hipnosuer
/
Zuma
forked from ShuriZma/suyu
1
0
Fork 0
Code Pull Requests Activity

weights refactor

This commit is contained in:
Ameer J 2023-07-30 12:26:48 -04:00
parent 6ff65abd62
commit ac09cc3504
1 changed files with 22 additions and 26 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

48
src/video_core/host_shaders/astc_decoder.comp
View File

@ -116,8 +116,6 @@ const uint REPLICATE_7_BIT_TO_8_TABLE[128] =
237, 239, 241, 243, 245, 247, 249, 251, 253, 255); 237, 239, 241, 243, 245, 247, 249, 251, 253, 255);
// Input ASTC texture globals // Input ASTC texture globals
uint current_index = 0;
int bitsread = 0;
int total_bitsread = 0; int total_bitsread = 0;
uvec4 local_buff; uvec4 local_buff;
@ -144,13 +142,6 @@ int texel_vector_index = 0;
uint unquantized_texel_weights[2][144]; uint unquantized_texel_weights[2][144];
uint SwizzleOffset(uvec2 pos) {
uint x = pos.x;
uint y = pos.y;
return ((x % 64) / 32) * 256 + ((y % 8) / 2) * 64 + ((x % 32) / 16) * 32 +
(y % 2) * 16 + (x % 16);
}
// Replicates low num_bits such that [(to_bit - 1):(to_bit - 1 - from_bit)] // Replicates low num_bits such that [(to_bit - 1):(to_bit - 1 - from_bit)]
// is the same as [(num_bits - 1):0] and repeats all the way down. // is the same as [(num_bits - 1):0] and repeats all the way down.
uint Replicate(uint val, uint num_bits, uint to_bit) { uint Replicate(uint val, uint num_bits, uint to_bit) {
@ -1224,33 +1215,40 @@ void DecompressBlock(ivec3 coord) {
uint local_partition = 0; uint local_partition = 0;
if (num_partitions > 1) { if (num_partitions > 1) {
local_partition = Select2DPartition(partition_index, i, j, num_partitions, local_partition = Select2DPartition(partition_index, i, j, num_partitions,
(block_dims.y * block_dims.x) < 32); (block_dims.y * block_dims.x) < 32);
} }
vec4 p; const uvec4 C0 = ReplicateByteTo16(endpoints[local_partition][0]);
uvec4 C0 = ReplicateByteTo16(endpoints[local_partition][0]); const uvec4 C1 = ReplicateByteTo16(endpoints[local_partition][1]);
uvec4 C1 = ReplicateByteTo16(endpoints[local_partition][1]); const uint weight_offset = (j * block_dims.x + i);
uvec4 plane_vec = uvec4(0); const uint primary_weight = unquantized_texel_weights[weight_offset][0];
uvec4 weight_vec = uvec4(0); uvec4 weight_vec = uvec4(primary_weight);
for (uint c = 0; c < 4; c++) { if (params.dual_plane) {
if (params.dual_plane && (((plane_index + 1) & 3) == c)) { const uint secondary_weight = unquantized_texel_weights[weight_offset][1];
plane_vec[c] = 1; for (uint c = 0; c < 4; c++) {
const bool is_secondary = ((plane_index + 1u) & 3u) == c;
weight_vec[c] = is_secondary ? secondary_weight : primary_weight;
} }
weight_vec[c] = unquantized_texel_weights[plane_vec[c]][j * block_dims.x + i];
} }
vec4 Cf = vec4((C0 * (uvec4(64) - weight_vec) + C1 * weight_vec + uvec4(32)) / 64); const vec4 Cf =
p = (Cf / 65535.0); vec4((C0 * (uvec4(64) - weight_vec) + C1 * weight_vec + uvec4(32)) / 64);
const vec4 p = (Cf / 65535.0);
imageStore(dest_image, coord + ivec3(i, j, 0), p.gbar); imageStore(dest_image, coord + ivec3(i, j, 0), p.gbar);
} }
} }
} }
uint SwizzleOffset(uvec2 pos) {
uint x = pos.x;
uint y = pos.y;
return ((x % 64) / 32) * 256 + ((y % 8) / 2) * 64 + ((x % 32) / 16) * 32 +
(y % 2) * 16 + (x % 16);
}
void main() { void main() {
uvec3 pos = gl_GlobalInvocationID; uvec3 pos = gl_GlobalInvocationID;
pos.x <<= BYTES_PER_BLOCK_LOG2; pos.x <<= BYTES_PER_BLOCK_LOG2;
// Read as soon as possible due to its latency
const uint swizzle = SwizzleOffset(pos.xy); const uint swizzle = SwizzleOffset(pos.xy);
const uint block_y = pos.y >> GOB_SIZE_Y_SHIFT; const uint block_y = pos.y >> GOB_SIZE_Y_SHIFT;
uint offset = 0; uint offset = 0;
@ -1264,8 +1262,6 @@ void main() {
if (any(greaterThanEqual(coord, imageSize(dest_image)))) { if (any(greaterThanEqual(coord, imageSize(dest_image)))) {
return; return;
} }
current_index = 0;
bitsread = 0;
local_buff = astc_data[offset / 16]; local_buff = astc_data[offset / 16];
DecompressBlock(coord); DecompressBlock(coord);
} }