mod6_table + mod8_table

src/video_core/host_shaders/astc_decoder.comp

@@ -153,26 +153,31 @@ uint ReplicateBitTo9(uint value) {
     return value * 511;
 }
 
+const uint mod6_table = 0
+    | ((6 % 1) << (2 * 1)) | ((6 % 2) << (2 * 2))
+    | ((6 % 3) << (2 * 3)) | ((6 % 4) << (2 * 4))
+    | ((6 % 5) << (2 * 5)) | ((6 % 6) << (2 * 6));
+const uint mod8_table = 0
+    | ((8 % 1) << (2 * 1)) | ((8 % 2) << (2 * 2))
+    | ((8 % 3) << (2 * 3)) | ((8 % 4) << (2 * 4))
+    | ((8 % 5) << (2 * 5)) | ((8 % 6) << (2 * 6))
+    | ((8 % 7) << (2 * 7)) | ((8 % 8) << (2 * 8));
 // Assumes num_bits < to_bit, num_bits and to_bit != 0
-uint ReplicateBits(uint value, uint num_bits, uint to_bit) {
+uint ReplicateBits(uint value, uint num_bits, uint to_bit, uint table) {
     const uint repl = value & ((1 << num_bits) - 1);
+    const uint shift = (table >> (num_bits * 2)) & 3;
     uint v = repl;
-    v |= v << (num_bits << 0); // [ xxxx xxxr ]
-    v |= v << (num_bits << 1); // [ xxxx xxrr ]
-    v |= v << (num_bits << 2); // [ xxxx rrrr ]
-    v |= v << (num_bits << 3); // [ rrrr rrrr ]
-    const uint shift = (to_bit % num_bits);
-    v <<= shift;
-    v |= repl >> (num_bits - shift);
+    v |= v << (num_bits << 0); // [ xxxx xxrr ]
+    v |= v << (num_bits << 1); // [ xxxx rrrr ]
+    v |= v << (num_bits << 2); // [ rrrr rrrr ]
+    v = (v << shift) | repl >> (num_bits - shift);
     return v & ((1 << to_bit) - 1);
 }
-
 uint FastReplicateTo8(uint value, uint num_bits) {
-    return ReplicateBits(value, num_bits, 8);
+    return ReplicateBits(value, num_bits, 8, mod8_table);
 }
-
 uint FastReplicateTo6(uint value, uint num_bits) {
-    return ReplicateBits(value, num_bits, 6);
+    return ReplicateBits(value, num_bits, 6, mod6_table);
 }
 
 uint Hash52(uint p) {