Merge remote-tracking branch 'qatar/master'

* qatar/master:
  FATE: add tests for targa
  ARM: fix Thumb-mode simple_idct_arm
  ARM: 4-byte align start of all asm functions
  rgb2rgb: rgb12to15()
  swscale-test: fix stack overread.
  swscale: fix invalid conversions and memory problems.
  cabac: split cabac.h into declarations and function definitions
  cabac: Mark ff_h264_mps_state array as static, it is only used within cabac.c.
  cabac: Remove ff_h264_lps_state array.

Conflicts:
	libswscale/rgb2rgb.h
	libswscale/swscale_unscaled.c
	tests/fate/image.mak

Merged-by: Michael Niedermayer <michaelni@gmx.at>

4 files changed, 53 insertions, 26 deletions

diff --git a/libswscale/rgb2rgb.c b/libswscale/rgb2rgb.c
index 31ebc20a68..96559202b9 100644
--- a/libswscale/rgb2rgb.c
+++ b/libswscale/rgb2rgb.c
@@ -183,6 +183,25 @@ void rgb16tobgr32(const uint8_t *src, uint8_t *dst, int src_size)
     }
 }
 
+void rgb12to15(const uint8_t *src, uint8_t *dst, int src_size)
+{
+    const uint16_t *end;
+    uint16_t *d = (uint16_t *)dst;
+    const uint16_t *s = (const uint16_t *)src;
+    uint16_t rgb, r, g, b;
+    end = s + src_size / 2;
+    while (s < end) {
+        rgb = *s++;
+        r = rgb & 0xF00;
+        g = rgb & 0x0F0;
+        b = rgb & 0x00F;
+        r = (r << 3) | ((r & 0x800) >> 1);
+        g = (g << 2) | ((g & 0x080) >> 2);
+        b = (b << 1) | ( b          >> 3);
+        *d++ = r | g | b;
+    }
+}
+
 void rgb16to24(const uint8_t *src, uint8_t *dst, int src_size)
 {
     const uint16_t *end;
diff --git a/libswscale/rgb2rgb.h b/libswscale/rgb2rgb.h
index 4e63124977..c4479862ef 100644
--- a/libswscale/rgb2rgb.h
+++ b/libswscale/rgb2rgb.h
@@ -63,6 +63,7 @@ void    rgb15to24(const uint8_t *src, uint8_t *dst, int src_size);
 void rgb15tobgr16(const uint8_t *src, uint8_t *dst, int src_size);
 void rgb15tobgr15(const uint8_t *src, uint8_t *dst, int src_size);
 void rgb12tobgr12(const uint8_t *src, uint8_t *dst, int src_size);
+void    rgb12to15(const uint8_t *src, uint8_t *dst, int src_size);
 
 void shuffle_bytes_0321(const uint8_t *src, uint8_t *dst, int src_size);
 void shuffle_bytes_1230(const uint8_t *src, uint8_t *dst, int src_size);
diff --git a/libswscale/swscale-test.c b/libswscale/swscale-test.c
index f382ce3d6a..ef6c55ce02 100644
--- a/libswscale/swscale-test.c
+++ b/libswscale/swscale-test.c
@@ -340,8 +340,8 @@ int main(int argc, char **argv)
     enum PixelFormat srcFormat = PIX_FMT_NONE;
     enum PixelFormat dstFormat = PIX_FMT_NONE;
     uint8_t *rgb_data   = av_malloc(W * H * 4);
-    uint8_t *rgb_src[3] = { rgb_data, NULL, NULL };
-    int rgb_stride[3]   = { 4 * W, 0, 0 };
+    uint8_t *rgb_src[4] = { rgb_data, NULL, NULL, NULL };
+    int rgb_stride[4]   = { 4 * W, 0, 0, 0 };
     uint8_t *data       = av_malloc(4 * W * H);
     uint8_t *src[4]     = { data, data + W * H, data + W * H * 2, data + W * H * 3 };
     int stride[4]       = { W, W, W, W };
diff --git a/libswscale/swscale_unscaled.c b/libswscale/swscale_unscaled.c
index d25c5facf4..eca377873b 100644
--- a/libswscale/swscale_unscaled.c
+++ b/libswscale/swscale_unscaled.c
@@ -396,17 +396,22 @@ static int planarRgbToRgbWrapper(SwsContext *c, const uint8_t* src[], int srcStr
         )
 
 /* {RGB,BGR}{15,16,24,32,32_1} -> {RGB,BGR}{15,16,24,32} */
-static int rgbToRgbWrapper(SwsContext *c, const uint8_t *src[], int srcStride[],
-                           int srcSliceY, int srcSliceH, uint8_t *dst[],
-                           int dstStride[])
+typedef void (* rgbConvFn) (const uint8_t *, uint8_t *, int);
+static rgbConvFn findRgbConvFn(SwsContext *c)
 {
     const enum PixelFormat srcFormat = c->srcFormat;
     const enum PixelFormat dstFormat = c->dstFormat;
-    const int srcBpp = (c->srcFormatBpp + 7) >> 3;
-    const int dstBpp = (c->dstFormatBpp + 7) >> 3;
     const int srcId = c->srcFormatBpp;
     const int dstId = c->dstFormatBpp;
-    void (*conv)(const uint8_t *src, uint8_t *dst, int src_size) = NULL;
+    rgbConvFn conv = NULL;
+
+#define IS_NOT_NE(bpp, fmt) \
+    (((bpp + 7) >> 3) == 2 && \
+     (!(av_pix_fmt_descriptors[fmt].flags & PIX_FMT_BE) != !HAVE_BIGENDIAN))
+
+    /* if this is non-native rgb444/555/565, don't handle it here. */
+    if (IS_NOT_NE(srcId, srcFormat) || IS_NOT_NE(dstId, dstFormat))
+        return NULL;
 
