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

Trying out more debugging

Browse Source
This commit is contained in:
Dynamitos
2024-10-27 14:41:23 +01:00
parent c7c7e4dd3e
commit 289b759498
9 changed files with 160 additions and 166 deletions
Download Patch File Download Diff File Expand all files Collapse all files
res/shaders/terrain/CBT.slang
+41 -53
View File
@@ -108,7 +108,7 @@ static const uint32_t OCBT_bit_count[18] = { 32, // Root 17
#define BUFFER_ELEMENT_PER_LANE ((OCBT_TREE_NUM_SLOTS + WORKGROUP_SIZE - 1) / WORKGROUP_SIZE)
#define BUFFER_ELEMENT_PER_LANE_NO_BITFIELD ((OCBT_TREE_NUM_SLOTS + WORKGROUP_SIZE - 1) / WORKGROUP_SIZE)
#define BITFIELD_ELEMENT_PER_LANE ((OCBT_BITFIELD_NUM_SLOTS + WORKGROUP_SIZE - 1) / WORKGROUP_SIZE)
#define WAVE_TREE_DEPTH uint(17)
#define WAVE_TREE_DEPTH uint(log2(OCBT_NUM_ELEMENTS))
uint32_t cbt_size()
{
@@ -154,7 +154,7 @@ uint get_bit(uint bitID)
return uint((pParams.bitFieldBuffer[slot] & (1uLL << local_id)) >> local_id);
}
uint get_heap_element(uint id)
uint get_heap_element(uint id, inout HeapDebug debug)
{
// Figure out the location of the first bit of this element
uint32_t real_heap_id = id - 1;
@@ -162,6 +162,12 @@ uint get_heap_element(uint id)
uint32_t level_first_element = (1u << depth) - 1;
uint32_t id_in_level = real_heap_id - level_first_element;
uint32_t first_bit = OCBT_depth_offset[depth] + OCBT_bit_count[depth] * id_in_level;
debug.id = id;
debug.real_heap_id = real_heap_id;
debug.depth = depth;
debug.level_first_element = level_first_element;
debug.id_in_level = id_in_level;
debug.first_bit = first_bit;
if (depth < FIRST_VIRTUAL_LEVEL)
{
uint32_t slot = first_bit / 32;
@@ -222,59 +228,37 @@ uint bit_count()
uint bit_count(uint depth, uint element)
{
return get_heap_element((1u << depth) + element);
}
// decodes the position of the i-th one in the bitfield
uint decode_bit(uint handle)
{
//uint bitID = 1;
//for (uint currentDepth = 0; currentDepth < WAVE_TREE_DEPTH; ++currentDepth)
//{
// uint heapValue = get_heap_element(2 * bitID);
// uint b = handle < heapValue ? 0 : 1;
//
// bitID = 2 * bitID + b;
// handle -= heapValue * b;
//}
//
//return (bitID ^ OCBT_NUM_ELEMENTS);
for(uint i = 0; i < OCBT_NUM_ELEMENTS; ++i)
{
if(get_bit(i) == 1)
{
if(handle == 0)
{
return i;
}
handle--;
}
}
HeapDebug debug;
return get_heap_element((1u << depth) + element, debug);
}
// decodes the position of the i-th zero in the bitfield
uint decode_bit_complement(uint handle)
uint decode_bit_complement(uint handle, uint dispatchID)
{
//uint bitID = 1u;
//uint c = OCBT_NUM_ELEMENTS / 2u;
//
//while (bitID < OCBT_NUM_ELEMENTS) {
// uint heapValue = c - get_heap_element(2u * bitID);
// uint b = handle < heapValue ? 0u : 1u;
//
// bitID = 2u * bitID + b;
// handle -= heapValue * b;
// c /= 2u;
//}
//
//return (bitID ^ OCBT_NUM_ELEMENTS);
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;
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--;
@@ -284,6 +268,7 @@ uint decode_bit_complement(uint handle)
void reduce(uint groupIndex)
{
HeapDebug debug;
// First we do a reduction until each lane has exactly one element to process
uint initial_pass_size = OCBT_NUM_ELEMENTS / WORKGROUP_SIZE;
