From fdb89d3e2fe58a1fb2c6c9931e513a608940c7b8 Mon Sep 17 00:00:00 2001
From: Qianfeng Zhang <Qianfeng.Zhang@amd.com>
Date: Tue, 14 Oct 2025 09:40:23 +0000
Subject: [PATCH] Add instances to consider for adding softmax support

---
 .../18_hstu_attention/generate_instances.py   | 101 ++++---
 .../hstu_attention_batched_forward_bf16.cpp   |  11 +-
 ...stu_attention_batched_forward_dispatch.hpp |  20 +-
 .../hstu_attention_batched_forward_fp16.cpp   |  11 +-
 .../hstu_attention_jagged_forward_bf16.cpp    |  11 +-
 ...hstu_attention_jagged_forward_dispatch.hpp |  20 +-
 .../hstu_attention_jagged_forward_fp16.cpp    |  11 +-
 .../hstu_attention_pipeline_problem.hpp       |   2 +
 ...x_false_has_bias_has_dropout_maxk_128.cpp} |   3 +-
 ...x_false_has_bias_has_dropout_maxk_256.cpp} |   5 +-
 ...ax_false_has_bias_has_dropout_maxk_64.cpp} |   3 +-
 ...ax_false_has_bias_no_dropout_maxk_128.cpp} |   3 +-
 ...ax_false_has_bias_no_dropout_maxk_256.cpp} |   3 +-
 ...max_false_has_bias_no_dropout_maxk_64.cpp} |   3 +-
 ...ax_false_no_bias_has_dropout_maxk_128.cpp} |   3 +-
 ...ax_false_no_bias_has_dropout_maxk_256.cpp} |   3 +-
 ...max_false_no_bias_has_dropout_maxk_64.cpp} |   3 +-
 ...max_false_no_bias_no_dropout_maxk_128.cpp} |   3 +-
 ...max_false_no_bias_no_dropout_maxk_256.cpp} |   3 +-
 ...tmax_false_no_bias_no_dropout_maxk_64.cpp} |   3 +-
 ...ax_true_has_bias_has_dropout_maxk_128.cpp} |   3 +-
 ...ax_true_has_bias_has_dropout_maxk_256.cpp} |   3 +-
 ...max_true_has_bias_has_dropout_maxk_64.cpp} |   3 +-
 ...max_true_has_bias_no_dropout_maxk_128.cpp} |   7 +-
 ...max_true_has_bias_no_dropout_maxk_256.cpp} |   5 +-
 ...tmax_true_has_bias_no_dropout_maxk_64.cpp} |   3 +-
 ...max_true_no_bias_has_dropout_maxk_128.cpp} |   3 +-
 ...max_true_no_bias_has_dropout_maxk_256.cpp} |   5 +-
 ...tmax_true_no_bias_has_dropout_maxk_64.cpp} |   5 +-
 ...tmax_true_no_bias_no_dropout_maxk_128.cpp} |   3 +-
 ...tmax_true_no_bias_no_dropout_maxk_256.cpp} |   3 +-
 ...ftmax_true_no_bias_no_dropout_maxk_64.cpp} |   3 +-
 ...ion_batched_forward_bf16_instances_ref.hpp | 264 ++++++++++++++++--
 ...ax_false_has_bias_has_dropout_maxk_128.cpp |  18 ++
 ...ax_false_has_bias_has_dropout_maxk_256.cpp |  18 ++
 ...max_false_has_bias_has_dropout_maxk_64.cpp |  18 ++
 ...max_false_has_bias_no_dropout_maxk_128.cpp |  18 ++
 ...max_false_has_bias_no_dropout_maxk_256.cpp |  18 ++
 ...tmax_false_has_bias_no_dropout_maxk_64.cpp |  18 ++
 ...max_false_no_bias_has_dropout_maxk_128.cpp |  18 ++
 ...max_false_no_bias_has_dropout_maxk_256.cpp |  18 ++
 ...tmax_false_no_bias_has_dropout_maxk_64.cpp |  18 ++
 ...tmax_false_no_bias_no_dropout_maxk_128.cpp |  18 ++
 ...tmax_false_no_bias_no_dropout_maxk_256.cpp |  18 ++
 ...ftmax_false_no_bias_no_dropout_maxk_64.cpp |  18 ++
 ...max_true_has_bias_has_dropout_maxk_128.cpp |  18 ++
 ...max_true_has_bias_has_dropout_maxk_256.cpp |  18 ++
 ...tmax_true_has_bias_has_dropout_maxk_64.cpp |  18 ++
 ...tmax_true_has_bias_no_dropout_maxk_128.cpp |  18 ++
 ...tmax_true_has_bias_no_dropout_maxk_256.cpp |  18 ++
 ...ftmax_true_has_bias_no_dropout_maxk_64.cpp |  18 ++
 ...tmax_true_no_bias_has_dropout_maxk_128.cpp |  18 ++
 ...tmax_true_no_bias_has_dropout_maxk_256.cpp |  18 ++
 ...ftmax_true_no_bias_has_dropout_maxk_64.cpp |  18 ++
 ...ftmax_true_no_bias_no_dropout_maxk_128.cpp |  18 ++
 ...ftmax_true_no_bias_no_dropout_maxk_256.cpp |  18 ++
 ...oftmax_true_no_bias_no_dropout_maxk_64.cpp |  18 ++
 ...x_false_has_bias_has_dropout_maxk_128.cpp} |   3 +-
 ...x_false_has_bias_has_dropout_maxk_256.cpp} |   5 +-
 ...ax_false_has_bias_has_dropout_maxk_64.cpp} |   3 +-
 ...ax_false_has_bias_no_dropout_maxk_128.cpp} |   3 +-
 ...ax_false_has_bias_no_dropout_maxk_256.cpp} |   3 +-
 ...max_false_has_bias_no_dropout_maxk_64.cpp} |   3 +-
 ...ax_false_no_bias_has_dropout_maxk_128.cpp} |   3 +-
 ...ax_false_no_bias_has_dropout_maxk_256.cpp} |   3 +-
 ...max_false_no_bias_has_dropout_maxk_64.cpp} |   3 +-
 ...max_false_no_bias_no_dropout_maxk_128.cpp} |   3 +-
 ...max_false_no_bias_no_dropout_maxk_256.cpp} |   3 +-
 ...tmax_false_no_bias_no_dropout_maxk_64.cpp} |   3 +-
 ...ax_true_has_bias_has_dropout_maxk_128.cpp} |   3 +-
 ...ax_true_has_bias_has_dropout_maxk_256.cpp} |   3 +-
 ...max_true_has_bias_has_dropout_maxk_64.cpp} |   3 +-
 ...max_true_has_bias_no_dropout_maxk_128.cpp} |   7 +-
 ...max_true_has_bias_no_dropout_maxk_256.cpp} |   5 +-
 ...tmax_true_has_bias_no_dropout_maxk_64.cpp} |   3 +-
 ...max_true_no_bias_has_dropout_maxk_128.cpp} |   3 +-
 ...max_true_no_bias_has_dropout_maxk_256.cpp} |   5 +-
 ...tmax_true_no_bias_has_dropout_maxk_64.cpp} |   5 +-
 ...tmax_true_no_bias_no_dropout_maxk_128.cpp} |   3 +-
 ...tmax_true_no_bias_no_dropout_maxk_256.cpp} |   3 +-
 ...ftmax_true_no_bias_no_dropout_maxk_64.cpp} |   7 +-
 ...ion_batched_forward_fp16_instances_ref.hpp | 264 ++++++++++++++++--
 ...ax_false_has_bias_has_dropout_maxk_128.cpp |  18 ++
 ...ax_false_has_bias_has_dropout_maxk_256.cpp |  18 ++
 ...max_false_has_bias_has_dropout_maxk_64.cpp |  18 ++
 ...max_false_has_bias_no_dropout_maxk_128.cpp |  18 ++
 ...max_false_has_bias_no_dropout_maxk_256.cpp |  18 ++
 ...tmax_false_has_bias_no_dropout_maxk_64.cpp |  18 ++
 ...max_false_no_bias_has_dropout_maxk_128.cpp |  18 ++
 ...max_false_no_bias_has_dropout_maxk_256.cpp |  18 ++
 ...tmax_false_no_bias_has_dropout_maxk_64.cpp |  18 ++
 ...tmax_false_no_bias_no_dropout_maxk_128.cpp |  18 ++
 ...tmax_false_no_bias_no_dropout_maxk_256.cpp |  18 ++
 ...ftmax_false_no_bias_no_dropout_maxk_64.cpp |  18 ++
 ...max_true_has_bias_has_dropout_maxk_128.cpp |  18 ++
 ...max_true_has_bias_has_dropout_maxk_256.cpp |  18 ++
 ...tmax_true_has_bias_has_dropout_maxk_64.cpp |  18 ++
 ...tmax_true_has_bias_no_dropout_maxk_128.cpp |  18 ++
 ...tmax_true_has_bias_no_dropout_maxk_256.cpp |  18 ++
 ...ftmax_true_has_bias_no_dropout_maxk_64.cpp |  18 ++
 ...tmax_true_no_bias_has_dropout_maxk_128.cpp |  18 ++
 ...tmax_true_no_bias_has_dropout_maxk_256.cpp |  18 ++
 ...ftmax_true_no_bias_has_dropout_maxk_64.cpp |  18 ++
 ...ftmax_true_no_bias_no_dropout_maxk_128.cpp |  18 ++
 ...ftmax_true_no_bias_no_dropout_maxk_256.cpp |  18 ++
 ...oftmax_true_no_bias_no_dropout_maxk_64.cpp |  18 ++
 ...x_false_has_bias_has_dropout_maxk_128.cpp} |   5 +-
 ...x_false_has_bias_has_dropout_maxk_256.cpp} |   5 +-
 ...ax_false_has_bias_has_dropout_maxk_64.cpp} |   5 +-
 ...ax_false_has_bias_no_dropout_maxk_128.cpp} |   3 +-
 ...ax_false_has_bias_no_dropout_maxk_256.cpp} |   3 +-
 ...max_false_has_bias_no_dropout_maxk_64.cpp} |   3 +-
 ...ax_false_no_bias_has_dropout_maxk_128.cpp} |   3 +-
 ...ax_false_no_bias_has_dropout_maxk_256.cpp} |   3 +-
 ...max_false_no_bias_has_dropout_maxk_64.cpp} |   3 +-
 ...max_false_no_bias_no_dropout_maxk_128.cpp} |   3 +-
 ...max_false_no_bias_no_dropout_maxk_256.cpp} |   3 +-
 ...tmax_false_no_bias_no_dropout_maxk_64.cpp} |   3 +-
 ...ax_true_has_bias_has_dropout_maxk_128.cpp} |   3 +-
 ...ax_true_has_bias_has_dropout_maxk_256.cpp} |   3 +-
 ...max_true_has_bias_has_dropout_maxk_64.cpp} |   3 +-
 ...max_true_has_bias_no_dropout_maxk_128.cpp} |   5 +-
