mupdf渲染过程（一）：颜色

mupdf除了解析PDF功能之外，还有一个强大的功能就是渲染文字和图像，本文介绍mupdf渲染过程中涉及到的颜色问题：包括颜色空间，颜色转换，lcms的使用。

1.初始化

    mupdf初始化第一步是实例化fz_context *ctx，fz_context是Mupdf最基本的数据结构，它是文件句柄，fz_context结构体fz_colorspace_context *colorspace成员变量就是颜色空间内容。

实例化函数fz_new_context，初始化了一些列Mupdf操作，代码fz_new_colorspace_context(ctx)初始化MUPDF使用的颜色空间：

void fz_new_colorspace_context(fz_context *ctx)
{
	ctx->colorspace = fz_malloc_struct(ctx, fz_colorspace_context);
	ctx->colorspace->ctx_refs = 1;
	set_no_icc(ctx->colorspace);
#ifdef NO_ICC
	fz_set_cmm_engine(ctx, NULL);
#else
	fz_set_cmm_engine(ctx, &fz_cmm_engine_lcms);
#endif
}

fz_new_colorspace_context内部使用了NO_ICC宏定义，通过宏定义，确定mupdf使用的颜色空间是否基于ICC文件，关于fz_set_cmm_engine函数，是否基于ICC也给出了明确的实现：

if (engine)
	{
		cct->gray = fz_new_icc_colorspace(ctx, FZ_ICC_PROFILE_GRAY, 1, NULL);
		cct->rgb = fz_new_icc_colorspace(ctx, FZ_ICC_PROFILE_RGB, 3, NULL);
		cct->bgr = fz_new_icc_colorspace(ctx, FZ_ICC_PROFILE_BGR, 3, NULL);
		cct->cmyk = fz_new_icc_colorspace(ctx, FZ_ICC_PROFILE_CMYK, 4, NULL);
		cct->lab = fz_new_icc_colorspace(ctx, FZ_ICC_PROFILE_LAB, 3, NULL);
	}
	else
		set_no_icc(cct);

对于icc方式，mupdf内部使用了5种颜色空间，分别是gray, rgb, bgr,cmyk,lab;各类颜色空间初始化方式：

fz_colorspace *
fz_new_icc_colorspace(fz_context *ctx, const char *name, int num, fz_buffer *buf)
{
	fz_colorspace *cs = NULL;
	fz_iccprofile *profile;
	int is_lab = 0;
	enum fz_colorspace_type type = FZ_COLORSPACE_NONE;
	int flags = FZ_COLORSPACE_IS_ICC;

	profile = fz_malloc_struct(ctx, fz_iccprofile);
	fz_try(ctx)
	{
		if (buf == NULL)
		{
			size_t size;
			const unsigned char *data;
			data = fz_lookup_icc(ctx, name, &size);
			profile->buffer = fz_new_buffer_from_shared_data(ctx, data, size);
			is_lab = (strcmp(name, FZ_ICC_PROFILE_LAB) == 0);
			profile->bgr = (strcmp(name, FZ_ICC_PROFILE_BGR) == 0);
			flags |= FZ_COLORSPACE_IS_DEVICE;
		}
		else
		{
			profile->buffer = fz_keep_buffer(ctx, buf);
		}

		fz_cmm_init_profile(ctx, profile);

        XXXXXX

		fz_md5_icc(ctx, profile);
        
        XXXXXX
	
		cs = fz_new_colorspace(ctx, name, type, flags, profile->num_devcomp, NULL, NULL, NULL, is_lab ? clamp_lab_icc : clamp_default_icc, free_icc, profile, sizeof(profile));


	return cs;
#endif
}

mupdf使用fz_iccprofile结构表示一个icc文件的解析结果；fz_new_icc_colorspace函数内部，有两个重要步骤：第一个是fz_lookup_icc，解析icc文件生成数据，

