saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-07-29 20:14:01 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

ppc64le: Optimizing the MlasMaximumPool() to use VSX instructions (#11216)

Browse source 
It runs on Power8, Power9, and Power10
This commit is contained in:
Maxiwell 2022-04-19 19:13:55 -03:00 committed by GitHub
parent bab9b80f1f
commit acb555c4c7
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 36 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

36
onnxruntime/core/mlas/lib/pooling.cpp
View file

@ -1538,6 +1538,42 @@ Return Value:
c -= 8;
}
#elif defined(MLAS_TARGET_POWER)
while (c >= 32) {
auto MaximumVector0 = vec_splats(std::numeric_limits<T8Bits>::lowest());
auto MaximumVector1 = vec_splats(std::numeric_limits<T8Bits>::lowest());
for (size_t k = 0; k < KernelSize; k++) {
auto InputVector0 = vec_xl(0, &Input[k][ChannelOffset]);
auto InputVector1 = vec_xl(16, &Input[k][ChannelOffset]);
MaximumVector0 = vec_max(MaximumVector0, InputVector0);
MaximumVector1 = vec_max(MaximumVector1, InputVector1);
}
vec_xst(MaximumVector0, 0, (T8Bits *) Output);
vec_xst(MaximumVector1, 16, (T8Bits *) Output);
Output += 32;
ChannelOffset += 32;
c -= 32;
}
while (c >= 16) {
auto MaximumVector = vec_splats(std::numeric_limits<T8Bits>::lowest());
for (size_t k = 0; k < KernelSize; k++) {
auto InputVector = vec_xl(0, &Input[k][ChannelOffset]);
MaximumVector = vec_max(MaximumVector, InputVector);
}
vec_xst(MaximumVector, 0, (T8Bits *) Output);
Output += 16;
ChannelOffset += 16;
c -= 16;
}
#endif
while (c > 0) {