diff --git a/src/gl/system/gl_swframebuffer.cpp b/src/gl/system/gl_swframebuffer.cpp
index 2bf3cf2ee..39b3307b6 100644
--- a/src/gl/system/gl_swframebuffer.cpp
+++ b/src/gl/system/gl_swframebuffer.cpp
@@ -2383,6 +2383,41 @@ bool OpenGLSWFrameBuffer::OpenGLPal::Update()
 	// See explanation in UploadPalette() for skipat rationale.
 	skipat = MIN(numEntries, DoColorSkip ? 256 - 8 : 256);
 
+#ifndef NO_SSE
+	// Manual SSE vectorized version here to workaround a bug in GCC's auto-vectorizer
+
+	int sse_count = skipat / 4 * 4;
+	for (i = 0; i < sse_count; i += 4)
+	{
+		_mm_storeu_si128((__m128i*)(&buff[i]), _mm_loadu_si128((__m128i*)(&pal[i])));
+	}
+	switch (skipat - i)
+	{
+	// fall through is intentional
+	case 3: buff[i] = pal[i].d; i++;
+	case 2: buff[i] = pal[i].d; i++;
+	case 1: buff[i] = pal[i].d; i++;
+	default: i++;
+	}
+	sse_count = numEntries / 4 * 4;
+	__m128i alphamask = _mm_set1_epi32(0xff000000);
+	while (i < sse_count)
+	{
+		__m128i lastcolor = _mm_loadu_si128((__m128i*)(&pal[i - 1]));
+		__m128i color = _mm_loadu_si128((__m128i*)(&pal[i]));
+		_mm_storeu_si128((__m128i*)(&buff[i]), _mm_or_si128(_mm_and_si128(alphamask, color), _mm_andnot_si128(alphamask, lastcolor)));
+		i += 4;
+	}
+	switch (numEntries - i)
+	{
+	// fall through is intentional
+	case 3: buff[i] = ColorARGB(pal[i].a, pal[i - 1].r, pal[i - 1].g, pal[i - 1].b); i++;
+	case 2: buff[i] = ColorARGB(pal[i].a, pal[i - 1].r, pal[i - 1].g, pal[i - 1].b); i++;
+	case 1: buff[i] = ColorARGB(pal[i].a, pal[i - 1].r, pal[i - 1].g, pal[i - 1].b); i++;
+	default: break;
+	}
+
+#else
 	for (i = 0; i < skipat; ++i)
 	{
 		buff[i] = ColorARGB(pal[i].a, pal[i].r, pal[i].g, pal[i].b);
@@ -2391,6 +2426,7 @@ bool OpenGLSWFrameBuffer::OpenGLPal::Update()
 	{
 		buff[i] = ColorARGB(pal[i].a, pal[i - 1].r, pal[i - 1].g, pal[i - 1].b);
 	}
+#endif
 	if (numEntries > 1)
 	{
 		i = numEntries - 1;