 ...max_true_has_bias_no_dropout_maxk_256.cpp} |   5 +-
 ...tmax_true_has_bias_no_dropout_maxk_64.cpp} |   5 +-
 ...max_true_no_bias_has_dropout_maxk_128.cpp} |   5 +-
 ...max_true_no_bias_has_dropout_maxk_256.cpp} |   5 +-
 ...tmax_true_no_bias_has_dropout_maxk_64.cpp} |   5 +-
 ...tmax_true_no_bias_no_dropout_maxk_128.cpp} |   3 +-
 ...tmax_true_no_bias_no_dropout_maxk_256.cpp} |   3 +-
 ...ftmax_true_no_bias_no_dropout_maxk_64.cpp} |   3 +-
 ...tion_jagged_forward_bf16_instances_ref.hpp | 264 ++++++++++++++++--
 ...ax_false_has_bias_has_dropout_maxk_128.cpp |  18 ++
 ...ax_false_has_bias_has_dropout_maxk_256.cpp |  18 ++
 ...max_false_has_bias_has_dropout_maxk_64.cpp |  18 ++
 ...max_false_has_bias_no_dropout_maxk_128.cpp |  18 ++
 ...max_false_has_bias_no_dropout_maxk_256.cpp |  18 ++
 ...tmax_false_has_bias_no_dropout_maxk_64.cpp |  18 ++
 ...max_false_no_bias_has_dropout_maxk_128.cpp |  18 ++
 ...max_false_no_bias_has_dropout_maxk_256.cpp |  18 ++
 ...tmax_false_no_bias_has_dropout_maxk_64.cpp |  18 ++
 ...tmax_false_no_bias_no_dropout_maxk_128.cpp |  18 ++
 ...tmax_false_no_bias_no_dropout_maxk_256.cpp |  18 ++
 ...ftmax_false_no_bias_no_dropout_maxk_64.cpp |  18 ++
 ...max_true_has_bias_has_dropout_maxk_128.cpp |  18 ++
 ...max_true_has_bias_has_dropout_maxk_256.cpp |  18 ++
 ...tmax_true_has_bias_has_dropout_maxk_64.cpp |  18 ++
 ...tmax_true_has_bias_no_dropout_maxk_128.cpp |  18 ++
 ...tmax_true_has_bias_no_dropout_maxk_256.cpp |  18 ++
 ...ftmax_true_has_bias_no_dropout_maxk_64.cpp |  18 ++
 ...tmax_true_no_bias_has_dropout_maxk_128.cpp |  18 ++
 ...tmax_true_no_bias_has_dropout_maxk_256.cpp |  18 ++
 ...ftmax_true_no_bias_has_dropout_maxk_64.cpp |  18 ++
 ...ftmax_true_no_bias_no_dropout_maxk_128.cpp |  18 ++
 ...ftmax_true_no_bias_no_dropout_maxk_256.cpp |  18 ++
 ...oftmax_true_no_bias_no_dropout_maxk_64.cpp |  18 ++
 ...x_false_has_bias_has_dropout_maxk_128.cpp} |   5 +-
 ...x_false_has_bias_has_dropout_maxk_256.cpp} |   5 +-
 ...ax_false_has_bias_has_dropout_maxk_64.cpp} |   5 +-
 ...ax_false_has_bias_no_dropout_maxk_128.cpp} |   3 +-
 ...ax_false_has_bias_no_dropout_maxk_256.cpp} |   3 +-
 ...max_false_has_bias_no_dropout_maxk_64.cpp} |   3 +-
 ...ax_false_no_bias_has_dropout_maxk_128.cpp} |   3 +-
 ...ax_false_no_bias_has_dropout_maxk_256.cpp} |   3 +-
 ...max_false_no_bias_has_dropout_maxk_64.cpp} |   3 +-
 ...max_false_no_bias_no_dropout_maxk_128.cpp} |   3 +-
 ...max_false_no_bias_no_dropout_maxk_256.cpp} |   3 +-
 ...tmax_false_no_bias_no_dropout_maxk_64.cpp} |   3 +-
 ...ax_true_has_bias_has_dropout_maxk_128.cpp} |   3 +-
 ...ax_true_has_bias_has_dropout_maxk_256.cpp} |   3 +-
 ...max_true_has_bias_has_dropout_maxk_64.cpp} |   3 +-
 ...max_true_has_bias_no_dropout_maxk_128.cpp} |   5 +-
 ...max_true_has_bias_no_dropout_maxk_256.cpp} |   5 +-
 ...tmax_true_has_bias_no_dropout_maxk_64.cpp} |   5 +-
 ...max_true_no_bias_has_dropout_maxk_128.cpp} |   5 +-
 ...max_true_no_bias_has_dropout_maxk_256.cpp} |   5 +-
 ...tmax_true_no_bias_has_dropout_maxk_64.cpp} |   5 +-
 ...tmax_true_no_bias_no_dropout_maxk_128.cpp} |   3 +-
 ...tmax_true_no_bias_no_dropout_maxk_256.cpp} |   3 +-
 ...ftmax_true_no_bias_no_dropout_maxk_64.cpp} |   3 +-
 ...tion_jagged_forward_fp16_instances_ref.hpp | 264 ++++++++++++++++--
 ...ax_false_has_bias_has_dropout_maxk_128.cpp |  18 ++
 ...ax_false_has_bias_has_dropout_maxk_256.cpp |  18 ++
 ...max_false_has_bias_has_dropout_maxk_64.cpp |  18 ++
 ...max_false_has_bias_no_dropout_maxk_128.cpp |  18 ++
 ...max_false_has_bias_no_dropout_maxk_256.cpp |  18 ++
 ...tmax_false_has_bias_no_dropout_maxk_64.cpp |  18 ++
 ...max_false_no_bias_has_dropout_maxk_128.cpp |  18 ++
 ...max_false_no_bias_has_dropout_maxk_256.cpp |  18 ++
 ...tmax_false_no_bias_has_dropout_maxk_64.cpp |  18 ++
 ...tmax_false_no_bias_no_dropout_maxk_128.cpp |  18 ++
 ...tmax_false_no_bias_no_dropout_maxk_256.cpp |  18 ++
 ...ftmax_false_no_bias_no_dropout_maxk_64.cpp |  18 ++
 ...max_true_has_bias_has_dropout_maxk_128.cpp |  18 ++
 ...max_true_has_bias_has_dropout_maxk_256.cpp |  18 ++
 ...tmax_true_has_bias_has_dropout_maxk_64.cpp |  18 ++
 ...tmax_true_has_bias_no_dropout_maxk_128.cpp |  18 ++
 ...tmax_true_has_bias_no_dropout_maxk_256.cpp |  18 ++
 ...ftmax_true_has_bias_no_dropout_maxk_64.cpp |  18 ++
 ...tmax_true_no_bias_has_dropout_maxk_128.cpp |  18 ++
 ...tmax_true_no_bias_has_dropout_maxk_256.cpp |  18 ++
 ...ftmax_true_no_bias_has_dropout_maxk_64.cpp |  18 ++
 ...ftmax_true_no_bias_no_dropout_maxk_128.cpp |  18 ++
 ...ftmax_true_no_bias_no_dropout_maxk_256.cpp |  18 ++
 ...oftmax_true_no_bias_no_dropout_maxk_64.cpp |  18 ++
 204 files changed, 3018 insertions(+), 305 deletions(-)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_has_causal_no_bias_has_dropout_maxk_128.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_has_causal_has_bias_no_dropout_maxk_256.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_has_causal_no_bias_has_dropout_maxk_64.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_no_causal_has_bias_no_dropout_maxk_128.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_no_causal_has_bias_no_dropout_maxk_256.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_no_causal_has_bias_no_dropout_maxk_64.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_no_causal_no_bias_has_dropout_maxk_128.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_no_causal_no_bias_has_dropout_maxk_256.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_no_causal_no_bias_has_dropout_maxk_64.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_no_causal_no_bias_no_dropout_maxk_128.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_no_causal_no_bias_no_dropout_maxk_256.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_no_causal_no_bias_no_dropout_maxk_64.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_has_causal_has_bias_has_dropout_maxk_128.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_has_causal_has_bias_has_dropout_maxk_256.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_has_causal_has_bias_has_dropout_maxk_64.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_no_causal_has_bias_has_dropout_maxk_128.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_has_causal_no_bias_has_dropout_maxk_256.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_has_causal_has_bias_no_dropout_maxk_64.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_has_causal_has_bias_no_dropout_maxk_128.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_no_causal_has_bias_has_dropout_maxk_256.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_no_causal_has_bias_has_dropout_maxk_64.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_has_causal_no_bias_no_dropout_maxk_128.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_has_causal_no_bias_no_dropout_maxk_256.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_bf16_has_causal_no_bias_no_dropout_maxk_64.cpp => hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp} (85%)
