ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-20 06:49:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

[Ck_tile] hot fix, fix rpcf param setting err (#1657)

Browse Source 
Co-authored-by: dummycoderfe <noplydummmycoder@163.com>
This commit is contained in:
feli
2024-11-14 14:06:36 +08:00
committed by GitHub
parent efd9261545
commit c1f8d53ce8
3 changed files with 21 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
include/ck_tile/ops/layernorm2d/pipeline/layernorm2d_fwd_pipeline_one_pass.hpp
View File

@@ -121,7 +121,7 @@ struct Layernorm2dFwdPipelineOnePass
auto [mean, var] = block_welford(acc, cur_count, max_count);
block_welford_sync(mean, var, cur_count);
block_welford_cross_warp_sync(mean, var, cur_count, smem);
block_tile_welford_post_scale_var(var, cur_count);
block_tile_welford_post_scale_var(var, cur_count, constant<kFastFDiv>{});
// compute inv-std
auto inv_std = tile_elementwise_in(

14
include/ck_tile/ops/layernorm2d/pipeline/layernorm2d_fwd_pipeline_two_pass.hpp
View File

@@ -35,6 +35,7 @@ struct Layernorm2dFwdPipelineTwoPass
static constexpr bool kNeedCrossWarpSync = Problem::kNeedCrossWarpSync;
static constexpr bool kPadM = false; // TODO - BlockLayernorm2dFwdProblem::kPadM
static constexpr bool kPadN = Problem::Traits::kPadN;
static constexpr bool kFastFDiv = Problem::Traits::kFastFDiv;
static constexpr auto kFusedAdd = Problem::Traits::kFusedAdd;
static constexpr auto kFusedQuant = Problem::Traits::kFusedQuant;
@@ -137,15 +138,22 @@ struct Layernorm2dFwdPipelineTwoPass
block_welford_sync(mean, var, cur_count);
block_welford_cross_warp_sync(mean, var, cur_count, smem);
block_tile_welford_post_scale_var(var, cur_count);
block_tile_welford_post_scale_var(var, cur_count, constant<kFastFDiv>{});
// compute inv-std
auto inv_std = tile_elementwise_in(
[&](const auto& v_) {
return type_convert<ComputeDataType>(1.0f) / (sqrt(v_ + epsilon));
if(kFastFDiv && std::is_same_v<ComputeDataType, float>)
{
return type_convert<ComputeDataType>(1.0f) *
__builtin_amdgcn_rcpf(sqrt(v_ + epsilon));
}
else
{
return type_convert<ComputeDataType>(1.0f) / sqrt(v_ + epsilon);
}
},
var);
if constexpr(kSaveMean)
store_tile(mean_window, cast_tile<MeanDataType>(mean));
if constexpr(kSaveInvStd)