const unsigned char *
fz_lookup_icc(fz_context *ctx, const char *name, size_t *size)
{
#ifndef NO_ICC
	if (fz_get_cmm_engine(ctx) == NULL)
		return *size = 0, NULL;
	if (!strcmp(name, FZ_ICC_PROFILE_GRAY)) {
		extern const int fz_resources_icc_gray_icc_size;
		extern const unsigned char fz_resources_icc_gray_icc[];
		*size = fz_resources_icc_gray_icc_size;
		return fz_resources_icc_gray_icc;
	}
	if (!strcmp(name, FZ_ICC_PROFILE_RGB) || !strcmp(name, FZ_ICC_PROFILE_BGR)) {
		extern const int fz_resources_icc_rgb_icc_size;
		extern const unsigned char fz_resources_icc_rgb_icc[];
		*size = fz_resources_icc_rgb_icc_size;
		return fz_resources_icc_rgb_icc;
	}
	if (!strcmp(name, FZ_ICC_PROFILE_CMYK)) {
		extern const int fz_resources_icc_cmyk_icc_size;
		extern const unsigned char fz_resources_icc_cmyk_icc[];
		*size = fz_resources_icc_cmyk_icc_size;
		return fz_resources_icc_cmyk_icc;
	}
	if (!strcmp(name, FZ_ICC_PROFILE_LAB)) {
		extern const int fz_resources_icc_lab_icc_size;
		extern const unsigned char fz_resources_icc_lab_icc[];
		*size = fz_resources_icc_lab_icc_size;
		return fz_resources_icc_lab_icc;
	}
#endif
	return *size = 0, NULL;
}

这里用CRAY颜色空间举例，找到fz_resources_icc_gray_icc变量的定义，它是一个全局变量：

const int fz_resources_icc_gray_icc_size = 416;
const unsigned char fz_resources_icc_gray_icc[] = {
0,0,1,160,0,0,0,0,2,16,0,0,109,110,116,114,71,82,65,89,88,89,90,32,0,0,0,
0,0,0,0,0,0,0,0,0,97,99,115,112,65,80,80,76,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,246,214,0,1,0,0,0,0,211,45,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,5,100,101,115,99,0,0,0,192,0,0,0,125,99,112,114,116,0,0,1,64,0,0,0,40,
119,116,112,116,0,0,1,104,0,0,0,20,98,107,112,116,0,0,1,124,0,0,0,20,107,
84,82,67,0,0,1,144,0,0,0,14,100,101,115,99,0,0,0,0,0,0,0,35,65,114,116,105,
102,101,120,32,83,111,102,116,119,97,114,101,32,115,71,114,97,121,32,73,67,
67,32,80,114,111,102,105,108,101,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,116,101,120,116,0,0,0,0,
67,111,112,121,114,105,103,104,116,32,65,114,116,105,102,101,120,32,83,111,
102,116,119,97,114,101,32,50,48,49,49,0,88,89,90,32,0,0,0,0,0,0,243,84,0,
1,0,0,0,1,22,207,88,89,90,32,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,99,117,114,118,
0,0,0,0,0,0,0,1,1,205,0,0,};

到这里，可以明白这是gray icc文件内容，它是Mupdf自定义的icc文件，其他颜色空间icc文件也是同理。fz_lookup_icc生成profile内容之后，第二个就是fz_cmm_init_profile(ctx, profile)了，他内部调用了fz_lcms_init_profile(fz_cmm_instance *instance, fz_iccprofile *profile)：

static void
fz_lcms_init_profile(fz_cmm_instance *instance, fz_iccprofile *profile)
{
	cmsContext cmm_ctx = (cmsContext)instance;
	fz_context *ctx = (fz_context *)cmsGetContextUserData(cmm_ctx);
	size_t size;
	unsigned char *data;

