X-Git-Url: https://git.sesse.net/?p=movit;a=blobdiff_plain;f=ycbcr_input.cpp;h=925bf2a11b2a89032b0d6f20f56110ceced0aa8f;hp=dc5e4d9fd2db9e619d9b680583b6c1f514f65220;hb=f99a1a6295d7809703f6b0dfb7ab4360916b002e;hpb=b757191bc6d258887445d88cdfe5b18666295660 diff --git a/ycbcr_input.cpp b/ycbcr_input.cpp index dc5e4d9..925bf2a 100644 --- a/ycbcr_input.cpp +++ b/ycbcr_input.cpp @@ -8,6 +8,7 @@ #include "effect_util.h" #include "resource_pool.h" #include "util.h" +#include "ycbcr.h" #include "ycbcr_input.h" using namespace Eigen; @@ -15,57 +16,16 @@ using namespace std; namespace movit { -namespace { - -// OpenGL has texel center in (0.5, 0.5), but different formats have -// chroma in various other places. If luma samples are X, the chroma -// sample is *, and subsampling is 3x3, the situation with chroma -// center in (0.5, 0.5) looks approximately like this: -// -// X X -// * -// X X -// -// If, on the other hand, chroma center is in (0.0, 0.5) (common -// for e.g. MPEG-4), the figure changes to: -// -// X X -// * -// X X -// -// In other words, (0.0, 0.0) means that the chroma sample is exactly -// co-sited on top of the top-left luma sample. Note, however, that -// this is _not_ 0.5 texels to the left, since the OpenGL's texel center -// is in (0.5, 0.5); it is in (0.25, 0.25). In a sense, the four luma samples -// define a square where chroma position (0.0, 0.0) is in texel position -// (0.25, 0.25) and chroma position (1.0, 1.0) is in texel position (0.75, 0.75) -// (the outer border shows the borders of the texel itself, ie. from -// (0, 0) to (1, 1)): -// -// --------- -// | | -// | X---X | -// | | * | | -// | X---X | -// | | -// --------- -// -// Also note that if we have no subsampling, the square will have zero -// area and the chroma position does not matter at all. -float compute_chroma_offset(float pos, unsigned subsampling_factor, unsigned resolution) -{ - float local_chroma_pos = (0.5 + pos * (subsampling_factor - 1)) / subsampling_factor; - return (0.5 - local_chroma_pos) / resolution; -} - -} // namespace - YCbCrInput::YCbCrInput(const ImageFormat &image_format, const YCbCrFormat &ycbcr_format, - unsigned width, unsigned height) + unsigned width, unsigned height, + YCbCrInputSplitting ycbcr_input_splitting, + GLenum type) : image_format(image_format), ycbcr_format(ycbcr_format), + ycbcr_input_splitting(ycbcr_input_splitting), needs_mipmaps(false), + type(type), width(width), height(height), resource_pool(NULL) @@ -73,42 +33,81 @@ YCbCrInput::YCbCrInput(const ImageFormat &image_format, pbos[0] = pbos[1] = pbos[2] = 0; texture_num[0] = texture_num[1] = texture_num[2] = 0; - 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; - - 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; + set_width(width); + set_height(height); pixel_data[0] = pixel_data[1] = pixel_data[2] = NULL; + owns_texture[0] = owns_texture[1] = owns_texture[2] = false; + + register_uniform_sampler2d("tex_y", &uniform_tex_y); + + if (ycbcr_input_splitting == YCBCR_INPUT_INTERLEAVED) { + num_channels = 1; + assert(ycbcr_format.chroma_subsampling_x == 1); + assert(ycbcr_format.chroma_subsampling_y == 1); + } else if (ycbcr_input_splitting == YCBCR_INPUT_SPLIT_Y_AND_CBCR) { + num_channels = 2; + register_uniform_sampler2d("tex_cbcr", &uniform_tex_cb); + } else { + assert(ycbcr_input_splitting == YCBCR_INPUT_PLANAR); + num_channels = 3; + register_uniform_sampler2d("tex_cb", &uniform_tex_cb); + register_uniform_sampler2d("tex_cr", &uniform_tex_cr); + } register_int("needs_mipmaps", &needs_mipmaps); } YCbCrInput::~YCbCrInput() { - for (unsigned channel = 0; channel < 3; ++channel) { - if (texture_num[channel] != 0) { - resource_pool->release_2d_texture(texture_num[channel]); - } + for (unsigned channel = 0; channel < num_channels; ++channel) { + possibly_release_texture(channel); } } void YCbCrInput::set_gl_state(GLuint glsl_program_num, const string& prefix, unsigned *sampler_num) { - for (unsigned channel = 0; channel < 3; ++channel) { + for (unsigned channel = 0; channel < num_channels; ++channel) { glActiveTexture(GL_TEXTURE0 + *sampler_num + channel); check_error(); - if (texture_num[channel] == 0) { + if (texture_num[channel] == 0 && (pbos[channel] != 0 || pixel_data[channel] != NULL)) { + GLenum format, internal_format; + if (channel == 0 && ycbcr_input_splitting == YCBCR_INPUT_INTERLEAVED) { + if (type == GL_UNSIGNED_INT_2_10_10_10_REV) { + format = GL_RGBA; + internal_format = GL_RGB10_A2; + } else if (type == GL_UNSIGNED_SHORT) { + format = GL_RGB; + internal_format = GL_RGB16; + } else { + assert(type == GL_UNSIGNED_BYTE); + format = GL_RGB; + internal_format = GL_RGB8; + } + } else if (channel == 1 && ycbcr_input_splitting == YCBCR_INPUT_SPLIT_Y_AND_CBCR) { + format = GL_RG; + if (type == GL_UNSIGNED_SHORT) { + internal_format = GL_RG16; + } else { + assert(type == GL_UNSIGNED_BYTE); + internal_format = GL_RG8; + } + } else { + format = GL_RED; + if (type == GL_UNSIGNED_SHORT) { + internal_format = GL_R16; + } else { + assert(type == GL_UNSIGNED_BYTE); + internal_format = GL_R8; + } + } + // (Re-)upload the texture. - texture_num[channel] = resource_pool->create_2d_texture(GL_R8, widths[channel], heights[channel]); + texture_num[channel] = resource_pool->create_2d_texture(internal_format, widths[channel], heights[channel]); glBindTexture(GL_TEXTURE_2D, texture_num[channel]); check_error(); - glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR); + glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, needs_mipmaps ? GL_LINEAR_MIPMAP_NEAREST : GL_LINEAR); check_error(); glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, pbos[channel]); check_error(); @@ -116,14 +115,19 @@ void YCbCrInput::set_gl_state(GLuint glsl_program_num, const string& prefix, uns check_error(); glPixelStorei(GL_UNPACK_ROW_LENGTH, pitch[channel]); check_error(); - glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, widths[channel], heights[channel], GL_RED, GL_UNSIGNED_BYTE, pixel_data[channel]); + glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, widths[channel], heights[channel], format, type, pixel_data[channel]); check_error(); glPixelStorei(GL_UNPACK_ROW_LENGTH, 0); check_error(); + if (needs_mipmaps) { + glGenerateMipmap(GL_TEXTURE_2D); + check_error(); + } glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); check_error(); glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE); check_error(); + owns_texture[channel] = true; } else { glBindTexture(GL_TEXTURE_2D, texture_num[channel]); check_error(); @@ -134,109 +138,53 @@ void YCbCrInput::set_gl_state(GLuint glsl_program_num, const string& prefix, uns check_error(); // Bind samplers. - set_uniform_int(glsl_program_num, prefix, "tex_y", *sampler_num + 0); - set_uniform_int(glsl_program_num, prefix, "tex_cb", *sampler_num + 1); - set_uniform_int(glsl_program_num, prefix, "tex_cr", *sampler_num + 2); + uniform_tex_y = *sampler_num + 0; + if (ycbcr_input_splitting != YCBCR_INPUT_INTERLEAVED) { + uniform_tex_cb = *sampler_num + 1; + } + if (ycbcr_input_splitting == YCBCR_INPUT_PLANAR) { + uniform_tex_cr = *sampler_num + 2; + } - *sampler_num += 3; + *sampler_num += num_channels; } string YCbCrInput::output_fragment_shader() { - float coeff[3], offset[3], scale[3]; - - switch (ycbcr_format.luma_coefficients) { - case YCBCR_REC_601: - // Rec. 601, page 2. - coeff[0] = 0.299; - coeff[1] = 0.587; - coeff[2] = 0.114; - break; - - case YCBCR_REC_709: - // Rec. 709, page 19. - coeff[0] = 0.2126; - coeff[1] = 0.7152; - coeff[2] = 0.0722; - break; - - case YCBCR_REC_2020: - // Rec. 2020, page 4. - coeff[0] = 0.2627; - coeff[1] = 0.6780; - coeff[2] = 0.0593; - break; - - default: - assert(false); - } - - if (ycbcr_format.full_range) { - offset[0] = 0.0 / 255.0; - offset[1] = 128.0 / 255.0; - offset[2] = 128.0 / 255.0; - - scale[0] = 1.0; - scale[1] = 1.0; - scale[2] = 1.0; - } else { - // Rec. 601, page 4; Rec. 709, page 19; Rec. 