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_has_causal_no_bias_has_dropout_maxk_128.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_has_causal_has_bias_no_dropout_maxk_256.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_has_causal_no_bias_has_dropout_maxk_64.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_no_causal_has_bias_no_dropout_maxk_128.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_no_causal_has_bias_no_dropout_maxk_256.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_no_causal_has_bias_no_dropout_maxk_64.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_no_causal_no_bias_has_dropout_maxk_128.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_no_causal_no_bias_has_dropout_maxk_256.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_has_causal_no_bias_no_dropout_maxk_64.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_no_causal_no_bias_no_dropout_maxk_128.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_no_causal_no_bias_no_dropout_maxk_256.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_no_causal_no_bias_no_dropout_maxk_64.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_has_causal_has_bias_has_dropout_maxk_128.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_has_causal_has_bias_has_dropout_maxk_256.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_has_causal_has_bias_has_dropout_maxk_64.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_no_causal_has_bias_has_dropout_maxk_128.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_has_causal_no_bias_has_dropout_maxk_256.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_has_causal_has_bias_no_dropout_maxk_64.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_has_causal_has_bias_no_dropout_maxk_128.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_no_causal_has_bias_has_dropout_maxk_256.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_no_causal_has_bias_has_dropout_maxk_64.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_has_causal_no_bias_no_dropout_maxk_128.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_has_causal_no_bias_no_dropout_maxk_256.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_batched_forward_fp16_no_causal_no_bias_has_dropout_maxk_64.cpp => hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp} (85%)
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_has_causal_has_bias_no_dropout_maxk_128.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_has_causal_has_bias_no_dropout_maxk_256.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_has_causal_has_bias_no_dropout_maxk_64.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_no_causal_has_bias_no_dropout_maxk_128.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_no_causal_has_bias_no_dropout_maxk_256.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_no_causal_has_bias_no_dropout_maxk_64.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_no_causal_no_bias_has_dropout_maxk_128.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_no_causal_no_bias_has_dropout_maxk_256.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_no_causal_no_bias_has_dropout_maxk_64.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_no_causal_no_bias_no_dropout_maxk_128.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_no_causal_no_bias_no_dropout_maxk_256.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_no_causal_no_bias_no_dropout_maxk_64.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_has_causal_has_bias_has_dropout_maxk_128.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_has_causal_has_bias_has_dropout_maxk_256.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_has_causal_has_bias_has_dropout_maxk_64.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_has_causal_no_bias_has_dropout_maxk_128.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_has_causal_no_bias_has_dropout_maxk_256.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_has_causal_no_bias_has_dropout_maxk_64.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_no_causal_has_bias_has_dropout_maxk_128.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_no_causal_has_bias_has_dropout_maxk_256.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_no_causal_has_bias_has_dropout_maxk_64.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_has_causal_no_bias_no_dropout_maxk_128.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_has_causal_no_bias_no_dropout_maxk_256.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_bf16_has_causal_no_bias_no_dropout_maxk_64.cpp => hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp} (85%)
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_has_causal_has_bias_no_dropout_maxk_128.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_has_causal_has_bias_no_dropout_maxk_256.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_has_causal_has_bias_no_dropout_maxk_64.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_no_causal_has_bias_no_dropout_maxk_128.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_no_causal_has_bias_no_dropout_maxk_256.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_no_causal_has_bias_no_dropout_maxk_64.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_no_causal_no_bias_has_dropout_maxk_128.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_no_causal_no_bias_has_dropout_maxk_256.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_no_causal_no_bias_has_dropout_maxk_64.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_no_causal_no_bias_no_dropout_maxk_128.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_no_causal_no_bias_no_dropout_maxk_256.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_no_causal_no_bias_no_dropout_maxk_64.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_has_causal_has_bias_has_dropout_maxk_128.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_has_causal_has_bias_has_dropout_maxk_256.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_has_causal_has_bias_has_dropout_maxk_64.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_has_causal_no_bias_has_dropout_maxk_128.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_has_causal_no_bias_has_dropout_maxk_256.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_has_causal_no_bias_has_dropout_maxk_64.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_no_causal_has_bias_has_dropout_maxk_128.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_no_causal_has_bias_has_dropout_maxk_256.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_no_causal_has_bias_has_dropout_maxk_64.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_has_causal_no_bias_no_dropout_maxk_128.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_has_causal_no_bias_no_dropout_maxk_256.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp} (85%)
 rename example/ck_tile/18_hstu_attention/instances/{hstu_attention_jagged_forward_fp16_has_causal_no_bias_no_dropout_maxk_64.cpp => hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp} (85%)
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
 create mode 100644 example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp

