1#version 450 core 2#define PRECISION ${PRECISION} 3#define FORMAT ${FORMAT} 4 5layout(std430) buffer; 6 7/* 8 * Output Image 9 */ 10layout(set = 0, binding = 0, FORMAT) uniform PRECISION image3D uOutput; 11 12/* 13 * Input Textures 14 */ 15layout(set = 0, binding = 1) uniform PRECISION sampler3D uInput; 16 17/* 18 * Params Buffer 19 */ 20layout(set = 0, binding = 2) uniform PRECISION restrict Block { 21 // output texture size (x=width,y=height,z=depth,w=unused) 22 ivec4 out_extents; 23 // input texture size (x=width,y=height,z=depth,w=unused) 24 ivec4 in_extents; 25 // x: size of output channel dim (values) 26 // y: size of output channel dim (texels) 27 uvec2 out_ch_info; 28 // x: size of input channel dim 29 // y: size of input channel dim up-aligned to 4 30 uvec2 in_ch_info; 31 // x: total number of channel values already appended 32 // y: offset to first channel texel being operated on 33 // z: number of channel texels being operated on 34 // w: padding 35 uvec4 appended_ch_info; 36} 37uBlock; 38 39/* 40 * Local Work Group 41 */ 42layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in; 43 44void main() { 45 const ivec3 pos = ivec3(gl_GlobalInvocationID); 46 47 if (any(greaterThanEqual(pos, uBlock.out_extents.xyz))) { 48 return; 49 } 50 51 // Determine the N, C indices that this invocation is writing to 52 const uint dst_n_idx = pos.z / uBlock.appended_ch_info.z; 53 const uint dst_c4_idx = (pos.z % uBlock.appended_ch_info.z) + uBlock.appended_ch_info.y; 54 uint dst_c_idx = dst_c4_idx * 4; 55 56 // Reconstruct the output write position based on the N, C indices 57 const uint dst_z_idx = dst_n_idx * uBlock.out_ch_info.y + dst_c4_idx; 58 const ivec3 write_pos = ivec3(pos.xy, dst_z_idx); 59 60 vec4 out_tex = imageLoad(uOutput, write_pos); 61 62 const uint src_n_offset = dst_n_idx * uBlock.in_ch_info.y; 63 64 uint dst_nc_idx = dst_z_idx * 4; 65 int src_c_idx = int(dst_c_idx - uBlock.appended_ch_info.x); 66 int src_nc_idx = int(src_n_offset) + src_c_idx; 67 68 // For each element of the output, extract the corresponding value from the 69 // input 70 for (uint i = 0; i < 4; ++i, ++dst_c_idx, ++dst_nc_idx, ++src_c_idx, ++src_nc_idx) { 71 if (src_c_idx >= 0) { 72 uint src_z_idx = src_nc_idx / 4; 73 74 vec4 in_tex = texelFetch(uInput, ivec3(pos.xy, src_z_idx), 0); 75 76 uint src_offset = src_nc_idx % 4; 77 uint dst_offset = dst_nc_idx % 4; 78 79 if (src_c_idx < uBlock.in_ch_info.x) { 80 out_tex[i] = in_tex[src_offset]; 81 } else { 82 out_tex[i] = 1.234; 83 } 84 } 85 } 86 87 imageStore(uOutput, write_pos, out_tex); 88} 89