	DEBUG_LCMS_MEM(("@@@@@@@ Create Profile Start:: mupdf ctx = %p lcms ctx = %p \n", (void*)ctx, (void*)cmm_ctx));

	size = fz_buffer_storage(ctx, profile->buffer, &data);
	profile->cmm_handle = cmsOpenProfileFromMemTHR(cmm_ctx, data, (cmsUInt32Number)size);
	if (profile->cmm_handle == NULL)
	{
		profile->num_devcomp = 0;
		fz_throw(ctx, FZ_ERROR_GENERIC, "cmsOpenProfileFromMem failed");
	}
	profile->num_devcomp = fz_lcms_num_devcomps(cmm_ctx, profile);

	DEBUG_LCMS_MEM(("@@@@@@@ Create Profile End:: mupdf ctx = %p lcms ctx = %p profile = %p profile_cmm = %p \n", (void*)ctx, (void*)cmm_ctx, (void*)profile, (void*)profile->cmm_handle));
}

cmsOpenProfileFromMemTHR是lcms的函数，用于颜色空间的转换，他初始化了一个基于icc的的文件句柄，fz_cmm_init_profile结束后，cs = fz_new_colorspace将profile赋给颜色空间，这样一个gray的颜色空间初始化完毕了。

2 颜色转换

        mupdf在渲染文字和图像的时候，都要对颜色进行转换，因为设备的颜色空间和pdf文件样本的颜色空间可能存在不一致情况，如果不进行转换，会出现显示效果和原PDF文件不一致情况。

mupdf颜色转换函数static fz_overprint *
resolve_color(fz_context *ctx, fz_overprint *op, const float *color, fz_colorspace *colorspace, float alpha, const fz_color_params *color_params, unsigned char *colorbv, fz_pixmap *dest)，需要传入输入颜色值，输入颜色空间，alpha值，PDF颜色参数，输出颜色值，目标图片；内部调用了fz_convert_color，它有两部：先查找合适的颜色转换器，然后使用颜色转换器对颜色进行转换：

void
fz_convert_color(fz_context *ctx, const fz_color_params *params, const fz_colorspace *is, const fz_colorspace *ds, float *dv, const fz_colorspace *ss, const float *sv)
{
	fz_color_converter cc;
	fz_find_color_converter(ctx, &cc, is, ds, ss, params);
	cc.convert(ctx, &cc, dv, sv);
	fz_drop_color_converter(ctx, &cc);
}

关于查找过程，如果非ICC模式，直接使用颜色值转换方式：

if (ds == default_gray)
			cc->convert = rgb2g;
		else if (ds == default_bgr)
			cc->convert = rgb2bgr;
		else if (ds == default_cmyk)
			cc->convert = rgb2cmyk;
		else
			cc->convert = std_conv_color;

如果是基于ICC模式，使用icc_conv_color函数，在使用icc_conv_color转换之前，需要做一个工作，就是要建立一个转换的句柄：cc->link = fz_get_icc_link(ctx, ds, 0, ss_base, 0, is, params, 2, 0, &cc->n);关于fz_get_icc_link，下面给出一组调用堆栈关系：

void
fz_lcms_init_link(fz_cmm_instance *instance, fz_icclink *link, const fz_iccprofile *dst, int dst_extras, const fz_iccprofile *src, int src_extras, const fz_iccprofile *prf, const fz_color_params *rend, int cmm_flags, int num_bytes, int copy_spots)
{
	cmsContext cmm_ctx = (cmsContext)instance;
	fz_context *ctx = (fz_context *)cmsGetContextUserData(cmm_ctx);

	cmsUInt32Number src_data_type, des_data_type;
	cmsColorSpaceSignature src_cs, des_cs;
	int src_num_chan, des_num_chan;
	int lcms_src_cs, lcms_des_cs;
	unsigned int flag = cmsFLAGS_LOWRESPRECALC | cmm_flags;

	DEBUG_LCMS_MEM(("@@@@@@@ Create Link Start:: mupdf ctx = %p lcms ctx = %p src = %p des = %p \n", (void*)ctx, (void*)cmm_ctx, (void*)src->cmm_handle, (void*)dst->cmm_handle));

