hwpplayer1/suyu
1
0
Fork 0
forked from suyu/suyu
Code Pull requests Activity

astc_decoder: Simplify Select2DPartition

Browse source
This commit is contained in:
ameerj 2021-07-05 18:51:51 -04:00
parent 5665d05547
commit a75d70fa90
1 changed files with 18 additions and 37 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

55
src/video_core/host_shaders/astc_decoder.comp
View file

@ -284,14 +284,10 @@ uint Hash52(uint p) {
return p; return p;
} }
uint SelectPartition(uint seed, uint x, uint y, uint z, uint partition_count, bool small_block) { uint Select2DPartition(uint seed, uint x, uint y, uint partition_count, bool small_block) {
if (partition_count == 1) {
return 0;
}
if (small_block) { if (small_block) {
x <<= 1; x <<= 1;
y <<= 1; y <<= 1;
z <<= 1;
} }
seed += (partition_count - 1) * 1024; seed += (partition_count - 1) * 1024;
@ -305,10 +301,6 @@ uint SelectPartition(uint seed, uint x, uint y, uint z, uint partition_count, bo
uint seed6 = uint((rnum >> 20) & 0xF); uint seed6 = uint((rnum >> 20) & 0xF);
uint seed7 = uint((rnum >> 24) & 0xF); uint seed7 = uint((rnum >> 24) & 0xF);
uint seed8 = uint((rnum >> 28) & 0xF); uint seed8 = uint((rnum >> 28) & 0xF);
uint seed9 = uint((rnum >> 18) & 0xF);
uint seed10 = uint((rnum >> 22) & 0xF);
uint seed11 = uint((rnum >> 26) & 0xF);
uint seed12 = uint(((rnum >> 30) | (rnum << 2)) & 0xF);
seed1 = (seed1 * seed1); seed1 = (seed1 * seed1);
seed2 = (seed2 * seed2); seed2 = (seed2 * seed2);
@ -318,12 +310,8 @@ uint SelectPartition(uint seed, uint x, uint y, uint z, uint partition_count, bo
seed6 = (seed6 * seed6); seed6 = (seed6 * seed6);
seed7 = (seed7 * seed7); seed7 = (seed7 * seed7);
seed8 = (seed8 * seed8); seed8 = (seed8 * seed8);
seed9 = (seed9 * seed9);
seed10 = (seed10 * seed10);
seed11 = (seed11 * seed11);
seed12 = (seed12 * seed12);
int sh1, sh2, sh3; uint sh1, sh2;
if ((seed & 1) > 0) { if ((seed & 1) > 0) {
sh1 = (seed & 2) > 0 ? 4 : 5; sh1 = (seed & 2) > 0 ? 4 : 5;
sh2 = (partition_count == 3) ? 6 : 5; sh2 = (partition_count == 3) ? 6 : 5;
@ -331,25 +319,19 @@ uint SelectPartition(uint seed, uint x, uint y, uint z, uint partition_count, bo
sh1 = (partition_count == 3) ? 6 : 5; sh1 = (partition_count == 3) ? 6 : 5;
sh2 = (seed & 2) > 0 ? 4 : 5; sh2 = (seed & 2) > 0 ? 4 : 5;
} }
sh3 = (seed & 0x10) > 0 ? sh1 : sh2; seed1 >>= sh1;
seed2 >>= sh2;
seed3 >>= sh1;
seed4 >>= sh2;
seed5 >>= sh1;
seed6 >>= sh2;
seed7 >>= sh1;
seed8 >>= sh2;
seed1 = (seed1 >> sh1); uint a = seed1 * x + seed2 * y + (rnum >> 14);
seed2 = (seed2 >> sh2); uint b = seed3 * x + seed4 * y + (rnum >> 10);
seed3 = (seed3 >> sh1); uint c = seed5 * x + seed6 * y + (rnum >> 6);
seed4 = (seed4 >> sh2); uint d = seed7 * x + seed8 * y + (rnum >> 2);
seed5 = (seed5 >> sh1);
seed6 = (seed6 >> sh2);
seed7 = (seed7 >> sh1);
seed8 = (seed8 >> sh2);
seed9 = (seed9 >> sh3);
seed10 = (seed10 >> sh3);
seed11 = (seed11 >> sh3);
seed12 = (seed12 >> sh3);
uint a = seed1 * x + seed2 * y + seed11 * z + (rnum >> 14);
uint b = seed3 * x + seed4 * y + seed12 * z + (rnum >> 10);
uint c = seed5 * x + seed6 * y + seed9 * z + (rnum >> 6);
uint d = seed7 * x + seed8 * y + seed10 * z + (rnum >> 2);
a &= 0x3F; a &= 0x3F;
b &= 0x3F; b &= 0x3F;
@ -374,10 +356,6 @@ uint SelectPartition(uint seed, uint x, uint y, uint z, uint partition_count, bo
} }
} }
uint Select2DPartition(uint seed, uint x, uint y, uint partition_count, bool small_block) {
return SelectPartition(seed, x, y, 0, partition_count, small_block);
}
uint ReadBit() { uint ReadBit() {
if (current_index >= local_buff.length()) { if (current_index >= local_buff.length()) {
return 0; return 0;
@ -1281,8 +1259,11 @@ void DecompressBlock(ivec3 coord, uint block_index) {
for (uint j = 0; j < block_dims.y; j++) { for (uint j = 0; j < block_dims.y; j++) {
for (uint i = 0; i < block_dims.x; i++) { for (uint i = 0; i < block_dims.x; i++) {
uint local_partition = Select2DPartition(partition_index, i, j, num_partitions, uint local_partition = 0;
if (num_partitions > 1) {
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; vec4 p;
uvec4 C0 = ReplicateByteTo16(endpoints[local_partition][0]); uvec4 C0 = ReplicateByteTo16(endpoints[local_partition][0]);
uvec4 C1 = ReplicateByteTo16(endpoints[local_partition][1]); uvec4 C1 = ReplicateByteTo16(endpoints[local_partition][1]);