Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[GPU] Increase FC tile_b size for INT4 shape agnostic kernel #23532

Merged
merged 2 commits into from
Mar 21, 2024
Merged

[GPU] Increase FC tile_b size for INT4 shape agnostic kernel #23532

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
57c5d1a
Increase fc tile_b to 16 for INT4 case for improving context processi…
yeonbok Mar 19, 2024
c75f94b
applied code review
yeonbok Mar 20, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 7 additions & 2 deletions ...intel_gpu/src/kernel_selector/kernels/fully_connected/fully_connected_kernel_bf_tiled.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -178,12 +178,13 @@ bool TuneParamsSelector::VerifyTuneParams(const fully_connected_params& params,
return false;

if (tparams.kernel_type == FullyConnected_bf_tiled::KernelType::SLM) {
bool is_i4_u4 = (params.weights.GetDType() == WeightsType::INT4 || params.weights.GetDType() == WeightsType::UINT4);
const auto required_batch_alignment = 64;
if (!params.is_shape_agnostic && (!IsAligned(output_b, required_batch_alignment) || output_b < 256))
return false;

const auto required_tile_b = 8;
if (tparams.tile_b != required_tile_b)
if ((tparams.tile_b != required_tile_b) && !is_i4_u4)
return false;

const auto required_tile_ofm = 2;
Expand Down Expand Up @@ -248,8 +249,12 @@ FullyConnected_bf_tiled::GetAutoTuneParams(const fully_connected_params& params,
} else {
// Try to use SLM kernels if possible
if (preferred_kernel_type != KernelType::DEFAULT) {
if (params.is_shape_agnostic) {
selector.Case(tune_params(16, 2, 2, 4, 1, 1, EXE_MODE_DEFAULT, KernelType::SLM))
.Case(tune_params(16, 2, 1, 4, 1, 1, EXE_MODE_DEFAULT, KernelType::SLM));
}
selector.Case(tune_params(8, 2, 2, 4, 1, 1, EXE_MODE_DEFAULT, KernelType::SLM))
.Case(tune_params(8, 2, 1, 4, 1, 1, EXE_MODE_DEFAULT, KernelType::SLM));
.Case(tune_params(8, 2, 1, 4, 1, 1, EXE_MODE_DEFAULT, KernelType::SLM));
andrew-k-park marked this conversation as resolved.
Show resolved Hide resolved
}
return selector.Default(tune_params(8, 2, 1, 4, 1, 1, EXE_MODE_DEFAULT));
}
Expand Down
Loading