for (uint it = initial_pass_size / 64, offset = OCBT_NUM_ELEMENTS / 64; it > 0 ; it >>=1, offset >>=1)
@@ -293,7 +278,7 @@ void reduce(uint groupIndex)
for (uint heapID = minHeapID; heapID < maxHeapID; ++heapID)
{
set_heap_element(heapID, get_heap_element(heapID * 2) + get_heap_element(heapID * 2 + 1));
set_heap_element(heapID, get_heap_element(heapID * 2, debug) + get_heap_element(heapID * 2 + 1, debug));
}
}
GroupMemoryBarrierWithGroupSync();
@@ -303,7 +288,7 @@ void reduce(uint groupIndex)
if (groupIndex < s)
{
uint v = s + groupIndex;
set_heap_element(v, get_heap_element(v * 2) + get_heap_element(v * 2 + 1));
set_heap_element(v, get_heap_element(v * 2, debug) + get_heap_element(v * 2 + 1, debug));
}
GroupMemoryBarrierWithGroupSync();
}
@@ -338,6 +323,7 @@ void reduce_prepass(uint dispatchThreadID)
void reduce_first_pass(uint dispatchThreadID, uint groupIndex)
{
HeapDebug debug;
// Load the lowest level (and only the last level)
const uint level0Offset = OCBT_depth_offset[TREE_LAST_LEVEL] / 32;
if (groupIndex % 2 == 0)
@@ -355,13 +341,13 @@ void reduce_first_pass(uint dispatchThreadID, uint groupIndex)
for (uint heapID = minHeapID; heapID < maxHeapID; ++heapID)
{
set_heap_element(heapID, get_heap_element(heapID * 2) + get_heap_element(heapID * 2 + 1));
set_heap_element(heapID, get_heap_element(heapID * 2, debug) + get_heap_element(heapID * 2 + 1, debug));
}
}
// Last pass needs to be atomic
uint heapID = offset + (dispatchThreadID * it);
set_heap_element_atomic(heapID, get_heap_element(heapID * 2) + get_heap_element(heapID * 2 + 1));
set_heap_element_atomic(heapID, get_heap_element(heapID * 2, debug) + get_heap_element(heapID * 2 + 1, debug));
GroupMemoryBarrierWithGroupSync();
@@ -376,6 +362,7 @@ void reduce_first_pass(uint dispatchThreadID, uint groupIndex)
void reduce_second_pass(uint groupIndex)
{
HeapDebug debug;
// Load the lowest level (and only the last level)
const uint level0Offset = OCBT_depth_offset[9] / 32;
for (uint e = 0; e < 4; ++e)
@@ -394,7 +381,7 @@ void reduce_second_pass(uint groupIndex)
for (uint heapID = minHeapID; heapID < maxHeapID; ++heapID)
{
set_heap_element(heapID, get_heap_element(heapID * 2) + get_heap_element(heapID * 2 + 1));
set_heap_element(heapID, get_heap_element(heapID * 2, debug) + get_heap_element(heapID * 2 + 1, debug));
}
}
GroupMemoryBarrierWithGroupSync();
@@ -404,7 +391,7 @@ void reduce_second_pass(uint groupIndex)
if (groupIndex < s)
{
uint v = s + groupIndex;
set_heap_element(v, get_heap_element(v * 2) + get_heap_element(v * 2 + 1));
set_heap_element(v, get_heap_element(v * 2, debug) + get_heap_element(v * 2 + 1, debug));
}
GroupMemoryBarrierWithGroupSync();
}
@@ -423,6 +410,7 @@ void reduce_second_pass(uint groupIndex)
void reduce_no_bitfield(uint groupIndex)
{
HeapDebug debug;
// First we do a reduction until each lane has exactly one element to process
uint initial_pass_size = OCBT_NUM_ELEMENTS / WORKGROUP_SIZE;
for (uint it = initial_pass_size / 128, offset = OCBT_NUM_ELEMENTS / 128; it > 0 ; it >>=1, offset >>=1)
@@ -432,7 +420,7 @@ void reduce_no_bitfield(uint groupIndex)
for (uint heapID = minHeapID; heapID < maxHeapID; ++heapID)
{
set_heap_element(heapID, get_heap_element(heapID * 2) + get_heap_element(heapID * 2 + 1));
