X-Git-Url: https://git.sesse.net/?p=movit;a=blobdiff_plain;f=ycbcr_input.h;h=31ba42bcbbacb9daa9e3fd1cadd9a80b6dee4a19;hp=2220f5d1ebd9392730f98439cbe3d98a7e6d8c30;hb=98f458e22ce732e6c50a9856d7fd636dca49b4c3;hpb=37f56fcbe571b2322243f6de59494bf9e0cbb37a diff --git a/ycbcr_input.h b/ycbcr_input.h index 2220f5d..31ba42b 100644 --- a/ycbcr_input.h +++ b/ycbcr_input.h @@ -1,51 +1,65 @@ -#ifndef _YCBCR_INPUT_H -#define _YCBCR_INPUT_H 1 - -// YCbCrInput is for handling planar 8-bit Y'CbCr (also sometimes, usually rather -// imprecisely, called “YUV”), which is typically what you get from a video decoder. -// It upsamples planes as needed, using the default linear upsampling OpenGL gives you. - -#include +#ifndef _MOVIT_YCBCR_INPUT_H +#define _MOVIT_YCBCR_INPUT_H 1 + +// YCbCrInput is for handling Y'CbCr (also sometimes, usually rather +// imprecisely, called “YUV”), which is typically what you get from a video +// decoder. It supports these formats: +// +// * 8-bit planar Y'CbCr, possibly subsampled (e.g. 4:2:0). +// * 8-bit semiplanar Y'CbCr (Y' in one plane, CbCr in another), +// possibly subsampled. +// * 8-bit interleaved (chunked) Y'CbCr, no subsampling (4:4:4 only). +// * 10-bit interleaved (chunked) Y'CbCr packed into 32-bit words +// (10:10:10:2), no subsampling (4:4:4 only). +// +// For the former case, it upsamples planes as needed, using the default linear +// upsampling OpenGL gives you. Note that YCbCr422InterleavedInput supports the +// important special case of 8-bit 4:2:2 interleaved. + +#include #include #include #include "effect.h" +#include "effect_chain.h" #include "image_format.h" #include "input.h" +#include "ycbcr.h" -struct YCbCrFormat { - // Which formula for Y' to use. - YCbCrLumaCoefficients luma_coefficients; +namespace movit { - // If true, assume Y'CbCr coefficients are full-range, ie. go from 0 to 255 - // instead of the limited 220/225 steps in classic MPEG. For instance, - // JPEG uses the Rec. 601 luma coefficients, but full range. - bool full_range; +class ResourcePool; - // Sampling factors for chroma components. For no subsampling (4:4:4), - // set both to 1. - unsigned chroma_subsampling_x, chroma_subsampling_y; +// Whether the data is planar (Y', Cb and Cr in one texture each) or not. +enum YCbCrInputSplitting { + // The standard, default case; Y', Cb and Cr in one texture each. + YCBCR_INPUT_PLANAR, - // Positioning of the chroma samples. MPEG-1 and JPEG is (0.5, 0.5); - // MPEG-2 and newer typically are (0.0, 0.5). - float cb_x_position, cb_y_position; - float cr_x_position, cr_y_position; + // Y' in one texture, and then Cb and Cr interleaved in one texture. + // In particular, this is a superset of the relatively popular NV12 mode. + // If you specify this mode, the “Cr” pointer texture will be unused + // (the ”Cb” texture contains both). + YCBCR_INPUT_SPLIT_Y_AND_CBCR, + + // Y', Cb and Cr interleaved in the same texture (the “Y” texture; + // “Cb” and “Cr” are unused). This means you cannot have any subsampling; + // 4:4:4 only. + YCBCR_INPUT_INTERLEAVED, }; class YCbCrInput : public Input { public: + // Type can be GL_UNSIGNED_BYTE for 8-bit, or GL_UNSIGNED_INT_2_10_10_10_REV + // for 10-bit (YCBCR_INPUT_INTERLEAVED only). YCbCrInput(const ImageFormat &image_format, const YCbCrFormat &ycbcr_format, - unsigned width, unsigned height); + unsigned width, unsigned height, + YCbCrInputSplitting ycbcr_input_splitting = YCBCR_INPUT_PLANAR, + GLenum type = GL_UNSIGNED_BYTE); ~YCbCrInput(); virtual std::string effect_type_id() const { return "YCbCrInput"; } - // Create the texture itself. We cannot do this in the constructor, - // because we don't necessarily know all the settings (sRGB texture, - // mipmap generation) at that point. - void finalize(); - virtual bool can_output_linear_gamma() const { return false; } virtual AlphaHandling alpha_handling() const { return OUTPUT_BLANK_ALPHA; } @@ -58,42 +72,108 @@ public: unsigned get_height() const { return height; } Colorspace get_color_space() const { return image_format.color_space; } GammaCurve