diff --git a/example/ck_tile/18_hstu_attention/generate_instances.py b/example/ck_tile/18_hstu_attention/generate_instances.py
index f085997e42..ca752cc3db 100644
--- a/example/ck_tile/18_hstu_attention/generate_instances.py
+++ b/example/ck_tile/18_hstu_attention/generate_instances.py
@@ -26,16 +26,17 @@ HSTU_FORWARD_INSTANCE_TEMPLATE_INC = """
 """
 
 HSTU_FORWARD_INSTANCE_TEMPLATE = """
-{extern}template void run_{mode}_forward_causal_bias_dropout_dispatch<
+{extern}template void run_{mode}_forward_causal_softmax_bias_dropout_dispatch<
     {dtype},
     {has_causal},
+    {use_softmax},
     {has_bias},
     {has_dropout},
     {max_k}>(HstuAttentionFwdParams& param, hipStream_t stream);
 """
 
 HSTU_FORWARD_INSTANCE_FNAME = (
-    "hstu_attention_{mode}_forward_{dtype_str}_{has_or_no_causal_str}_"
+    "hstu_attention_{mode}_forward_{dtype_str}_{has_or_no_causal_str}_{use_softmax_or_not_str}_"
     "{has_or_no_bias_str}_{has_or_no_dropout_str}_{max_k_str}.cpp"
 )
 
@@ -48,6 +49,11 @@ BOOL_MAP_CAUSAL = {
     False: "no_causal",
 }
 
+BOOL_MAP_SOFTMAX = {
+    True: "softmax_true",
+    False: "softmax_false",
+}
+
 BOOL_MAP_BIAS = {
     True: "has_bias",
     False: "no_bias",
@@ -79,38 +85,41 @@ def create_forward_instances(instance_dir: Path, headdims: List) -> None:
     for mode in ["batched", "jagged"]:
         for dtype in ["fp16", "bf16"]:
             for has_causal in [True, False]:
-                for has_bias in [True, False]:
-                    for has_dropout in [True, False]:
-                        for max_k in headdims:
-                            fname = HSTU_FORWARD_INSTANCE_FNAME.format(
-                                mode=mode,
-                                dtype_str=dtype,
-                                has_or_no_causal_str=BOOL_MAP_CAUSAL[has_causal],
-                                has_or_no_bias_str=BOOL_MAP_BIAS[has_bias],
-                                has_or_no_dropout_str=BOOL_MAP_DROPOUT[has_dropout],
-                                max_k_str=INT_MAP_MAX_K[max_k],
-                            )
-                            forward_instance_inc = (
-                                HSTU_FORWARD_INSTANCE_TEMPLATE_INC.format(
+                for use_softmax in [True, False]:
+                    for has_bias in [True, False]:
+                        for has_dropout in [True, False]:
+                            for max_k in headdims:
+                                fname = HSTU_FORWARD_INSTANCE_FNAME.format(
                                     mode=mode,
-                                    dtype_file=TYPE_FNAME_MAP[dtype],
+                                    dtype_str=dtype,
+                                    has_or_no_causal_str=BOOL_MAP_CAUSAL[has_causal],
+                                    use_softmax_or_not_str=BOOL_MAP_SOFTMAX[use_softmax],
+                                    has_or_no_bias_str=BOOL_MAP_BIAS[has_bias],
+                                    has_or_no_dropout_str=BOOL_MAP_DROPOUT[has_dropout],
+                                    max_k_str=INT_MAP_MAX_K[max_k],
                                 )
-                            )
-                            forward_instance = HSTU_FORWARD_INSTANCE_TEMPLATE.format(
-                                extern="",
-                                mode=mode,
-                                dtype=TYPE_CTYPE_MAP[dtype],
-                                has_causal=BOOL_MAP[has_causal],
-                                has_bias=BOOL_MAP[has_bias],
-                                has_dropout=BOOL_MAP[has_dropout],
-                                max_k=max_k,
-                                cap_mode=MODE_NAME_MAP[mode],
-                            )
-                            (instance_dir / fname).write_text(
-                                HSTU_COPYRIGHT_HEADER
-                                + forward_instance_inc
-                                + forward_instance
-                            )
+                                forward_instance_inc = (
+                                    HSTU_FORWARD_INSTANCE_TEMPLATE_INC.format(
+                                        mode=mode,
+                                        dtype_file=TYPE_FNAME_MAP[dtype],
+                                    )
+                                )
+                                forward_instance = HSTU_FORWARD_INSTANCE_TEMPLATE.format(
+                                    extern="",
+                                    mode=mode,
+                                    dtype=TYPE_CTYPE_MAP[dtype],
+                                    has_causal=BOOL_MAP[has_causal],
+                                    use_softmax=BOOL_MAP[use_softmax],
+                                    has_bias=BOOL_MAP[has_bias],
+                                    has_dropout=BOOL_MAP[has_dropout],
+                                    max_k=max_k,
+                                    cap_mode=MODE_NAME_MAP[mode],
+                                )
+                                (instance_dir / fname).write_text(
+                                     HSTU_COPYRIGHT_HEADER
+                                     + forward_instance_inc
+                                     + forward_instance
+                                )    
 
 
 def create_forward_instances_ref(instance_dir: Path, headdims: List) -> None:
@@ -133,19 +142,21 @@ def create_forward_instances_ref(instance_dir: Path, headdims: List) -> None:
                     for has_bias in [True, False]:
                         for has_dropout in [True, False]:
                             for has_causal in [True, False]:
-                                forward_instance = (
-                                    HSTU_FORWARD_INSTANCE_TEMPLATE.format(
-                                        extern="extern ",
-                                        mode=mode,
-                                        dtype=TYPE_CTYPE_MAP[dtype],
-                                        has_causal=BOOL_MAP[has_causal],
-                                        has_bias=BOOL_MAP[has_bias],
-                                        has_dropout=BOOL_MAP[has_dropout],
-                                        max_k=max_k,
-                                        cap_mode=MODE_NAME_MAP[mode],
-                                    )
-                                )
-                                file.write(forward_instance)
+                                for use_softmax in [True, False]:
+                                    forward_instance = (
+                                        HSTU_FORWARD_INSTANCE_TEMPLATE.format(
+                                            extern="extern ",
+                                            mode=mode,
+                                            dtype=TYPE_CTYPE_MAP[dtype],
+                                            has_causal=BOOL_MAP[has_causal],
+                                            use_softmax=BOOL_MAP[use_softmax],
+                                            has_bias=BOOL_MAP[has_bias],
+                                            has_dropout=BOOL_MAP[has_dropout],
+                                            max_k=max_k,
+                                            cap_mode=MODE_NAME_MAP[mode],
+                                        )
+                                    )   
+                                    file.write(forward_instance)
 
