#include "chroma_subsampler.h"
+#include "v210_converter.h"
#include <vector>
"#version 130 \n"
"in vec2 tc0, tc1; \n"
"uniform sampler2D cbcr_tex; \n"
- "out vec4 FragColor; \n"
+ "out vec4 FragColor, FragColor2; \n"
"void main() { \n"
" FragColor = 0.5 * (texture(cbcr_tex, tc0) + texture(cbcr_tex, tc1)); \n"
+ " FragColor2 = FragColor; \n"
"} \n";
cbcr_program_num = resource_pool->compile_glsl_program(cbcr_vert_shader, cbcr_frag_shader, frag_shader_outputs);
check_error();
+ cbcr_chroma_offset_0_location = get_uniform_location(cbcr_program_num, "foo", "chroma_offset_0");
+ check_error();
+ cbcr_chroma_offset_1_location = get_uniform_location(cbcr_program_num, "foo", "chroma_offset_1");
+ check_error();
cbcr_texture_sampler_uniform = glGetUniformLocation(cbcr_program_num, "cbcr_tex");
check_error();
uyvy_program_num = resource_pool->compile_glsl_program(uyvy_vert_shader, uyvy_frag_shader, frag_shader_outputs);
check_error();
+ uyvy_luma_offset_0_location = get_uniform_location(uyvy_program_num, "foo", "luma_offset_0");
+ check_error();
+ uyvy_luma_offset_1_location = get_uniform_location(uyvy_program_num, "foo", "luma_offset_1");
+ check_error();
+ uyvy_chroma_offset_0_location = get_uniform_location(uyvy_program_num, "foo", "chroma_offset_0");
+ check_error();
+ uyvy_chroma_offset_1_location = get_uniform_location(uyvy_program_num, "foo", "chroma_offset_1");
+ check_error();
uyvy_y_texture_sampler_uniform = glGetUniformLocation(uyvy_program_num, "y_tex");
check_error();
};
vbo = generate_vbo(2, GL_FLOAT, sizeof(vertices), vertices);
check_error();
+
+ // v210 compute shader.
+ if (v210Converter::has_hardware_support()) {
+ string v210_shader_src = R"(#version 150
+#extension GL_ARB_compute_shader : enable
+#extension GL_ARB_shader_image_load_store : enable
+layout(local_size_x=2, local_size_y=16) in;
+layout(r16) uniform restrict readonly image2D in_y;
+uniform sampler2D in_cbcr; // Of type RG16.
+layout(rgb10_a2) uniform restrict writeonly image2D outbuf;
+uniform float inv_width, inv_height;
+
+void main()
+{
+ int xb = int(gl_GlobalInvocationID.x); // X block number.
+ int y = int(gl_GlobalInvocationID.y); // Y (actual line).
+ float yf = (gl_GlobalInvocationID.y + 0.5f) * inv_height; // Y float coordinate.
+
+ // Load and scale CbCr values, sampling in-between the texels to get
+ // to (left/4 + center/2 + right/4).
+ vec2 pix_cbcr[3];
+ for (int i = 0; i < 3; ++i) {
+ vec2 a = texture(in_cbcr, vec2((xb * 6 + i * 2) * inv_width, yf)).xy;
+ vec2 b = texture(in_cbcr, vec2((xb * 6 + i * 2 + 1) * inv_width, yf)).xy;
+ pix_cbcr[i] = (a + b) * (0.5 * 65535.0 / 1023.0);
+ }
+
+ // Load and scale the Y values. Note that we use integer coordinates here,
+ // so we don't need to offset by 0.5.
