From c4998f09bd02bf525b1b953cbfc6e9cc37b138cc Mon Sep 17 00:00:00 2001
From: MHYangAMD <meng-hsuan.yang@amd.com>
Date: Fri, 30 Jan 2026 10:52:19 +0800
Subject: [PATCH] Fix redundant cast in model sensitive rmsnorm (#3681)

* Fix redundant cast

* Fix linting

[ROCm/composable_kernel commit: 6ff073784321a55ee276f38af195532d8d812670]
---
 .../rmsnorm2d_fwd_pipeline_model_sensitive_pass.hpp    | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)

diff --git a/include/ck_tile/ops/rmsnorm2d/pipeline/rmsnorm2d_fwd_pipeline_model_sensitive_pass.hpp b/include/ck_tile/ops/rmsnorm2d/pipeline/rmsnorm2d_fwd_pipeline_model_sensitive_pass.hpp
index de27b15952..f94d220b94 100644
--- a/include/ck_tile/ops/rmsnorm2d/pipeline/rmsnorm2d_fwd_pipeline_model_sensitive_pass.hpp
+++ b/include/ck_tile/ops/rmsnorm2d/pipeline/rmsnorm2d_fwd_pipeline_model_sensitive_pass.hpp
@@ -181,12 +181,10 @@ struct Rmsnorm2dFwdPipelineModelSensitiveT5Pass
 
             if constexpr(std::is_same_v<XDataType, ck_tile::bf16_t>)
             {
-                const auto tmp0 =
-                    float_to_bf16<bf16_rounding_mode::standard>(acc[idx] * inv_rms_[i_idx]);
-                const auto tmp1 = float_to_bf16<bf16_rounding_mode::standard>(
-                    type_convert<ComputeDataType>(tmp0) * gamma_);
-                const auto rmsn_ = type_convert<ComputeDataType>(tmp1);
-                rmsn(idx)        = rmsn_;
+                const auto tmp      = acc[idx] * inv_rms_[i_idx];
+                const auto tmp_bf16 = float_to_bf16<bf16_rounding_mode::standard>(tmp);
+                const auto rmsn_    = type_convert<ComputeDataType>(tmp_bf16) * gamma_;
+                rmsn(idx)           = rmsn_;
             }
             else
             {