get_gamma_curve() const { return image_format.gamma_curve; } + virtual bool can_supply_mipmaps() const { return false; } // Tells the input where to fetch the actual pixel data. Note that if you change // this data, you must either call set_pixel_data() again (using the same pointer // is fine), or invalidate_pixel_data(). Otherwise, the texture won't be re-uploaded // on subsequent frames. - void set_pixel_data(unsigned channel, const unsigned char *pixel_data) + // + // The data can either be a regular pointer (if pbo==0), or a byte offset + // into a PBO. The latter will allow you to start uploading the texture data + // asynchronously to the GPU, if you have any CPU-intensive work between the + // call to set_pixel_data() and the actual rendering. In either case, + // the pointer (and PBO, if set) has to be valid at the time of the render call. + void set_pixel_data(unsigned channel, const unsigned char *pixel_data, GLuint pbo = 0) { - assert(channel >= 0 && channel < 3); + assert(type == GL_UNSIGNED_BYTE || type == GL_UNSIGNED_INT_2_10_10_10_REV); + assert(channel >= 0 && channel < num_channels); this->pixel_data[channel] = pixel_data; + this->pbos[channel] = pbo; + invalidate_pixel_data(); + } + + void set_pixel_data(unsigned channel, const uint32_t *pixel_data, GLuint pbo = 0) + { + assert(type == GL_UNSIGNED_INT_2_10_10_10_REV); + assert(channel == 0); + this->pixel_data[channel] = reinterpret_cast(pixel_data); + this->pbos[channel] = pbo; + invalidate_pixel_data(); + } + + void invalidate_pixel_data(); + + // Note: Sets pitch to width, so even if your pitch is unchanged, + // you will need to re-set it after this call. + void set_width(unsigned width) + { + assert(width != 0); + this->width = width; + + assert(width % ycbcr_format.chroma_subsampling_x == 0); + pitch[0] = widths[0] = width; + pitch[1] = widths[1] = width / ycbcr_format.chroma_subsampling_x; + pitch[2] = widths[2] = width / ycbcr_format.chroma_subsampling_x; invalidate_pixel_data(); } - void invalidate_pixel_data() + void set_height(unsigned height) { - needs_update = true; + assert(height != 0); + this->height = height; + + assert(height % ycbcr_format.chroma_subsampling_y == 0); + heights[0] = height; + heights[1] = height / ycbcr_format.chroma_subsampling_y; + heights[2] = height / ycbcr_format.chroma_subsampling_y; + invalidate_pixel_data(); + } + + void set_pitch(unsigned channel, unsigned pitch) + { + assert(pitch != 0); + assert(channel >= 0 && channel < num_channels); + this->pitch[channel] = pitch; + invalidate_pixel_data(); + } + + // Tells the input to use the specific OpenGL texture as pixel data for the given + // channel. The comments on FlatInput::set_texture_num() also apply here, except + // that this input generally does not use mipmaps. + void set_texture_num(unsigned channel, GLuint texture_num) + { + possibly_release_texture(channel); + this->texture_num[channel] = texture_num; + this->owns_texture[channel] = false; } - void set_pitch(unsigned channel, unsigned pitch) { - assert(channel >= 0 && channel < 3); - if (this->pitch[channel] != pitch) { - this->pitch[channel] = pitch; - needs_pbo_recreate = true; - } + virtual void inform_added(EffectChain *chain) + { + resource_pool = chain->get_resource_pool(); } + bool set_int(const std::string& key, int value); + private: + // Release the texture in the given channel if we have any, and it is owned by us. + void possibly_release_texture(unsigned channel); + ImageFormat image_format; YCbCrFormat ycbcr_format; + GLuint num_channels; + YCbCrInputSplitting ycbcr_input_splitting; + GLenum type; GLuint pbos[3], texture_num[3]; - bool needs_update, needs_pbo_recreate, finalized; - - int needs_mipmaps; + GLint uniform_tex_y, uniform_tex_cb, uniform_tex_cr; unsigned width, height, widths[3], heights[3]; const unsigned char *pixel_data[3]; unsigned pitch[3]; + bool owns_texture[3]; + ResourcePool *resource_pool; }; -#endif // !defined(_YCBCR_INPUT_H) +} // namespace movit + +#endif // !defined(_MOVIT_YCBCR_INPUT_H)