+ float pix_y[6];
+ for (int i = 0; i < 6; ++i) {
+ pix_y[i] = imageLoad(in_y, ivec2(xb * 6 + i, y)).x * (65535.0 / 1023.0);
+ }
+
+ imageStore(outbuf, ivec2(xb * 4 + 0, y), vec4(pix_cbcr[0].x, pix_y[0], pix_cbcr[0].y, 1.0));
+ imageStore(outbuf, ivec2(xb * 4 + 1, y), vec4(pix_y[1], pix_cbcr[1].x, pix_y[2], 1.0));
+ imageStore(outbuf, ivec2(xb * 4 + 2, y), vec4(pix_cbcr[1].y, pix_y[3], pix_cbcr[2].x, 1.0));
+ imageStore(outbuf, ivec2(xb * 4 + 3, y), vec4(pix_y[4], pix_cbcr[2].y, pix_y[5], 1.0));
+}
+)";
+ GLuint shader_num = movit::compile_shader(v210_shader_src, GL_COMPUTE_SHADER);
+ check_error();
+ v210_program_num = glCreateProgram();
+ check_error();
+ glAttachShader(v210_program_num, shader_num);
+ check_error();
+ glLinkProgram(v210_program_num);
+ check_error();
+
+ GLint success;
+ glGetProgramiv(v210_program_num, GL_LINK_STATUS, &success);
+ check_error();
+ if (success == GL_FALSE) {
+ GLchar error_log[1024] = {0};
+ glGetProgramInfoLog(v210_program_num, 1024, nullptr, error_log);
+ fprintf(stderr, "Error linking program: %s\n", error_log);
+ exit(1);
+ }
+
+ v210_in_y_pos = glGetUniformLocation(v210_program_num, "in_y");
+ check_error();
+ v210_in_cbcr_pos = glGetUniformLocation(v210_program_num, "in_cbcr");
+ check_error();
+ v210_outbuf_pos = glGetUniformLocation(v210_program_num, "outbuf");
+ check_error();
+ v210_inv_width_pos = glGetUniformLocation(v210_program_num, "inv_width");
+ check_error();
+ v210_inv_height_pos = glGetUniformLocation(v210_program_num, "inv_height");
+ check_error();
+ } else {
+ v210_program_num = 0;
+ }
}
ChromaSubsampler::~ChromaSubsampler()
check_error();
glDeleteBuffers(1, &vbo);
check_error();
+ if (v210_program_num != 0) {
+ glDeleteProgram(v210_program_num);
+ check_error();
+ }
}
-void ChromaSubsampler::subsample_chroma(GLuint cbcr_tex, unsigned width, unsigned height, GLuint dst_tex)
+void ChromaSubsampler::subsample_chroma(GLuint cbcr_tex, unsigned width, unsigned height, GLuint dst_tex, GLuint dst2_tex)
{
- GLuint vao;
- glGenVertexArrays(1, &vao);
- check_error();
-
+ GLuint vao = resource_pool->create_vec2_vao({ cbcr_position_attribute_index, cbcr_texcoord_attribute_index }, vbo);
glBindVertexArray(vao);
check_error();
// Extract Cb/Cr.
- GLuint fbo = resource_pool->create_fbo(dst_tex);
+ GLuint fbo;
+ if (dst2_tex <= 0) {
+ fbo = resource_pool->create_fbo(dst_tex);
+ } else {
+ fbo = resource_pool->create_fbo(dst_tex, dst2_tex);
+ }
glBindFramebuffer(GL_FRAMEBUFFER, fbo);
glViewport(0, 0, width/2, height/2);
check_error();
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
check_error();
- float chroma_offset_0[] = { -1.0f / width, 0.0f };
- float chroma_offset_1[] = { -0.0f / width, 0.0f };
- set_uniform_vec2(cbcr_program_num, "foo", "chroma_offset_0", chroma_offset_0);
- set_uniform_vec2(cbcr_program_num, "foo", "chroma_offset_1", chroma_offset_1);
-
- glUniform1i(cbcr_texture_sampler_uniform, 0);
-
- glBindBuffer(GL_ARRAY_BUFFER, vbo);
+ glUniform2f(cbcr_chroma_offset_0_location, -1.0f / width, 0.0f);
check_error();
-
- for (GLint attr_index : { cbcr_position_attribute_index, cbcr_texcoord_attribute_index }) {
- glEnableVertexAttribArray(attr_index);
- check_error();
- glVertexAttribPointer(attr_index, 2, GL_FLOAT, GL_FALSE, 0, BUFFER_OFFSET(0));
- check_error();
- }
+ glUniform2f(cbcr_chroma_offset_1_location, -0.0f / width, 0.0f);
+ check_error();
+ glUniform1i(cbcr_texture_sampler_uniform, 0);
glDrawArrays(GL_TRIANGLES, 0, 3);
check_error();
- for (GLint attr_index : { cbcr_position_attribute_index, cbcr_texcoord_attribute_index }) {
- glDisableVertexAttribArray(attr_index);
- check_error();
- }
-
glUseProgram(0);
check_error();
glBindFramebuffer(GL_FRAMEBUFFER, 0);
check_error();
+ glBindVertexArray(0);
+ check_error();
resource_pool->release_fbo(fbo);
