diff --git a/examples/07-callback/callback.cpp b/examples/07-callback/callback.cpp
index 349f9c10..483918ce 100644
--- a/examples/07-callback/callback.cpp
+++ b/examples/07-callback/callback.cpp
@@ -219,8 +219,65 @@ struct BgfxCallback : public bgfx::CallbackI
 
 	virtual void screenShot(const char* _filePath, uint32_t _width, uint32_t _height, uint32_t _pitch, const void* _data, uint32_t /*_size*/, bool _yflip) BX_OVERRIDE
 	{
+		char temp[1024];
+
 		// Save screen shot as TGA.
-		saveTga(_filePath, _width, _height, _pitch, _data, false, _yflip);
+		bx::snprintf(temp, BX_COUNTOF(temp), "%s.mip0.tga", _filePath);
+		saveTga(temp, _width, _height, _pitch, _data, false, _yflip);
+
+		uint32_t width = _width;
+		uint32_t height = _height;
+		uint32_t pitch = _pitch;
+
+		uint8_t* data = (uint8_t*)_data;
+
+		// Generate mip maps.
+		uint32_t mip = 1;
+		for (; 2 <= width && 2 <= height; ++mip)
+		{
+			bx::snprintf(temp, BX_COUNTOF(temp), "%s.mip%d.tga", _filePath, mip);
+			bgfx::imageRgba8Downsample2x2(width, height, pitch, data, data);
+
+			width >>= 1;
+			height >>= 1;
+			pitch = width*4;
+
+			saveTga(temp, width, height, pitch, _data, false, _yflip);
+		}
+
+		if (width > height)
+		{
+			for (; 2 <= width; ++mip)
+			{
+				memcpy(&data[width*4], data, width*4);
+
+				bx::snprintf(temp, BX_COUNTOF(temp), "%s.mip%d.tga", _filePath, mip);
+				bgfx::imageRgba8Downsample2x2(width, 2, pitch, data, data);
+
+				width >>= 1;
+				pitch = width*4;
+
+				saveTga(temp, width, 2, pitch, _data, false, _yflip);
+			}
+		}
+		else
+		{
+			for (; 2 <= height; ++mip)
+			{
+				uint32_t* src = (uint32_t*)data;
+				for (uint32_t ii = 0; ii < height; ++ii, src += 2)
+				{
+					src[1] = src[0];
+				}
+
+				bx::snprintf(temp, BX_COUNTOF(temp), "%s.mip%d.tga", _filePath, mip);
+				bgfx::imageRgba8Downsample2x2(2, height, 8, data, data);
+
+				height >>= 1;
+
+				saveTga(temp, 2, height, 8, _data, false, _yflip);
+			}
+		}
 	}
 
 	virtual void captureBegin(uint32_t _width, uint32_t _height, uint32_t /*_pitch*/, bgfx::TextureFormat::Enum /*_format*/, bool _yflip) BX_OVERRIDE
@@ -403,7 +460,7 @@ int _main_(int /*_argc*/, char** /*_argv*/)
 		// Take screen shot at frame 150.
 		if (150 == frame)
 		{
-			bgfx::saveScreenShot("temp/frame150.tga");
+			bgfx::saveScreenShot("temp/frame150");
 		}
 
 		// Advance to next frame. Rendering thread will be kicked to 
diff --git a/include/bgfx.h b/include/bgfx.h
index fc89a995..c0a9a566 100644
--- a/include/bgfx.h
+++ b/include/bgfx.h
@@ -512,7 +512,26 @@ namespace bgfx
 	void vertexConvert(const VertexDecl& _destDecl, void* _destData, const VertexDecl& _srcDecl, const void* _srcData, uint32_t _num = 1);
 
 	/// Swizzle RGBA8 image to BGRA8.
-	void imageSwizzleBGRA8(uint8_t* _rgbaData, uint32_t _width, uint32_t _height);
+	///
+	/// @param _width Width of input image (pixels).
+	/// @param _height Height of input image (pixels).
+	/// @param _pitch Pitch of input image (bytes).
+	/// @param _src Source image.
+	/// @param _dst Destination image. Must be the same size as input image.
+	///   _dst might be pointer to the same memory as _src.
+	///
+	void imageSwizzleBgra8(uint32_t _width, uint32_t _height, const void* _src, void* _dst);
+
+	/// Downsample RGBA8 image with 2x2 pixel average filter.
+	///
+	/// @param _width Width of input image (pixels).
+	/// @param _height Height of input image (pixels).
+	/// @param _pitch Pitch of input image (bytes).
+	/// @param _src Source image.
+	/// @param _dst Destination image. Must be at least quarter size of
+	///   input image. _dst might be pointer to the same memory as _src.
+	///
+	void imageRgba8Downsample2x2(uint32_t _width, uint32_t _height, uint32_t _pitch, const void* _src, void* _dst);
 