set_heap_element(heapID, get_heap_element(heapID * 2, debug) + get_heap_element(heapID * 2 + 1, debug));
}
}
GroupMemoryBarrierWithGroupSync();
@@ -442,7 +430,7 @@ void reduce_no_bitfield(uint groupIndex)
if (groupIndex < s)
{
uint v = s + groupIndex;
set_heap_element(v, get_heap_element(v * 2) + get_heap_element(v * 2 + 1));
set_heap_element(v, get_heap_element(v * 2, debug) + get_heap_element(v * 2 + 1, debug));
}
GroupMemoryBarrierWithGroupSync();
}
res/shaders/terrain/CBTCompute.slang
+1 -1
View File
@@ -78,7 +78,7 @@ void Allocate(uint groupIndex : SV_GroupIndex, uint dispatchID : SV_DispatchThre
//for(uint i = 0; i < pParams.allocateBuffer[0]; ++i)
//{
// Allocate the required bits
AllocateElement(pParams.allocateBuffer[1 + dispatchID]);
AllocateElement(pParams.allocateBuffer[1 + dispatchID], dispatchID);
//}
}
res/shaders/terrain/LEB.slang
+4 -4
View File
@@ -310,9 +310,9 @@ void EvaluateLEB(uint currentID : SV_DispatchThreadID, uint groupIndex: SV_Group
EvaluateElementPosition(cHeapID, 0, pParams.geometry.baseDepth, pParams.currentVertexBuffer, parentTri, childTri);
// Export the child
pParams.lebPositionBuffer[3 * currentID + 0] = float4(1000 * (childTri.p[0]), 1.0f);
pParams.lebPositionBuffer[3 * currentID + 1] = float4(1000 * (childTri.p[1]), 1.0f);
pParams.lebPositionBuffer[3 * currentID + 2] = float4(1000 * (childTri.p[2]), 1.0f);
pParams.lebPositionBuffer[3 * currentID + 0] = float4((childTri.p[0]), 1.0f);
pParams.lebPositionBuffer[3 * currentID + 1] = float4((childTri.p[1]), 1.0f);
pParams.lebPositionBuffer[3 * currentID + 2] = float4((childTri.p[2]), 1.0f);
// Export the fourth element
if (pParams.geometry.baseDepth < depth)
@@ -321,6 +321,6 @@ void EvaluateLEB(uint currentID : SV_DispatchThreadID, uint groupIndex: SV_Group
const uint parentOffset = 3 * pParams.geometry.totalNumElements;
// Transform the coordinate to planet space
pParams.lebPositionBuffer[parentOffset + currentID] = float4(1000 * (cHeapID % 2 == 0 ? parentTri.p[0] : parentTri.p[2]), 1.0f);
pParams.lebPositionBuffer[parentOffset + currentID] = float4((cHeapID % 2 == 0 ? parentTri.p[0] : parentTri.p[2]), 1.0f);
}
}
res/shaders/terrain/Parameters.slang
+11 -5
View File
@@ -26,13 +26,19 @@ struct UpdateCB
float farPlaneDistance;
}
struct HeapDebug
{
uint id;
uint real_heap_id;
uint depth;
uint level_first_element;
uint id_in_level;
uint first_bit;
};
struct DebugStruct
{
uint4 indices;
uint baseHeapID;
uint subdivision;
uint propagateLocation;
uint numSiblings;
HeapDebug heapValues[18];
};
res/shaders/terrain/update_utils.slang
+2 -2
View File
@@ -320,7 +320,7 @@ void SplitElement(uint currentID, uint baseDepth)
InterlockedAdd(pParams.memoryBuffer[1], maxRequiredMemory - usedMemory, remainingMemory);
}
void AllocateElement(uint currentID)
void AllocateElement(uint currentID, uint dispatchID)
{
// Load the bisector for this element
BisectorData bisectorData = pParams.bisectorDataBuffer[currentID];
@@ -338,7 +338,7 @@ void AllocateElement(uint currentID)
// llocate the bits we need
for (uint bitId = 0; bitId < numSlots; ++bitId)
{
bisectorData.indices[bitId] = decode_bit_complement(firstBitIndex + bitId);
bisectorData.indices[bitId] = decode_bit_complement(firstBitIndex + bitId, dispatchID);
}
// Output