- glDeleteVertexArrays(1, &vao);
- check_error();
+ resource_pool->release_vec2_vao(vao);
}
void ChromaSubsampler::create_uyvy(GLuint y_tex, GLuint cbcr_tex, unsigned width, unsigned height, GLuint dst_tex)
{
- GLuint vao;
- glGenVertexArrays(1, &vao);
+ GLuint vao = resource_pool->create_vec2_vao({ cbcr_position_attribute_index, cbcr_texcoord_attribute_index }, vbo);
+ glBindVertexArray(vao);
check_error();
glBindVertexArray(vao);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
check_error();
- float y_offset_0[] = { -0.5f / width, 0.0f };
- float y_offset_1[] = { 0.5f / width, 0.0f };
- float cbcr_offset0[] = { -1.0f / width, 0.0f };
- float cbcr_offset1[] = { -0.0f / width, 0.0f };
- set_uniform_vec2(uyvy_program_num, "foo", "luma_offset_0", y_offset_0);
- set_uniform_vec2(uyvy_program_num, "foo", "luma_offset_1", y_offset_1);
- set_uniform_vec2(uyvy_program_num, "foo", "chroma_offset_0", cbcr_offset0);
- set_uniform_vec2(uyvy_program_num, "foo", "chroma_offset_1", cbcr_offset1);
+ glUniform2f(uyvy_luma_offset_0_location, -0.5f / width, 0.0f);
+ check_error();
+ glUniform2f(uyvy_luma_offset_1_location, 0.5f / width, 0.0f);
+ check_error();
+ glUniform2f(uyvy_chroma_offset_0_location, -1.0f / width, 0.0f);
+ check_error();
+ glUniform2f(uyvy_chroma_offset_1_location, -0.0f / width, 0.0f);
+ check_error();
glBindBuffer(GL_ARRAY_BUFFER, vbo);
check_error();
- for (GLint attr_index : { uyvy_position_attribute_index, uyvy_texcoord_attribute_index }) {
- if (attr_index == -1) continue;
- glEnableVertexAttribArray(attr_index);
- check_error();
- glVertexAttribPointer(attr_index, 2, GL_FLOAT, GL_FALSE, 0, BUFFER_OFFSET(0));
- check_error();
- }
-
glDrawArrays(GL_TRIANGLES, 0, 3);
check_error();
- for (GLint attr_index : { uyvy_position_attribute_index, uyvy_texcoord_attribute_index }) {
- if (attr_index == -1) continue;
- glDisableVertexAttribArray(attr_index);
- check_error();
- }
-
glActiveTexture(GL_TEXTURE0);
check_error();
glUseProgram(0);
check_error();
glBindFramebuffer(GL_FRAMEBUFFER, 0);
check_error();
+ glBindVertexArray(0);
+ check_error();
resource_pool->release_fbo(fbo);
- glDeleteVertexArrays(1, &vao);
+ resource_pool->release_vec2_vao(vao);
+}
+
+void ChromaSubsampler::create_v210(GLuint y_tex, GLuint cbcr_tex, unsigned width, unsigned height, GLuint dst_tex)
+{
+ assert(v210_program_num != 0);
+
+ glUseProgram(v210_program_num);
+ check_error();
+
+ glUniform1i(v210_in_y_pos, 0);
+ check_error();
+ glUniform1i(v210_in_cbcr_pos, 1);
+ check_error();
+ glUniform1i(v210_outbuf_pos, 2);
+ check_error();
+ glUniform1f(v210_inv_width_pos, 1.0 / width);
+ check_error();
+ glUniform1f(v210_inv_height_pos, 1.0 / height);
+ check_error();
+
+ glActiveTexture(GL_TEXTURE0);
+ check_error();
+ glBindTexture(GL_TEXTURE_2D, y_tex); // We don't actually need to bind it, but we need to set the state.
+ check_error();
+ glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
+ 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();
+ glBindImageTexture(0, y_tex, 0, GL_FALSE, 0, GL_READ_ONLY, GL_R16); // This is the real bind.
+ check_error();
+
+ glActiveTexture(GL_TEXTURE1);
+ check_error();
+ glBindTexture(GL_TEXTURE_2D, cbcr_tex);
+ check_error();
+ glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
+ 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();
+
+ glBindImageTexture(2, dst_tex, 0, GL_FALSE, 0, GL_WRITE_ONLY, GL_RGB10_A2);
+ check_error();
+
+ // Actually run the shader. We use workgroups of size 2x16 threadst , and each thread
+ // processes 6x1 input pixels, so round up to number of 12x16 pixel blocks.
+ glDispatchCompute((width + 11) / 12, (height + 15) / 16, 1);
+
+ glBindTexture(GL_TEXTURE_2D, 0);
+ check_error();
+ glActiveTexture(GL_TEXTURE0);
+ check_error();
+ glUseProgram(0);
+ check_error();
}