 if __name__ == "__main__":
     headdims_fwd = [64, 128, 256]
diff --git a/example/ck_tile/18_hstu_attention/hstu_attention_batched_forward_bf16.cpp b/example/ck_tile/18_hstu_attention/hstu_attention_batched_forward_bf16.cpp
index 490d55784a..764fa96c42 100644
--- a/example/ck_tile/18_hstu_attention/hstu_attention_batched_forward_bf16.cpp
+++ b/example/ck_tile/18_hstu_attention/hstu_attention_batched_forward_bf16.cpp
@@ -17,11 +17,12 @@ void hstu_attention_batched_forward_bf16(HstuAttentionFwdParams& param, hipStrea
     const bool use_causal  = param.use_causal;
     BOOL_SWITCH_3(has_bias, kHasBias, has_dropout, kHasDropout, use_causal, kUseCausal, [&] {
         HDIM_SWITCH(param.hdim_qk, param.hdim_v, MaxK, [&] {
-            run_batched_forward_causal_bias_dropout_dispatch<ck_tile::bf16_t,
-                                                             kUseCausal,
-                                                             kHasBias,
-                                                             kHasDropout,
-                                                             MaxK>(param, stream);
+            run_batched_forward_causal_softmax_bias_dropout_dispatch<ck_tile::bf16_t,
+                                                                     kUseCausal,
+                                                                     false, // using softmax
+                                                                     kHasBias,
+                                                                     kHasDropout,
+                                                                     MaxK>(param, stream);
         });
     });
 };
diff --git a/example/ck_tile/18_hstu_attention/hstu_attention_batched_forward_dispatch.hpp b/example/ck_tile/18_hstu_attention/hstu_attention_batched_forward_dispatch.hpp
index b5dbba7cd8..c64bcb1a93 100644
--- a/example/ck_tile/18_hstu_attention/hstu_attention_batched_forward_dispatch.hpp
+++ b/example/ck_tile/18_hstu_attention/hstu_attention_batched_forward_dispatch.hpp
@@ -21,10 +21,11 @@
 
 template <typename InOutDataType,
           bool kUseCausal,
+          bool kUseSoftmax,
           bool kHasBias,
           bool kHasDropout,
           ck_tile::index_t MaxK>
-struct batched_forward_causal_bias_dropout_dispatch
+struct batched_forward_causal_softmax_bias_dropout_dispatch
 {
     using HstuAttentionTileSetting = typename HstuAttentionFwdTileSetting<MaxK>::Type;
 
@@ -38,6 +39,7 @@ struct batched_forward_causal_bias_dropout_dispatch
         kHasBias,
         kHasDropout,
         kUseCausal,
+        kUseSoftmax,
         HstuAttentionTileSetting,
         HstuTraits>;
 
@@ -137,15 +139,17 @@ struct batched_forward_causal_bias_dropout_dispatch
 
 template <typename InOutDataType,
           bool kUseCausal,
+          bool kUseSoftmax,
           bool kHasBias,
           bool kHasDropout,
           ck_tile::index_t MaxK>
-void run_batched_forward_causal_bias_dropout_dispatch(HstuAttentionFwdParams& param,
-                                                      hipStream_t stream)
+void run_batched_forward_causal_softmax_bias_dropout_dispatch(HstuAttentionFwdParams& param,
+                                                              hipStream_t stream)
 {
-    batched_forward_causal_bias_dropout_dispatch<InOutDataType,
-                                                 kUseCausal,
-                                                 kHasBias,
-                                                 kHasDropout,
-                                                 MaxK>::Run(param, stream);
+    batched_forward_causal_softmax_bias_dropout_dispatch<InOutDataType,
+                                                         kUseCausal,
+                                                         kUseSoftmax,
+                                                         kHasBias,
+                                                         kHasDropout,
+                                                         MaxK>::Run(param, stream);
 };
diff --git a/example/ck_tile/18_hstu_attention/hstu_attention_batched_forward_fp16.cpp b/example/ck_tile/18_hstu_attention/hstu_attention_batched_forward_fp16.cpp
index a75ada209a..75575e151e 100644
--- a/example/ck_tile/18_hstu_attention/hstu_attention_batched_forward_fp16.cpp
+++ b/example/ck_tile/18_hstu_attention/hstu_attention_batched_forward_fp16.cpp
@@ -17,11 +17,12 @@ void hstu_attention_batched_forward_fp16(HstuAttentionFwdParams& param, hipStrea
     const bool use_causal  = param.use_causal;
     BOOL_SWITCH_3(has_bias, kHasBias, has_dropout, kHasDropout, use_causal, kUseCausal, [&] {
         HDIM_SWITCH(param.hdim_qk, param.hdim_v, MaxK, [&] {
-            run_batched_forward_causal_bias_dropout_dispatch<ck_tile::fp16_t,
-                                                             kUseCausal,
-                                                             kHasBias,
-                                                             kHasDropout,
-                                                             MaxK>(param, stream);
+            run_batched_forward_causal_softmax_bias_dropout_dispatch<ck_tile::fp16_t,
+                                                                     kUseCausal,
+                                                                     false, // using softmax
+                                                                     kHasBias,
+                                                                     kHasDropout,
+                                                                     MaxK>(param, stream);
         });
     });
 };
diff --git a/example/ck_tile/18_hstu_attention/hstu_attention_jagged_forward_bf16.cpp b/example/ck_tile/18_hstu_attention/hstu_attention_jagged_forward_bf16.cpp
index 723c14921a..0d07914ff3 100644
--- a/example/ck_tile/18_hstu_attention/hstu_attention_jagged_forward_bf16.cpp
+++ b/example/ck_tile/18_hstu_attention/hstu_attention_jagged_forward_bf16.cpp
@@ -17,11 +17,12 @@ void hstu_attention_jagged_forward_bf16(HstuAttentionFwdParams& param, hipStream
     const bool use_causal  = param.use_causal;
     BOOL_SWITCH_3(has_bias, kHasBias, has_dropout, kHasDropout, use_causal, kUseCausal, [&] {
         HDIM_SWITCH(param.hdim_qk, param.hdim_v, MaxK, [&] {
-            run_jagged_forward_causal_bias_dropout_dispatch<ck_tile::bf16_t,
-                                                            kUseCausal,
-                                                            kHasBias,
-                                                            kHasDropout,
-                                                            MaxK>(param, stream);
+            run_jagged_forward_causal_softmax_bias_dropout_dispatch<ck_tile::bf16_t,
+                                                                    kUseCausal,
+                                                                    false, // using softmax
+                                                                    kHasBias,
+                                                                    kHasDropout,
+                                                                    MaxK>(param, stream);
         });
     });
 };
diff --git a/example/ck_tile/18_hstu_attention/hstu_attention_jagged_forward_dispatch.hpp b/example/ck_tile/18_hstu_attention/hstu_attention_jagged_forward_dispatch.hpp
index f885c75d7a..2165cd79ec 100644
--- a/example/ck_tile/18_hstu_attention/hstu_attention_jagged_forward_dispatch.hpp
+++ b/example/ck_tile/18_hstu_attention/hstu_attention_jagged_forward_dispatch.hpp
@@ -21,10 +21,11 @@
 
 template <typename InOutDataType,
           bool kUseCausal,
+          bool kUseSoftmax,
           bool kHasBias,
           bool kHasDropout,
           ck_tile::index_t MaxK>
-struct jagged_forward_causal_bias_dropout_dispatch
+struct jagged_forward_causal_softmax_bias_dropout_dispatch
 {
     using HstuAttentionTileSetting = typename HstuAttentionFwdTileSetting<MaxK>::Type;
 