 	/// Returns renderer backend API type.
 	RendererType::Enum getRendererType();
diff --git a/src/bgfx.cpp b/src/bgfx.cpp
index 8185c61c..bf820b0c 100755
--- a/src/bgfx.cpp
+++ b/src/bgfx.cpp
@@ -98,7 +98,10 @@ namespace bgfx
 
 		virtual void screenShot(const char* _filePath, uint32_t _width, uint32_t _height, uint32_t _pitch, const void* _data, uint32_t /*_size*/, bool _yflip) BX_OVERRIDE
 		{
-			saveTga(_filePath, _width, _height, _pitch, _data, false, _yflip);
+			bx::CrtFileWriter writer;
+			writer.open(_filePath);
+			imageWriteTga(&writer, _width, _height, _pitch, _data, false, _yflip);
+			writer.close();
 		}
 
 		virtual void captureBegin(uint32_t /*_width*/, uint32_t /*_height*/, uint32_t /*_pitch*/, TextureFormat::Enum /*_format*/, bool /*_yflip*/) BX_OVERRIDE
@@ -187,57 +190,6 @@ namespace bgfx
 		_result[15] = 1.0f;
 	}
 
-	void saveTga(const char* _filePath, uint32_t _width, uint32_t _height, uint32_t _srcPitch, const void* _src, bool _grayscale, bool _yflip)
-	{
-		FILE* file = fopen(_filePath, "wb");
-		if ( NULL != file )
-		{
-			uint8_t type = _grayscale ? 3 : 2;
-			uint8_t bpp = _grayscale ? 8 : 32;
-
-			putc(0, file);
-			putc(0, file);
-			putc(type, file);
-			putc(0, file); 
-			putc(0, file);
-			putc(0, file); 
-			putc(0, file);
-			putc(0, file);
-			putc(0, file); 
-			putc(0, file);
-			putc(0, file); 
-			putc(0, file);
-			putc(_width&0xff, file);
-			putc( (_width>>8)&0xff, file);
-			putc(_height&0xff, file);
-			putc( (_height>>8)&0xff, file);
-			putc(bpp, file);
-			putc(32, file);
-
-			uint32_t dstPitch = _width*bpp/8;
-			if (_yflip)
-			{
-				uint8_t* data = (uint8_t*)_src + _srcPitch*_height - _srcPitch;
-				for (uint32_t yy = 0; yy < _height; ++yy)
-				{
-					fwrite(data, dstPitch, 1, file);
-					data -= _srcPitch;
-				}
-			}
-			else
-			{
-				uint8_t* data = (uint8_t*)_src;
-				for (uint32_t yy = 0; yy < _height; ++yy)
-				{
-					fwrite(data, dstPitch, 1, file);
-					data += _srcPitch;
-				}
-			}
-
-			fclose(file);
-		}
-	}
-
 #include "charset.h"
 
 	void charsetFillTexture(const uint8_t* _charset, uint8_t* _rgba, uint32_t _height, uint32_t _pitch, uint32_t _bpp)
diff --git a/src/bgfx_p.h b/src/bgfx_p.h
index 04dc7844..08879d66 100755
--- a/src/bgfx_p.h
+++ b/src/bgfx_p.h
@@ -229,7 +229,7 @@ namespace bgfx
 	extern FreeFn g_free;
 
 	void release(const Memory* _mem);
-	void saveTga(const char* _filePath, uint32_t _width, uint32_t _height, uint32_t _srcPitch, const void* _src, bool _grayscale = false, bool _yflip = false);
+	void imageWriteTga(bx::WriterI* _writer, uint32_t _width, uint32_t _height, uint32_t _srcPitch, const void* _src, bool _grayscale = false, bool _yflip = false);
 	const char* getAttribName(Attrib::Enum _attr);
 	bool renderFrame();
 
diff --git a/src/image.cpp b/src/image.cpp
index 6cb22c97..a2d1bd56 100644
--- a/src/image.cpp
+++ b/src/image.cpp
@@ -8,27 +8,187 @@
 
 namespace bgfx
 {
-	static void imageSwizzleBGRA8Ref(uint8_t* _rgbaData, uint32_t _width, uint32_t _height) 
+	void imageSolid(uint32_t _width, uint32_t _height, uint32_t _solid, void* _dst)
 	{
-		uint8_t* ptr = _rgbaData;
-
-		for (uint32_t xx = 0, num = _width*_height; xx < num; ++xx)
+		uint32_t* dst = (uint32_t*)_dst;
+		for (uint32_t ii = 0, num = _width*_height; ii < num; ++ii)
 		{
-			uint8_t tmp = ptr[0];
-			ptr[0] = ptr[2];
-			ptr[2] = tmp;
-			ptr += 4;
+			*dst++ = _solid;
 		}
 	}
 
