Loosen up some restrictions on YCbCrInput if we have interleaved mode.

[movit] / ycbcr_input.cpp

diff --git a/ycbcr_input.cpp b/ycbcr_input.cpp
index 4c824c6d092ddc94617a2ecf18e96be4c3a51202..f748f1c617df20fcbb98a7eeb728490c335530a3 100644 (file)
--- a/ycbcr_input.cpp
+++ b/ycbcr_input.cpp
@@ -19,10 +19,12 @@ namespace movit {
 YCbCrInput::YCbCrInput(const ImageFormat &image_format,
                        const YCbCrFormat &ycbcr_format,
                        unsigned width, unsigned height,
-                       YCbCrInputSplitting ycbcr_input_splitting)
+                       YCbCrInputSplitting ycbcr_input_splitting,
+                       GLenum type)
        : image_format(image_format),
          ycbcr_format(ycbcr_format),
          ycbcr_input_splitting(ycbcr_input_splitting),
+         type(type),
          width(width),
          height(height),
          resource_pool(NULL)
@@ -30,22 +32,19 @@ 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_SPLIT_Y_AND_CBCR) {
+       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 {
@@ -69,14 +68,36 @@ void YCbCrInput::set_gl_state(GLuint glsl_program_num, const string& prefix, uns
                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 == 1 && ycbcr_input_splitting == YCBCR_INPUT_SPLIT_Y_AND_CBCR) {
+                       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;
-                               internal_format = GL_RG8;
+                               if (type == GL_UNSIGNED_SHORT) {
+                                       internal_format = GL_RG16;
+                               } else {
+                                       assert(type == GL_UNSIGNED_BYTE);
+                                       internal_format = GL_RG8;
+                               }
                        } else {
                                format = GL_RED;
-                               internal_format = GL_R8;
+                               if (type == GL_UNSIGNED_SHORT) {
+                                       internal_format = GL_R16;
+                               } else {
+                                       assert(type == GL_UNSIGNED_BYTE);
+                                       internal_format = GL_R8;
+                               }
                        }
 
                        // (Re-)upload the texture.
@@ -91,7 +112,7 @@ 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], format, 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();
@@ -125,6 +146,19 @@ string YCbCrInput::output_fragment_shader()
        Matrix3d ycbcr_to_rgb;
        compute_ycbcr_matrix(ycbcr_format, offset, &ycbcr_to_rgb);
 
+       if (type == GL_UNSIGNED_SHORT) {
+               // For 10-bit or 12-bit packed into 16-bit, we need to scale the values
+               // so that the max value goes from 1023 (or 4095) to 65535. We do this
+               // by folding the scaling into the conversion matrix, so it comes essentially
+               // for free. However, the offset is before the scaling (and thus assumes
+               // correctly scaled values), so we need to adjust that the other way.
+               double scale = 65535.0 / (ycbcr_format.num_levels - 1);
+               offset[0] /= scale;
+               offset[1] /= scale;
+               offset[2] /= scale;
+               ycbcr_to_rgb *= scale;
+       }
+
        string frag_shader;
 
        frag_shader = output_glsl_mat3("PREFIX(inv_ycbcr_matrix)", ycbcr_to_rgb);
@@ -142,13 +176,15 @@ string YCbCrInput::output_fragment_shader()
                ycbcr_format.cr_y_position, ycbcr_format.chroma_subsampling_y, heights[2]);
        frag_shader += output_glsl_vec2("PREFIX(cr_offset)", cr_offset_x, cr_offset_y);
 
-       if (ycbcr_input_splitting == YCBCR_INPUT_SPLIT_Y_AND_CBCR) {
+       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) {
                char buf[256];
-               snprintf(buf, sizeof(buf), "#define CB_CR_SAME_TEXTURE 1\n#define CB_CR_OFFSETS_EQUAL %d\n",
+               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",
                        (fabs(ycbcr_format.cb_x_position - ycbcr_format.cr_x_position) < 1e-6));
                frag_shader += buf;
        } else {
-               frag_shader += "#define CB_CR_SAME_TEXTURE 0\n";
+               frag_shader += "#define Y_CB_CR_SAME_TEXTURE 0\n#define CB_CR_SAME_TEXTURE 0\n";
        }
 
        frag_shader += read_file("ycbcr_input.frag");