Dynamitos/Seele
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Very strange mask bug

Browse Source
This commit is contained in:
Dynamitos
2024-10-30 10:28:10 +01:00
parent 289b759498
commit 2811757476
11 changed files with 276 additions and 950 deletions
Download Patch File Download Diff File Expand all files Collapse all files
res/shaders/terrain/CBT.slang
+71 -31
View File
@@ -60,7 +60,33 @@ static const uint32_t OCBT_depth_offset[18] = { 0, // Level 0
0, // Level 18
};
static const uint64_t OCBT_bit_mask[18] = { 0xffffffff, // Root 17
static const uint OCBT_bit_mask_high[18] = {
0x0, // Root 17
0x0, // Level 16
0x0, // level 15
0x0, // level 14
0x0, // level 13
0x0, // level 12
0x0, // level 11
0x0, // level 10
0x0, // level 9
0x0, // level 8
0x0, // level 8
0xffffffff, // level 7
0x0, // Level 6
0x0, // level 5
0x0, // level 4
0x0, // level 3
0x0, // level 2
0x0, // level 1
};
static const uint OCBT_bit_mask_low[18] = {
0xffffffff, // Root 17
0xffffffff, // Level 16
0xffffffff, // level 15
0xffffffff, // level 14
@@ -73,7 +99,7 @@ static const uint64_t OCBT_bit_mask[18] = { 0xffffffff, // Root 17
0xffff, // level 8
0xff, // level 8
0xffffffffffffffff, // level 7
0xffffffff, // level 7
0xffffffff, // Level 6
0xffff, // level 5
0xff, // level 4
@@ -172,14 +198,16 @@ uint get_heap_element(uint id, inout HeapDebug debug)
{
uint32_t slot = first_bit / 32;
uint32_t local_id = first_bit % 32;
uint32_t target_bits = (gs_cbtTree[slot] >> local_id) & uint32_t(OCBT_bit_mask[depth]);
return (gs_cbtTree[slot] >> local_id) & uint32_t(OCBT_bit_mask[depth]);
return (pParams.cbtBuffer[slot] >> local_id) & OCBT_bit_mask_low[depth];
}
else
{
uint32_t slot = first_bit / 64;
uint32_t local_id = first_bit % 64;
uint64_t target_bits = (pParams.bitFieldBuffer[slot] >> local_id) & OCBT_bit_mask[depth];
uint64_t mask = OCBT_bit_mask_high[depth];
mask = (mask << 32) | OCBT_bit_mask_low[depth];
uint64_t target_bits = (pParams.bitFieldBuffer[slot] >> local_id) & mask;
debug.target_bits = target_bits;
return countbits(uint(target_bits >> 32)) + countbits(uint(target_bits));
}
}
@@ -198,8 +226,8 @@ void set_heap_element(uint id, uint value)
uint local_id = first_bit % 32;
// Extract the relevant bits
gs_cbtTree[slot] &= ~(uint32_t(OCBT_bit_mask[depth]) << local_id);
gs_cbtTree[slot] |= ((uint32_t(OCBT_bit_mask[depth]) & value) << local_id);
gs_cbtTree[slot] &= ~(uint32_t(OCBT_bit_mask_low[depth]) << local_id);
gs_cbtTree[slot] |= ((uint32_t(OCBT_bit_mask_low[depth]) & value) << local_id);
}
// Should not be called if depth > TREE_LAST_LEVEL
@@ -216,8 +244,8 @@ void set_heap_element_atomic(uint id, uint value)
uint local_id = first_bit % 32;
// Extract the relevant bits
InterlockedAnd(gs_cbtTree[slot], ~(uint32_t(OCBT_bit_mask[depth]) << local_id));
InterlockedOr(gs_cbtTree[slot], ((uint32_t(OCBT_bit_mask[depth]) & value) << local_id));
InterlockedAnd(gs_cbtTree[slot], ~(uint32_t(OCBT_bit_mask_low[depth]) << local_id));
InterlockedOr(gs_cbtTree[slot], ((uint32_t(OCBT_bit_mask_low[depth]) & value) << local_id));
}
// Function that returns the number of active bits
@@ -233,37 +261,49 @@ uint bit_count(uint depth, uint element)
}
// decodes the position of the i-th zero in the bitfield
uint decode_bit_complement(uint handle, uint dispatchID)
uint decode_bit_complement(uint handle, inout DebugStruct debug)
{
uint temp = handle;
DebugStruct debug;
uint x = 0;
uint bitID = 1u;
uint c = OCBT_NUM_ELEMENTS / 2u;
while (bitID < OCBT_NUM_ELEMENTS) {
uint heapValue = c - get_heap_element(2u * bitID, debug.heapValues[x++]);
uint b = handle < heapValue ? 0u : 1u;
bitID = 2u * bitID + b;
handle -= heapValue * b;
uint b2 = 2u * bitID;
uint h = get_heap_element(b2, debug.heapValues[x]);
uint heapValue = c - h;
uint b;
if(handle < heapValue)
{
b = 0u;
}
else
{
b = 1u;
}
debug.heapValues[x].handle = handle;
debug.heapValues[x].heapValue = heapValue;
debug.heapValues[x].c = c;
debug.heapValues[x].h = h;
x++;
bitID = b2 + b;
uint t = heapValue * b;
handle = handle - t;
c /= 2u;
}
uint result = (bitID ^ OCBT_NUM_ELEMENTS);
handle = temp;
for(uint i = 0; i < OCBT_NUM_ELEMENTS; ++i)
{
if(get_bit(i) == 0)
{
if(handle == 0)
{
pParams.debugBuffer[dispatchID] = debug;
return i;
}
handle--;
}
}
return (bitID ^ OCBT_NUM_ELEMENTS);
//handle = temp;
//for(uint i = 0; i < OCBT_NUM_ELEMENTS; ++i)
//{
// if(get_bit(i) == 0)
// {
// if(handle == 0)
// {
// return i;
// }
// handle--;
// }
//}
}
void reduce(uint groupIndex)