-	void imageSwizzleBGRA8(uint8_t* _rgbaData, uint32_t _width, uint32_t _height)
+	void imageChessboard(uint32_t _width, uint32_t _height, uint32_t _step, uint32_t _0, uint32_t _1, void* _dst)
+	{
+		uint32_t* dst = (uint32_t*)_dst;
+		for (uint32_t yy = 0; yy < _height; ++yy)
+		{
+			for (uint32_t xx = 0; xx < _width; ++xx)
+			{
+				uint32_t abgr = ( (xx/_step)&1) ^ ( (yy/_step)&1) ? _1 : _0;
+				*dst++ = abgr;
+			}
+		}
+	}
+
+	void imageRgba8Downsample2x2Ref(uint32_t _width, uint32_t _height, uint32_t _pitch, const void* _src, void* _dst)
+	{
+		const uint32_t dstwidth  = _width/2;
+		const uint32_t dstheight = _height/2;
+
+		if (0 == dstwidth
+		||  0 == dstheight)
+		{
+			return;
+		}
+
+		uint8_t* dst = (uint8_t*)_dst;
+		const uint8_t* src = (const uint8_t*)_src;
+		
+		for (uint32_t yy = 0, ystep = _pitch*2; yy < dstheight; ++yy, src += ystep)
+		{
+			const uint8_t* rgba = src;
+			for (uint32_t xx = 0; xx < dstwidth; ++xx, rgba += 8, dst += 4)
+			{
+				float rr = powf(rgba[       0], 2.2f);
+				float gg = powf(rgba[       1], 2.2f);
+				float bb = powf(rgba[       2], 2.2f);
+				float aa =      rgba[       3];
+				rr      += powf(rgba[       4], 2.2f);
+				gg      += powf(rgba[       5], 2.2f);
+				bb      += powf(rgba[       6], 2.2f);
+				aa      +=      rgba[       7];
+				rr      += powf(rgba[_pitch+0], 2.2f);
+				gg      += powf(rgba[_pitch+1], 2.2f);
+				bb      += powf(rgba[_pitch+2], 2.2f);
+				aa      +=      rgba[_pitch+3];
+				rr      += powf(rgba[_pitch+4], 2.2f);
+				gg      += powf(rgba[_pitch+5], 2.2f);
+				bb      += powf(rgba[_pitch+6], 2.2f);
+				aa      +=      rgba[_pitch+7];
+
+				rr *= 0.25f;
+				gg *= 0.25f;
+				bb *= 0.25f;
+				aa *= 0.25f;
+				rr = powf(rr, 1.0f/2.2f);
+				gg = powf(gg, 1.0f/2.2f);
+				bb = powf(bb, 1.0f/2.2f);
+				aa =      aa;
+				dst[0] = (uint8_t)rr;
+				dst[1] = (uint8_t)gg;
+				dst[2] = (uint8_t)bb;
+				dst[3] = (uint8_t)aa;
+			}
+		}
+	}
+
+	void imageRgba8Downsample2x2(uint32_t _width, uint32_t _height, uint32_t _pitch, const void* _src, void* _dst)
+	{
+		const uint32_t dstwidth  = _width/2;
+		const uint32_t dstheight = _height/2;
+
+		if (0 == dstwidth
+		||  0 == dstheight)
+		{
+			return;
+		}
+
+		uint8_t* dst = (uint8_t*)_dst;
+		const uint8_t* src = (const uint8_t*)_src;
+
+		using namespace bx;
+		const float4_t unpack = float4_ld(1.0f, 1.0f/256.0f, 1.0f/65536.0f, 1.0f/16777216.0f);
+		const float4_t pack   = float4_ld(1.0f, 256.0f*0.5f, 65536.0f, 16777216.0f*0.5f);
+		const float4_t umask  = float4_ild(0xff, 0xff00, 0xff0000, 0xff000000);
+		const float4_t pmask  = float4_ild(0xff, 0x7f80, 0xff0000, 0x7f800000);
+		const float4_t wflip  = float4_ild(0, 0, 0, 0x80000000);
+		const float4_t wadd   = float4_ld(0.0f, 0.0f, 0.0f, 32768.0f*65536.0f);
+		const float4_t gamma  = float4_ld(1.0f/2.2f, 1.0f/2.2f, 1.0f/2.2f, 1.0f);
+		const float4_t linear = float4_ld(2.2f, 2.2f, 2.2f, 1.0f);
+		const float4_t quater = float4_splat(0.25f);
+
+		for (uint32_t yy = 0, ystep = _pitch*2; yy < dstheight; ++yy, src += ystep)
+		{
+			const uint8_t* rgba = src;
+			for (uint32_t xx = 0; xx < dstwidth; ++xx, rgba += 8, dst += 4)
+			{