2020, page 4. - offset[0] = 16.0 / 255.0; - offset[1] = 128.0 / 255.0; - offset[2] = 128.0 / 255.0; - - scale[0] = 255.0 / 219.0; - scale[1] = 255.0 / 224.0; - scale[2] = 255.0 / 224.0; - } - - // Matrix to convert RGB to YCbCr. See e.g. Rec. 601. - Matrix3d rgb_to_ycbcr; - rgb_to_ycbcr(0,0) = coeff[0]; - rgb_to_ycbcr(0,1) = coeff[1]; - rgb_to_ycbcr(0,2) = coeff[2]; - - float cb_fac = (224.0 / 219.0) / (coeff[0] + coeff[1] + 1.0f - coeff[2]); - rgb_to_ycbcr(1,0) = -coeff[0] * cb_fac; - rgb_to_ycbcr(1,1) = -coeff[1] * cb_fac; - rgb_to_ycbcr(1,2) = (1.0f - coeff[2]) * cb_fac; - - float cr_fac = (224.0 / 219.0) / (1.0f - coeff[0] + coeff[1] + coeff[2]); - rgb_to_ycbcr(2,0) = (1.0f - coeff[0]) * cr_fac; - rgb_to_ycbcr(2,1) = -coeff[1] * cr_fac; - rgb_to_ycbcr(2,2) = -coeff[2] * cr_fac; - - // Inverting the matrix gives us what we need to go from YCbCr back to RGB. - Matrix3d ycbcr_to_rgb = rgb_to_ycbcr.inverse(); + float offset[3]; + Matrix3d ycbcr_to_rgb; + compute_ycbcr_matrix(ycbcr_format, offset, &ycbcr_to_rgb, type); string frag_shader; frag_shader = output_glsl_mat3("PREFIX(inv_ycbcr_matrix)", ycbcr_to_rgb); - - char buf[256]; - sprintf(buf, "const vec3 PREFIX(offset) = vec3(%.8f, %.8f, %.8f);\n", - offset[0], offset[1], offset[2]); - frag_shader += buf; - - sprintf(buf, "const vec3 PREFIX(scale) = vec3(%.8f, %.8f, %.8f);\n", - scale[0], scale[1], scale[2]); - frag_shader += buf; + frag_shader += output_glsl_vec3("PREFIX(offset)", offset[0], offset[1], offset[2]); float cb_offset_x = compute_chroma_offset( ycbcr_format.cb_x_position, ycbcr_format.chroma_subsampling_x, widths[1]); float cb_offset_y = compute_chroma_offset( ycbcr_format.cb_y_position, ycbcr_format.chroma_subsampling_y, heights[1]); - sprintf(buf, "const vec2 PREFIX(cb_offset) = vec2(%.8f, %.8f);\n", - cb_offset_x, cb_offset_y); - frag_shader += buf; + frag_shader += output_glsl_vec2("PREFIX(cb_offset)", cb_offset_x, cb_offset_y); float cr_offset_x = compute_chroma_offset( ycbcr_format.cr_x_position, ycbcr_format.chroma_subsampling_x, widths[2]); float cr_offset_y = compute_chroma_offset( ycbcr_format.cr_y_position, ycbcr_format.chroma_subsampling_y, heights[2]); - sprintf(buf, "const vec2 PREFIX(cr_offset) = vec2(%.8f, %.8f);\n", - cr_offset_x, cr_offset_y); - frag_shader += buf; + frag_shader += output_glsl_vec2("PREFIX(cr_offset)", cr_offset_x, cr_offset_y); + + if (ycbcr_input_splitting == YCBCR_INPUT_INTERLEAVED) { + frag_shader += "#define Y_CB_CR_SAME_TEXTURE 1\n"; + } else if (ycbcr_input_splitting == YCBCR_INPUT_SPLIT_Y_AND_CBCR) { + bool cb_cr_offsets_equal = + (fabs(ycbcr_format.cb_x_position - ycbcr_format.cr_x_position) < 1e-6) && + (fabs(ycbcr_format.cb_y_position - ycbcr_format.cr_y_position) < 1e-6); + char buf[256]; + snprintf(buf, sizeof(buf), "#define Y_CB_CR_SAME_TEXTURE 0\n#define CB_CR_SAME_TEXTURE 1\n#define CB_CR_OFFSETS_EQUAL %d\n", + cb_cr_offsets_equal); + frag_shader += buf; + } else { + frag_shader += "#define Y_CB_CR_SAME_TEXTURE 0\n#define CB_CR_SAME_TEXTURE 0\n"; + } frag_shader += read_file("ycbcr_input.frag"); return frag_shader; @@ -245,11 +193,28 @@ string YCbCrInput::output_fragment_shader() void YCbCrInput::invalidate_pixel_data() { for (unsigned channel = 0; channel < 3; ++channel) { - if (texture_num[channel] != 0) { - resource_pool->release_2d_texture(texture_num[channel]); - texture_num[channel] = 0; + possibly_release_texture(channel); + } +} + +bool YCbCrInput::set_int(const std::string& key, int value) +{ + if (key == "needs_mipmaps") { + if (ycbcr_input_splitting != YCBCR_INPUT_INTERLEAVED && value != 0) { + // We do not currently support this. + return false; } } + return Effect::set_int(key, value); +} + +void YCbCrInput::possibly_release_texture(unsigned channel) +{ + if (texture_num[channel] != 0 && owns_texture[channel]) { + resource_pool->release_2d_texture(texture_num[channel]); + texture_num[channel] = 0; + owns_texture[channel] = false; + } } } // namespace movit