MirrorHub/construct
0
0
Fork 0
mirror of https://github.com/matrix-construct/construct synced 2024-12-27 07:54:05 +01:00
Code Issues Releases Wiki Activity

ircd::simt: Simplify various loop/mask conditionals; minor cleanup.

Browse source
This commit is contained in:
Jason Volk 2021-04-12 15:19:53 -07:00
parent 1e08339955
commit de166dfab8
2 changed files with 11 additions and 17 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

13
include/ircd/simt/mean.h
View file

@ -20,16 +20,11 @@ ircd_simt_math_mean_f4lldr(__local float4 *const restrict out,
ircd_simt_reduce_add_f4lldr(out, num, i); ircd_simt_reduce_add_f4lldr(out, num, i);
if(i == 0) if(i == 0)
{ for(uint k = 1; k < 4; ++k)
float numerator = 0.0f; out[i][0] += out[i][k];
float4 numeratorv = out[i];
for(uint k = 0; k < 4; ++k)
numerator += numeratorv[k];
out[i] = numerator; if(i == 0)
} out[i] = out[i][0] / (num * 4);
ircd_simt_broadcast_f4lldr(out, num, i); ircd_simt_broadcast_f4lldr(out, num, i);
const float4 numeratorv = out[i];
out[i] = numeratorv / (num * 4);
} }

15
include/ircd/simt/sort.h
View file

@ -18,15 +18,14 @@ ircd_simt_sort_idx16_flldr(__local ushort *const idx,
for(uint stride = ln >> 1; stride > 0; stride >>= 1) for(uint stride = ln >> 1; stride > 0; stride >>= 1)
{ {
barrier(CLK_LOCAL_MEM_FENCE); barrier(CLK_LOCAL_MEM_FENCE);
if(li >= stride || val[idx[li]] >= val[idx[li + stride]])
continue;
if(li < stride && val[idx[li]] < val[idx[li + stride]]) const ushort
{ ours = idx[li],
const ushort theirs = idx[li + stride];
ours = idx[li],
theirs = idx[li + stride];
idx[li] = theirs; idx[li] = theirs;
idx[li + stride] = ours; idx[li + stride] = ours;
}
} }
} }