@@ -38,6 +39,7 @@ struct jagged_forward_causal_bias_dropout_dispatch
         kHasBias,
         kHasDropout,
         kUseCausal,
+        kUseSoftmax,
         HstuAttentionTileSetting,
         HstuTraits>;
 
@@ -128,15 +130,17 @@ struct jagged_forward_causal_bias_dropout_dispatch
 
 template <typename InOutDataType,
           bool kUseCausal,
+          bool kUseSoftmax,
           bool kHasBias,
           bool kHasDropout,
           ck_tile::index_t MaxK>
-void run_jagged_forward_causal_bias_dropout_dispatch(HstuAttentionFwdParams& param,
-                                                     hipStream_t stream)
+void run_jagged_forward_causal_softmax_bias_dropout_dispatch(HstuAttentionFwdParams& param,
+                                                             hipStream_t stream)
 {
-    jagged_forward_causal_bias_dropout_dispatch<InOutDataType,
-                                                kUseCausal,
-                                                kHasBias,
-                                                kHasDropout,
-                                                MaxK>::Run(param, stream);
+    jagged_forward_causal_softmax_bias_dropout_dispatch<InOutDataType,
+                                                        kUseCausal,
+                                                        kUseSoftmax,
+                                                        kHasBias,
+                                                        kHasDropout,
+                                                        MaxK>::Run(param, stream);
 };
diff --git a/example/ck_tile/18_hstu_attention/hstu_attention_jagged_forward_fp16.cpp b/example/ck_tile/18_hstu_attention/hstu_attention_jagged_forward_fp16.cpp
index 3d80cf9bfd..6cf51f502e 100644
--- a/example/ck_tile/18_hstu_attention/hstu_attention_jagged_forward_fp16.cpp
+++ b/example/ck_tile/18_hstu_attention/hstu_attention_jagged_forward_fp16.cpp
@@ -17,11 +17,12 @@ void hstu_attention_jagged_forward_fp16(HstuAttentionFwdParams& param, hipStream
     const bool use_causal  = param.use_causal;
     BOOL_SWITCH_3(has_bias, kHasBias, has_dropout, kHasDropout, use_causal, kUseCausal, [&] {
         HDIM_SWITCH(param.hdim_qk, param.hdim_v, MaxK, [&] {
-            run_jagged_forward_causal_bias_dropout_dispatch<ck_tile::fp16_t,
-                                                            kUseCausal,
-                                                            kHasBias,
-                                                            kHasDropout,
-                                                            MaxK>(param, stream);
+            run_jagged_forward_causal_softmax_bias_dropout_dispatch<ck_tile::fp16_t,
+                                                                    kUseCausal,
+                                                                    false, // using softmax
+                                                                    kHasBias,
+                                                                    kHasDropout,
+                                                                    MaxK>(param, stream);
         });
     });
 };
diff --git a/example/ck_tile/18_hstu_attention/hstu_attention_pipeline_problem.hpp b/example/ck_tile/18_hstu_attention/hstu_attention_pipeline_problem.hpp
index ec68dc1558..6bd1c71dd0 100644
--- a/example/ck_tile/18_hstu_attention/hstu_attention_pipeline_problem.hpp
+++ b/example/ck_tile/18_hstu_attention/hstu_attention_pipeline_problem.hpp
@@ -21,6 +21,7 @@ template <typename InOutDataType_,
           bool kHasBias_,
           bool kHasDropout_,
           bool kHasCausal_,
+          bool kUseSoftmax_,
           typename AttentionTileSetting_,
           typename Traits_>
 struct HstuAttentionFwdPipelineProblem
@@ -42,6 +43,7 @@ struct HstuAttentionFwdPipelineProblem
     static constexpr bool kHasBias    = kHasBias_;
     static constexpr bool kHasDropout = kHasDropout_;
     static constexpr bool kHasCausal  = kHasCausal_;
+    static constexpr bool kUseSoftmax = kUseSoftmax_;
 
     using HstuAttentionTileSetting = remove_cvref_t<AttentionTileSetting_>;
 
