diff --git a/src/codec/SkSwizzler.cpp b/src/codec/SkSwizzler.cpp
index e1e2ecb..05636eb 100644
--- a/src/codec/SkSwizzler.cpp
+++ b/src/codec/SkSwizzler.cpp
@@ -1216,6 +1216,15 @@
     fSwizzleWidth = get_scaled_dimension(fSrcWidth, sampleX);
     fAllocatedWidth = get_scaled_dimension(fDstWidth, sampleX);
 
+    if (fDstOffsetBytes > 0) {
+        const size_t dstSwizzleBytes   = fSwizzleWidth   * fDstBPP;
+        const size_t dstAllocatedBytes = fAllocatedWidth * fDstBPP;
+        if (fDstOffsetBytes + dstSwizzleBytes > dstAllocatedBytes) {
+            SkASSERT(dstSwizzleBytes < dstAllocatedBytes);
+            fDstOffsetBytes = dstAllocatedBytes - dstSwizzleBytes;
+        }
+    }
+
     // The optimized swizzler functions do not support sampling.  Sampled swizzles
     // are already fast because they skip pixels.  We haven't seen a situation
     // where speeding up sampling has a significant impact on total decode time.