cwpearson/astaroth
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Added commented out pragma unrolls to remind how packing could be improved. Though at the moment unrolls actually make the performance much worse, reasons unknown.

Browse Source
This commit is contained in:
jpekkila
2020-01-22 19:27:45 +02:00
parent 41f8e9aebb
commit 5de163e8d1
1 changed files with 4 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/core/kernels/packing.cu
View File

@@ -53,6 +53,8 @@ kernel_pack_data(const VertexBufferArray vba, const int3 vba_start, PackedData p
k_packed * packed.dims.x * packed.dims.y; k_packed * packed.dims.x * packed.dims.y;
const size_t vtxbuf_offset = packed.dims.x * packed.dims.y * packed.dims.z; const size_t vtxbuf_offset = packed.dims.x * packed.dims.y * packed.dims.z;
//#pragma unroll
for (int i = 0; i < NUM_VTXBUF_HANDLES; ++i) for (int i = 0; i < NUM_VTXBUF_HANDLES; ++i)
packed.data[packed_idx + i * vtxbuf_offset] = vba.in[i][unpacked_idx]; packed.data[packed_idx + i * vtxbuf_offset] = vba.in[i][unpacked_idx];
} }
@@ -82,6 +84,8 @@ kernel_unpack_data(const PackedData packed, const int3 vba_start, VertexBufferAr
k_packed * packed.dims.x * packed.dims.y; k_packed * packed.dims.x * packed.dims.y;
const size_t vtxbuf_offset = packed.dims.x * packed.dims.y * packed.dims.z; const size_t vtxbuf_offset = packed.dims.x * packed.dims.y * packed.dims.z;
//#pragma unroll
for (int i = 0; i < NUM_VTXBUF_HANDLES; ++i) for (int i = 0; i < NUM_VTXBUF_HANDLES; ++i)
vba.in[i][unpacked_idx] = packed.data[packed_idx + i * vtxbuf_offset]; vba.in[i][unpacked_idx] = packed.data[packed_idx + i * vtxbuf_offset];
} }