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_no_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_no_bias_has_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
index 5da34dc283..36cb9a65f2 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_no_bias_has_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
     false,
     true,
+    true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_has_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_has_bias_no_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
index 8d2076e4a2..71febdb83e 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_has_bias_no_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
-    true,
     false,
+    true,
+    true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_no_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_no_bias_has_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
index 1e5f3d8a34..bb398341f6 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_no_bias_has_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
     false,
     true,
+    true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_has_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_has_bias_no_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
index 5869131cac..842c81d46f 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_has_bias_no_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     true,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_has_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_has_bias_no_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
index 1fe75755bb..de1bab9bba 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_has_bias_no_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     true,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_has_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_has_bias_no_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
index f8a29bae1e..a391cdefeb 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_has_bias_no_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     true,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_no_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_no_bias_has_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
index 06e46a8016..7e38f57564 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_no_bias_has_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     true,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_no_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_no_bias_has_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
index 7850b7c5a2..fc1312e0b9 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_no_bias_has_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     true,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_no_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_no_bias_has_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
index 23130f6570..8f21011a2f 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_no_bias_has_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     true,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_no_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_no_bias_no_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
index f0a46ce2de..b57fcf1ae6 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_no_bias_no_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_no_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_no_bias_no_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
index 93c13f38c1..e0eda1fd86 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_no_bias_no_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_no_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_no_bias_no_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
index 6bf55948d6..20a2c94eb6 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_no_bias_no_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_has_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_has_bias_has_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
index 83c5bdd5ff..847dade837 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_has_bias_has_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
     true,
     true,
+    true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_has_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_has_bias_has_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
index f30110917a..cd4db455a0 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_has_bias_has_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
     true,
     true,
+    true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_has_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_has_bias_has_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
index 90b8557133..b0fe7e9a0c 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_has_bias_has_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
     true,
     true,
+    true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_has_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_has_bias_has_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
index 88d54e9023..c2d4368586 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_has_bias_has_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
+    true,
+    true,
     false,
-    true,
-    true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_no_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_no_bias_has_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
index 26d62e70ca..ae9a9820cd 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_no_bias_has_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
-    false,
     true,
+    true,
+    false,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_has_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_has_bias_no_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
index 217aa242b4..2a99b9be99 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_has_bias_no_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
     true,
+    true,
     false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_has_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_has_bias_no_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
index 67d6c22f04..5f6437e2f2 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_has_bias_no_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
     true,
     false,
+    true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_has_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_has_bias_has_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
index 7e94d0fecb..2084aa6e87 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_has_bias_has_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
+    true,
     false,
     true,
-    true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_has_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_has_bias_has_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
index 2803ee1722..26d25bd909 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_has_bias_has_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
+    true,
     false,
     true,
-    true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_no_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_no_bias_no_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
index 39f99c69b4..e006dee199 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_no_bias_no_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
+    true,
     false,
     false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_no_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_no_bias_no_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
index 142dc101ec..a097c97db5 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_no_bias_no_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
+    true,
     false,
     false,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_no_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_no_bias_no_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
index 48daf07cab..3209277570 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_no_bias_no_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
+    true,
     false,
     false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_instances_ref.hpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_instances_ref.hpp
index e427616a8d..92be40644d 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_instances_ref.hpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_instances_ref.hpp
@@ -9,169 +9,385 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
     true,
     true,
+    true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     true,
     true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     true,
     false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     true,
     false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     false,
     true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     false,
     false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     true,
     true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     true,
     true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     true,
     false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     true,
     false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     false,
     true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     false,
     false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     true,
     true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     true,
     true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     true,
     false,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     true,
     false,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     false,
     true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     false,
     false,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
+    false,
+    false,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
     false,
     false,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..44e9e9115f
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..cc10734807
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..2303ca8181
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..745ace2b60
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..f1957e30be
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..2c3f9b0ee3
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..2001840777
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..3e799fc731
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..f4b4000478
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..2016f8ebda
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..ec2a5e7c94
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..e99e8cfedc
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..c21fe134be
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..7c285b0079
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..7fe019360b
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..7418ceafbe
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..71ad9b6cf3
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..1cd866c432
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..a641156a39
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..2085f81314
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..36aa6cd03b
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..6ca7ad1e75
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..b218e0965c
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..a75ad6a2b4
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_no_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_no_bias_has_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
index d5f156b831..aee00936ab 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_no_bias_has_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
     false,
     true,
+    true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_has_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_has_bias_no_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
index 2de6890d7b..de9d2248ab 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_has_bias_no_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
-    true,
     false,
+    true,
+    true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_no_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_no_bias_has_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
index 0c38bd5fc0..6f2ddd517a 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_no_bias_has_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
     false,
     true,
+    true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_has_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_has_bias_no_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
index 07a2e191ff..9a247781f1 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_has_bias_no_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     true,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_has_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_has_bias_no_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
index e93483d743..0a133a4528 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_has_bias_no_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     true,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_has_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_has_bias_no_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
index 4974885839..1f73b0903a 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_has_bias_no_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     true,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_no_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_no_bias_has_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
index c8052fc524..0a41097b22 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_no_bias_has_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     false,
     true,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_no_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_no_bias_has_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
index 9ea678a5f8..54cfbb4c51 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_no_bias_has_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     false,
     true,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_no_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_no_bias_no_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
index c144ffc9c8..3f179b6895 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_no_bias_no_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
     false,
     false,
+    true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_no_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_no_bias_no_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
index 26af585661..82cd2f4b84 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_no_bias_no_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     false,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_no_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_no_bias_no_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
index 659b548f65..f8403d811f 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_no_bias_no_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     false,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_no_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_no_bias_no_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
index 5c32779c32..9244f2d58c 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_no_bias_no_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     false,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_has_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_has_bias_has_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
index bc2807fb64..d0e54b40b9 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_has_bias_has_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
     true,
     true,
+    true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_has_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_has_bias_has_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
index 0993862073..dba42fb653 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_has_bias_has_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
     true,
     true,
+    true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_has_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_has_bias_has_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
index f23cb03f49..70d653d799 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_has_bias_has_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
     true,
     true,
+    true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_has_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_has_bias_has_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
index fae90c5abf..077ad80e73 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_has_bias_has_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
+    true,
+    true,
     false,
-    true,
-    true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_no_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_no_bias_has_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
index f00bbcedc1..a9cbc16d54 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_no_bias_has_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
-    false,
     true,
+    true,
+    false,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_has_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_has_bias_no_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
index 13019d228a..49daa7c9a5 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_has_bias_no_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
     true,
+    true,
     false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_has_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_has_bias_no_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
index 6daa05688f..72476a7d86 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_has_bias_no_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
     true,
     false,
+    true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_has_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_has_bias_has_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
index b7ca6dbcd3..a817cc54df 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_has_bias_has_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
+    true,
     false,
     true,
-    true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_has_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_has_bias_has_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
index 2857284fd1..4c014c2019 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_has_bias_has_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
+    true,
     false,
     true,
-    true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_no_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_no_bias_no_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
index 84f4aedd3a..50a6491efb 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_no_bias_no_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
+    true,
     false,
     false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_no_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_no_bias_no_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
index 57dc438e87..aa2545ce90 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_no_bias_no_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
+    true,
     false,
     false,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_no_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_no_bias_has_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
index fd265fa161..b68fbc7297 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_no_bias_has_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-template void run_batched_forward_causal_bias_dropout_dispatch<
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
-    false,
-    false,
     true,
+    true,
+    false,
+    false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_instances_ref.hpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_instances_ref.hpp
index fd049beeb8..9b49db3cf3 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_instances_ref.hpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_instances_ref.hpp
@@ -9,169 +9,385 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_batched_forward_dispatch.hpp"
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
     true,
     true,
+    true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     true,
     true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     true,
     false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     true,
     false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     false,
     true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     false,
     true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     false,
     false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     false,
     false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     true,
     true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     true,
     true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     true,
     false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     true,
     false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     false,
     true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     false,
     true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     false,
     false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     false,
     false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     true,
     true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     true,
     true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     true,
     false,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     true,
     false,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     false,
     true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     false,
     true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     false,
     false,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_batched_forward_causal_bias_dropout_dispatch<
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
+    false,
+    false,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
     false,
     false,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..cec5df8883
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..46a4760dca
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..98e60c8c65
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..49e5c27d4f
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..beeeef97df
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..abeb6f6fa4
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..5d8f42117f
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..0f5ff1a61b
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..3cdc33115f
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..7d46856d36
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..32042ed616
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..64e37de126
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..bc8984b4e5
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..4a6929b18f
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..87b7a71bca
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..3d882587a0
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..66e0d89883
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..b25b21e9bb
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..7014050cff
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..99eeccd6e5
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..5a829be981
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..7211600ef9