	/* src */
	src_cs = cmsGetColorSpace(cmm_ctx, src->cmm_handle);
	lcms_src_cs = _cmsLCMScolorSpace(cmm_ctx, src_cs);
	if (lcms_src_cs < 0)
		lcms_src_cs = 0;
	src_num_chan = cmsChannelsOf(cmm_ctx, src_cs);
	src_data_type = (COLORSPACE_SH(lcms_src_cs) | CHANNELS_SH(src_num_chan) | DOSWAP_SH(src->bgr) | SWAPFIRST_SH(src->bgr && (src_extras != 0)) | BYTES_SH(num_bytes) | EXTRA_SH(src_extras));

	/* dst */
	des_cs = cmsGetColorSpace(cmm_ctx, dst->cmm_handle);
	lcms_des_cs = _cmsLCMScolorSpace(cmm_ctx, des_cs);
	if (lcms_des_cs < 0)
		lcms_des_cs = 0;
	des_num_chan = cmsChannelsOf(cmm_ctx, des_cs);
	des_data_type =  (COLORSPACE_SH(lcms_des_cs) | CHANNELS_SH(des_num_chan) | DOSWAP_SH(dst->bgr) | SWAPFIRST_SH(dst->bgr && (dst_extras != 0)) | BYTES_SH(num_bytes) | EXTRA_SH(dst_extras));

	/* flags */
	if (rend->bp)
		flag |= cmsFLAGS_BLACKPOINTCOMPENSATION;

	if (copy_spots)
		flag |= cmsFLAGS_COPY_ALPHA;

	link->depth = num_bytes;
	link->src_extras = src_extras;
	link->dst_extras = dst_extras;
	link->copy_spots = copy_spots;

	if (prf == NULL)
	{
		link->cmm_handle = cmsCreateTransformTHR(cmm_ctx, src->cmm_handle, src_data_type, dst->cmm_handle, des_data_type, rend->ri, flag);
		if (!link->cmm_handle)
			fz_throw(ctx, FZ_ERROR_GENERIC, "cmsCreateTransform failed");
	}
}

颜色空间转换句柄生成核心函数fz_lcms_init_link，依然是调用lcms接口cmsCreateTransformTHR，它需要输入源颜色空间lcms句柄，源颜色空间数据格式，关于数据格式，lcms提供了宏定义，这里主要使用了颜色空间值（COLORSPACE_SH），通道数（CHANNELS_SH），采样比特（多少字节的颜色值）（BYTES_SH）。颜色空间和通道数的获取，也是使用了lcms接口cmsGetColorSpace，_cmsLCMScolorSpace和cmsChannelsOf。

转换句柄生成之后，就要调用icc_conv_color进行颜色转换：

void
fz_lcms_transform_color(fz_cmm_instance *instance, fz_icclink *link, unsigned short *dst, const unsigned short *src)
{
	cmsContext cmm_ctx = (cmsContext)instance;
	cmsHTRANSFORM hTransform = (cmsHTRANSFORM) link->cmm_handle;

	cmsDoTransform(cmm_ctx, hTransform, src, dst, 1);
}

icc_conv_color调用了fz_lcms_transform_color，其内部调用的cmsDoTransform也是lcms接口，结束之后，一次颜色和颜色空间的转换也结束了，最后生成和目标颜色空间对应的颜色值。

3 总结

      对于Mupdf颜色转换，总体过程总结如下

      颜色空间初始化->初始化mupdf定义的5种颜色空间->判断是否icc模式（如果是）->使用mupdf定义pro文件初始化数据->调用lcms函数生成profile句柄。

      颜色空间转换->生成转换器->判断是否icc模式（如果是)->创建icc转换句柄->调用lcms生成句柄->根据转换器转换颜色空间->调用lcms转换接口。

     以上都是基于icc模式的流程，对于非icc模式流程，则简单许多，不需要调用lcms接口，直接使用mupdf颜色值转换即可，流程也和icc差不多，这里不做介绍

     上面介绍了MUPDF颜色转换的全部流程，但是还有很多细节没有写到，还有非icc模式的转换，颜色空间hash表去重存储，颜色空间md5比对，颜色值预处理等。