 #define CONV_IS(src, dst) (srcFormat == PIX_FMT_##src && dstFormat == PIX_FMT_##dst)
 
@@ -428,6 +433,7 @@ static int rgbToRgbWrapper(SwsContext *c, const uint8_t *src[], int srcStride[],
     if ((isBGRinInt(srcFormat) && isBGRinInt(dstFormat)) ||
         (isRGBinInt(srcFormat) && isRGBinInt(dstFormat))) {
         switch (srcId | (dstId << 16)) {
+        case 0x000F000C: conv = rgb12to15; break;
         case 0x000F0010: conv = rgb16to15; break;
         case 0x000F0018: conv = rgb24to15; break;
         case 0x000F0020: conv = rgb32to15; break;
@@ -463,6 +469,21 @@ static int rgbToRgbWrapper(SwsContext *c, const uint8_t *src[], int srcStride[],
         }
     }
 
+    return conv;
+}
+
+/* {RGB,BGR}{15,16,24,32,32_1} -> {RGB,BGR}{15,16,24,32} */
+static int rgbToRgbWrapper(SwsContext *c, const uint8_t *src[], int srcStride[],
+                           int srcSliceY, int srcSliceH, uint8_t *dst[],
+                           int dstStride[])
+
+{
+    const enum PixelFormat srcFormat = c->srcFormat;
+    const enum PixelFormat dstFormat = c->dstFormat;
+    const int srcBpp = (c->srcFormatBpp + 7) >> 3;
+    const int dstBpp = (c->dstFormatBpp + 7) >> 3;
+    rgbConvFn conv = findRgbConvFn(c);
+
     if (!conv) {
         av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
                av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
@@ -694,6 +715,8 @@ static int planarCopyWrapper(SwsContext *c, const uint8_t *src[],
             } else {
                 if (is16BPS(c->srcFormat) && is16BPS(c->dstFormat))
                     length *= 2;
+                else if (!av_pix_fmt_descriptors[c->srcFormat].comp[0].depth_minus1)
+                    length >>= 3; // monowhite/black
                 for (i = 0; i < height; i++) {
                     memcpy(dstPtr, srcPtr, length);
                     srcPtr += srcStride[plane];
@@ -748,24 +771,8 @@ void ff_get_unscaled_swscale(SwsContext *c)
         c->swScale = bgr24ToYv12Wrapper;
 
     /* RGB/BGR -> RGB/BGR (no dither needed forms) */
-    if (   isAnyRGB(srcFormat)
-        && isAnyRGB(dstFormat)
-        && srcFormat != PIX_FMT_BGR8      && dstFormat != PIX_FMT_BGR8
-        && srcFormat != PIX_FMT_RGB8      && dstFormat != PIX_FMT_RGB8
-        && srcFormat != PIX_FMT_BGR4      && dstFormat != PIX_FMT_BGR4
-        && srcFormat != PIX_FMT_RGB4      && dstFormat != PIX_FMT_RGB4
-        && srcFormat != PIX_FMT_BGR4_BYTE && dstFormat != PIX_FMT_BGR4_BYTE
-        && srcFormat != PIX_FMT_RGB4_BYTE && dstFormat != PIX_FMT_RGB4_BYTE
-        && srcFormat != PIX_FMT_MONOBLACK && dstFormat != PIX_FMT_MONOBLACK
-        && srcFormat != PIX_FMT_MONOWHITE && dstFormat != PIX_FMT_MONOWHITE
-        && srcFormat != PIX_FMT_RGB48LE   && dstFormat != PIX_FMT_RGB48LE
-        && srcFormat != PIX_FMT_RGB48BE   && dstFormat != PIX_FMT_RGB48BE
-        && srcFormat != PIX_FMT_BGR48LE   && dstFormat != PIX_FMT_BGR48LE
-        && srcFormat != PIX_FMT_BGR48BE   && dstFormat != PIX_FMT_BGR48BE
-        && (!needsDither || (c->flags&(SWS_FAST_BILINEAR|SWS_POINT)))
-        && (!(av_pix_fmt_descriptors[srcFormat].flags & PIX_FMT_BE) == !HAVE_BIGENDIAN || (c->srcFormatBpp+7)/8!=2)
-        && (!(av_pix_fmt_descriptors[dstFormat].flags & PIX_FMT_BE) == !HAVE_BIGENDIAN || (c->dstFormatBpp+7)/8!=2)
-    )
+    if (isAnyRGB(srcFormat) && isAnyRGB(dstFormat) && findRgbConvFn(c)
+        && (!needsDither || (c->flags&(SWS_FAST_BILINEAR|SWS_POINT))))
         c->swScale= rgbToRgbWrapper;
 
 #define isByteRGB(f) (\