Dx9 vs opengl Texture Lock

[omaremad]

Just wanted to point out a bug where opengl driver when locking any texture always converts to rgba8 even if the texture is a 16 bit fixed or floating point one,.

Dx textures just return a pointer to the data rather than doing the rgba8 conversion.

[omaremad]

The dx version doesnt even use IImage as a intermediate to generate the pointer.

Code: Select all

void* COpenGLTexture::lock(bool readOnly, u32 mipmapLevel)
{
	// store info about which image is locked
	IImage* image = (mipmapLevel==0)?Image:MipImage;
	ReadOnlyLock |= readOnly;
	MipLevelStored = mipmapLevel;

	// if data not available or might have changed on GPU download it
	if (!image || IsRenderTarget)
	{
		// prepare the data storage if necessary
		if (!image)
		{
			if (mipmapLevel)
			{
				u32 i=0;
				u32 width = TextureSize.Width;
				u32 height = TextureSize.Height;
				do
				{
					if (width>1)
						width>>=1;
					if (height>1)
						height>>=1;
					++i;
				}
				while (i != mipmapLevel);
				MipImage = image = new CImage(ECF_A8R8G8B8, core::dimension2du(width,height));
			}
			else
				Image = image = new CImage(ECF_A8R8G8B8, ImageSize);
			ColorFormat = ECF_A8R8G8B8;
		}
		if (!image)
			return 0;

		u8* pixels = static_cast<u8*>(image->lock());
		if (!pixels)
			return 0;

		// we need to keep the correct texture bound later on
		GLint tmpTexture;
		glGetIntegerv(GL_TEXTURE_BINDING_2D, &tmpTexture);
		glBindTexture(GL_TEXTURE_2D, TextureName);

		// allows to read pixels in top-to-bottom order
#ifdef GL_MESA_pack_invert
		if (Driver->queryOpenGLFeature(COpenGLExtensionHandler::IRR_MESA_pack_invert))
			glPixelStorei(GL_PACK_INVERT_MESA, GL_TRUE);
#endif

		// download GPU data as ARGB8 to pixels;
		glGetTexImage(GL_TEXTURE_2D, mipmapLevel, GL_BGRA_EXT, GL_UNSIGNED_BYTE, pixels);

#ifdef GL_MESA_pack_invert
		if (Driver->queryOpenGLFeature(COpenGLExtensionHandler::IRR_MESA_pack_invert))
			glPixelStorei(GL_PACK_INVERT_MESA, GL_FALSE);
		else
#endif
		{
			// opengl images are horizontally flipped, so we have to fix that here.
			const s32 pitch=image->getPitch();
			u8* p2 = pixels + (image->getDimension().Height - 1) * pitch;
			u8* tmpBuffer = new u8[pitch];
			for (u32 i=0; i < image->getDimension().Height; i += 2)
			{
				memcpy(tmpBuffer, pixels, pitch);
				memcpy(pixels, p2, pitch);
				memcpy(p2, tmpBuffer, pitch);
				pixels += pitch;
				p2 -= pitch;
			}
			delete [] tmpBuffer;
		}
		image->unlock();

		//reset old bound texture
		glBindTexture(GL_TEXTURE_2D, tmpTexture);
	}
	return image->lock();
}

[robmar]

Is it a bug, or a feature? There´s a flag used to create the irrlicht device, to force ARGB8 mode, I think what youy´re seeing is a default mode only used if the data is no longer available, as its cached, I guess, no?

[hybrid]

It's basically a not yet fully implemented feature, so more a bug than a feature in the current state. Correct behavior would be to convert the d3d textures as well, though, as we usually don't give access to the floating point textures so far.