+ uint16 *d1 = dPtr;
+
+ for (uint32 index1 = 0; index1 < count1; index1++)
+ {
+
+ uint16 *d2 = d1;
+
+ for (uint32 index2 = 0; index2 < count2; index2++)
+ {
+
+ d2 [0] = map [d2 [0]];
+
+ d2 += step2;
+
+ }
+
+ d1 += step1;
+
+ }
+
+ dPtr += step0;
+
+ }
+
+ }
+
+ }
+
+/*****************************************************************************/
|