+				const float4_t abgr0  = float4_splat(rgba);
+				const float4_t abgr1  = float4_splat(rgba+4);
+				const float4_t abgr2  = float4_splat(rgba+_pitch);
+				const float4_t abgr3  = float4_splat(rgba+_pitch+4);
+
+				const float4_t abgr0m = float4_and(abgr0, umask);
+				const float4_t abgr1m = float4_and(abgr1, umask);
+				const float4_t abgr2m = float4_and(abgr2, umask);
+				const float4_t abgr3m = float4_and(abgr3, umask);
+				const float4_t abgr0x = float4_xor(abgr0m, wflip);
+				const float4_t abgr1x = float4_xor(abgr1m, wflip);
+				const float4_t abgr2x = float4_xor(abgr2m, wflip);
+				const float4_t abgr3x = float4_xor(abgr3m, wflip);
+				const float4_t abgr0f = float4_itof(abgr0x);
+				const float4_t abgr1f = float4_itof(abgr1x);
+				const float4_t abgr2f = float4_itof(abgr2x);
+				const float4_t abgr3f = float4_itof(abgr3x);
+				const float4_t abgr0c = float4_add(abgr0f, wadd);
+				const float4_t abgr1c = float4_add(abgr1f, wadd);
+				const float4_t abgr2c = float4_add(abgr2f, wadd);
+				const float4_t abgr3c = float4_add(abgr3f, wadd);
+				const float4_t abgr0n = float4_mul(abgr0c, unpack);
+				const float4_t abgr1n = float4_mul(abgr1c, unpack);
+				const float4_t abgr2n = float4_mul(abgr2c, unpack);
+				const float4_t abgr3n = float4_mul(abgr3c, unpack);
+
+				const float4_t abgr0l = float4_pow(abgr0n, linear);
+				const float4_t abgr1l = float4_pow(abgr1n, linear);
+				const float4_t abgr2l = float4_pow(abgr2n, linear);
+				const float4_t abgr3l = float4_pow(abgr3n, linear);
+
+				const float4_t sum0   = float4_add(abgr0l, abgr1l);
+				const float4_t sum1   = float4_add(abgr2l, abgr3l);
+				const float4_t sum2   = float4_add(sum0, sum1);
+				const float4_t avg0   = float4_mul(sum2, quater);
+				const float4_t avg1   = float4_pow(avg0, gamma);
+
+				const float4_t avg2   = float4_mul(avg1, pack);
+				const float4_t ftoi0  = float4_ftoi(avg2);
+				const float4_t ftoi1  = float4_and(ftoi0, pmask);
+				const float4_t zwxy   = float4_swiz_zwxy(ftoi1);
+				const float4_t tmp0   = float4_or(ftoi1, zwxy);
+				const float4_t yyyy   = float4_swiz_yyyy(tmp0);
+				const float4_t tmp1   = float4_iadd(yyyy, yyyy);
+				const float4_t result = float4_or(tmp0, tmp1);
+
+				float4_stx(dst, result);
+			}
+		}
+	}
+
+	static void imageSwizzleBgra8Ref(uint32_t _width, uint32_t _height, const void* _src, void* _dst)
+	{
+		const uint8_t* src = (uint8_t*) _src;
+		uint8_t* dst = (uint8_t*)_dst;
+
+		for (uint32_t xx = 0, num = _width*_height; xx < num; ++xx, src += 4, dst += 4)
+		{
+			uint8_t rr = src[0];
+			uint8_t gg = src[1];
+			uint8_t bb = src[2];
+			uint8_t aa = src[3];
+			dst[0] = bb;
+			dst[1] = gg;
+			dst[2] = rr;
+			dst[3] = aa;
+		}
+	}
+
+	void imageSwizzleBgra8(uint32_t _width, uint32_t _height, const void* _src, void* _dst)
 	{
 		// Test can we do four 4-byte pixels at the time.
 		if (0 != (_width&0x3)
 		||  _width < 4)
 		{
 			BX_WARN(_width < 4, "Image swizzle is taking slow path. Image width must be multiple of 4 (width %d).", _width);
-			imageSwizzleBGRA8Ref(_rgbaData, _width, _height);
+			imageSwizzleBgra8Ref(_width, _height, _src, _dst);
 			return;
 		}
 
