experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Add changes for P100 and [u]int8.

Browse Source
This commit is contained in:
Frederic Bastien 2020-01-31 10:37:30 -08:00
parent c9878bb912
commit 7d9ababee5
1 changed files with 3 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
tensorflow/compiler/xla/service/gpu/ir_emission_utils.cc
View File

@ -146,9 +146,10 @@ std::array<int64, 3> GetReductionTiling(
&cc_major, &cc_minor); &cc_major, &cc_minor);
} }
int unroll_x = 8; int unroll_x = 8;
if (cc_major >= 7 && smallest_input_dtype_bits <= 16) { if (cc_major >= 6 && smallest_input_dtype_bits == 16) {
// TODO: Benchmark with int8 too.
unroll_x = 32; unroll_x = 32;
} else if (cc_major >= 6 && smallest_input_dtype_bits == 8) {
unroll_x = 64;
} }
return {tile_z, 1, unroll_x}; return {tile_z, 1, unroll_x};
} }