@@ -38,19 +198,60 @@ namespace bgfx
 
 		const float4_t mf0f0 = float4_isplat(0xff00ff00);
 		const float4_t m0f0f = float4_isplat(0x00ff00ff);
-		uint8_t* ptr = _rgbaData;
+		const uint8_t* src = (uint8_t*) _src;
+		uint8_t* dst = (uint8_t*)_dst;
 
-		for (uint32_t xx = 0, num = dstpitch/16*_height; xx < num; ++xx)
+		for (uint32_t xx = 0, num = dstpitch/16*_height; xx < num; ++xx, src += 16, dst += 16)
 		{
-			const float4_t tabgr = float4_ld(ptr);
+			const float4_t tabgr = float4_ld(src);
 			const float4_t t00ab = float4_srl(tabgr, 16);
 			const float4_t tgr00 = float4_sll(tabgr, 16);
 			const float4_t tgrab = float4_or(t00ab, tgr00);
 			const float4_t ta0g0 = float4_and(tabgr, mf0f0);
 			const float4_t t0r0b = float4_and(tgrab, m0f0f);
 			const float4_t targb = float4_or(ta0g0, t0r0b);
-			float4_st(ptr, targb);
-			ptr += 16;
+			float4_st(dst, targb);
+		}
+	}
+
+	void imageWriteTga(bx::WriterI* _writer, uint32_t _width, uint32_t _height, uint32_t _srcPitch, const void* _src, bool _grayscale, bool _yflip)
+	{
+		uint8_t type = _grayscale ? 3 : 2;
+		uint8_t bpp = _grayscale ? 8 : 32;
+
+		uint8_t header[18] = {};
+		header[2] = type;
+		header[12] = _width&0xff;
+		header[13] = (_width>>8)&0xff;
+		header[14] = _height&0xff;
+		header[15] = (_height>>8)&0xff;
+		header[16] = bpp;
+		header[17] = 32;
+
+		bx::write(_writer, header, sizeof(header) );
+
+		uint32_t dstPitch = _width*bpp/8;
+		if (_yflip)
+		{
+			uint8_t* data = (uint8_t*)_src + _srcPitch*_height - _srcPitch;
+			for (uint32_t yy = 0; yy < _height; ++yy)
+			{
+				bx::write(_writer, data, dstPitch);
+				data -= _srcPitch;
+			}
+		}
+		else if (_srcPitch == dstPitch)
+		{
+			bx::write(_writer, _src, _height*_srcPitch);
+		}
+		else
+		{
+			uint8_t* data = (uint8_t*)_src;
+			for (uint32_t yy = 0; yy < _height; ++yy)
+			{
+				bx::write(_writer, data, dstPitch);
+				data += _srcPitch;
+			}
 		}
 	}
 
diff --git a/src/renderer_gl.cpp b/src/renderer_gl.cpp
index 5efdb43d..52eaf991 100644
--- a/src/renderer_gl.cpp
+++ b/src/renderer_gl.cpp
@@ -758,7 +758,7 @@ namespace bgfx
 
 			if (GL_RGBA == fmt)
 			{
-				imageSwizzleBGRA8(data, width, height);
+				imageSwizzleBgra8(width, height, data, data);
 			}
 
 			g_callback->screenShot(_filePath
@@ -1455,7 +1455,7 @@ namespace bgfx
 
 							if (swizzle)
 							{
-								imageSwizzleBGRA8(bits, width, height);
+								imageSwizzleBgra8(width, height, bits, bits);
 							}
 
 							texImage(target+side
@@ -1604,7 +1604,7 @@ namespace bgfx
 							if (NULL != data
 							&&  swizzle)
 							{
-								imageSwizzleBGRA8(data, width, height);
+								imageSwizzleBgra8(width, height, data, data);
 							}
 
 							texImage(target+side