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..86427b0848
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..9daefa8cf1
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_batched_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_batched_forward_dispatch.hpp"
+
+template void run_batched_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_has_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_has_bias_no_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
index b035a633ac..47fa39006b 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_has_bias_no_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
-    true,
     false,
+    true,
+    true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_has_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_has_bias_no_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
index b0dedb77fe..8dd7fed807 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_has_bias_no_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
-    true,
     false,
+    true,
+    true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_has_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_has_bias_no_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
index 3c97baa783..3b221a0639 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_has_bias_no_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
-    true,
     false,
+    true,
+    true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_has_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_has_bias_no_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
index d2cc31743a..c9ea8c9f85 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_has_bias_no_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     true,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_has_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_has_bias_no_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
index 90a53861f6..66495aa174 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_has_bias_no_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     true,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_has_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_has_bias_no_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
index cbdeb383d9..65d0869668 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_has_bias_no_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     true,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_no_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_no_bias_has_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
index bc936753d2..4e85352335 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_no_bias_has_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     true,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_no_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_no_bias_has_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
index 1f79e75b58..4a08ec78c0 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_no_bias_has_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     true,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_no_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_no_bias_has_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
index 8353360002..2ecf0f834a 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_no_bias_has_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     true,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_no_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_no_bias_no_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
index c5edd4c845..801d843934 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_no_bias_no_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_no_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_no_bias_no_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
index 588a11259f..be24f955ab 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_no_bias_no_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_no_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_no_bias_no_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
index 2c4949dbd7..4a8c665074 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_no_bias_no_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_has_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_has_bias_has_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
index e6bc0acf12..2f35f8b934 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_has_bias_has_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
     true,
     true,
+    true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_has_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_has_bias_has_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
index 445e49af15..b5bffe92c5 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_has_bias_has_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
     true,
     true,
+    true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_has_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_has_bias_has_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
index 2e949e4b9b..895c495536 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_has_bias_has_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
     true,
     true,
+    true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_no_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_no_bias_has_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
index e54e4ea4db..76e162a819 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_no_bias_has_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
-    false,
     true,
+    true,
+    false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_no_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_no_bias_has_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
index 1b156f0d04..9afb1ff339 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_no_bias_has_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
-    false,
     true,
+    true,
+    false,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_no_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_no_bias_has_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
index 2da30a2b76..952559d871 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_no_bias_has_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
-    false,
     true,
+    true,
+    false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_has_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_has_bias_has_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
index 391663aa33..10301fff99 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_has_bias_has_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
+    true,
     false,
     true,
-    true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_has_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_has_bias_has_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
index 5699dcd8cd..b0de249e88 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_has_bias_has_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
+    true,
     false,
     true,
-    true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_has_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_has_bias_has_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
index 77f7b68ad2..f73442f7c2 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_has_bias_has_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
+    true,
     false,
     true,
-    true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_no_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_no_bias_no_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
index 8b4fd5554b..2f93f9a67e 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_no_bias_no_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
+    true,
     false,
     false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_no_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_no_bias_no_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
index f45bda223c..763bb9d6de 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_no_bias_no_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
+    true,
     false,
     false,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_no_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_no_bias_no_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
index cabcff4a20..1891d19185 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_no_bias_no_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_has_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
+    true,
     false,
     false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_instances_ref.hpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_instances_ref.hpp
index 9592f8e1b0..9689dc395f 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_instances_ref.hpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_instances_ref.hpp
@@ -9,169 +9,385 @@
 #include <ck_tile/core/numeric/half.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
     true,
     true,
     true,
+    true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     true,
     true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     true,
     false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     true,
     false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     false,
     true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     false,
     false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     true,
     true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     true,
     true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     true,
     false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     true,
     false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     false,
     true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     false,
     false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     true,
     true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     true,
     true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     true,
     false,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     true,
     false,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     false,
     true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
     false,
     false,
     true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    true,
     true,
     false,
     false,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::bf16_t,
+    true,
+    false,
+    false,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
     false,
     false,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..9bf339ee25
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..2dc97f8687
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..b6c7fc3072
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..417c579a49
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..c7c41e843f
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..01ee526d18
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    true,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..fe604c177e
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..4ff7c6449b
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..4b8e548247
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..69450e0993
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..5581c4c525
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..56c4b2fabf
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    false,
+    false,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..60b9f399ce
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..65294641ca
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..50d0570d54
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..4ef06ef3ff
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..01715cb711
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..60d34a2dba
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    true,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..9ab39a0d70
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..14f91e6df7
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..ab06a435bd
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..c531054ec5
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..63e7a21dbf
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..a9a9b9450f
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_bf16_no_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/half.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::bf16_t,
+    false,
+    true,
+    false,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_has_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_has_bias_no_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
index 1a9dee4904..487e2f3979 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_has_bias_no_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
-    true,
     false,
+    true,
+    true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_has_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_has_bias_no_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
index 1cd93c371b..2cd3628a64 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_has_bias_no_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
-    true,
     false,
+    true,
+    true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_has_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_has_bias_no_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
index 9338479037..c5782adb4e 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_has_bias_no_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
-    true,
     false,
+    true,
+    true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_has_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_has_bias_no_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
index 9b5be023b0..3a64e883a9 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_has_bias_no_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     true,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_has_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_has_bias_no_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
index 90b75d5176..6fedc16471 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_has_bias_no_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     true,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_has_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_has_bias_no_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
index 356c944c50..5d10a6bb43 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_has_bias_no_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     true,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_no_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_no_bias_has_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
index 42f024c736..aa8387967d 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_no_bias_has_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     false,
     true,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_no_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_no_bias_has_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
index 82eb5592b3..e47ba94d3d 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_no_bias_has_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     false,
     true,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_no_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_no_bias_has_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
index 59656a53d3..834773eba9 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_no_bias_has_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     false,
     true,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_no_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_no_bias_no_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
index 168868b98c..0d9075f495 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_no_bias_no_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     false,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_no_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_no_bias_no_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
index 413b8d0539..8788e61c95 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_no_bias_no_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     false,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_no_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_no_bias_no_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
index 09dc46d17d..c9f5926653 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_no_bias_no_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
@@ -9,8 +9,9 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     false,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_has_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_has_bias_has_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
index 69e9de96c8..87bf14777e 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_has_bias_has_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
     true,
     true,
+    true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_has_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_has_bias_has_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
index b4980956f7..e12bfe1f5b 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_has_bias_has_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
     true,
     true,
+    true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_has_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_has_bias_has_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
index 2f8ad1c663..93bb83bddb 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_has_bias_has_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
     true,
     true,
+    true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_no_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_no_bias_has_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
index 16cdbbe212..87a9d42429 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_no_bias_has_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
-    false,
     true,
+    true,
+    false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_no_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_no_bias_has_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
index 55fe8f70fd..14f29bbad9 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_no_bias_has_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
-    false,
     true,
+    true,
+    false,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_no_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_no_bias_has_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
index 527c1d55da..1ce8e85516 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_no_bias_has_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
-    false,
     true,
+    true,
+    false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_has_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_has_bias_has_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
index bdbc1ba394..a8fc8e0b09 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_has_bias_has_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
+    true,
     false,
     true,
-    true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_has_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_has_bias_has_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
index 01a673a840..e29fcc1d5e 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_has_bias_has_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
+    true,
     false,
     true,
-    true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_has_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_has_bias_has_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
index 61aab431fe..9bc937b0cd 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_has_bias_has_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
+    true,
     false,
     true,
-    true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_no_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_no_bias_no_dropout_maxk_128.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
index f2af3fc71f..2463c4bc7b 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_no_bias_no_dropout_maxk_128.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
+    true,
     false,
     false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_no_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_no_bias_no_dropout_maxk_256.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
index fb528405f3..9f531bee1a 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_no_bias_no_dropout_maxk_256.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
+    true,
     false,
     false,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_no_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
similarity index 85%
rename from example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_no_bias_no_dropout_maxk_64.cpp
rename to example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
index f446651bf5..504cc5c3d0 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_no_bias_no_dropout_maxk_64.cpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_has_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
@@ -9,9 +9,10 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-template void run_jagged_forward_causal_bias_dropout_dispatch<
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
+    true,
     false,
     false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_instances_ref.hpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_instances_ref.hpp
index 2494277eac..702ec002b0 100644
--- a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_instances_ref.hpp
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_instances_ref.hpp
@@ -9,169 +9,385 @@
 #include <ck_tile/core/numeric/bfloat16.hpp>
 #include "hstu_attention_jagged_forward_dispatch.hpp"
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
     true,
     true,
     true,
+    true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     true,
     true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     true,
     false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     true,
     false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     false,
     true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     false,
     true,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     false,
     false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     false,
     false,
     64>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     true,
     true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     true,
     true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     true,
     false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     true,
     false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     false,
     true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     false,
     true,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     false,
     false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     false,
     false,
     128>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     true,
     true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     true,
     true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     true,
     false,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     true,
     false,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     false,
     true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
     false,
     false,
     true,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    true,
     true,
     false,
     false,
     256>(HstuAttentionFwdParams& param, hipStream_t stream);
 
-extern template void run_jagged_forward_causal_bias_dropout_dispatch<
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
     ck_tile::fp16_t,
+    true,
+    false,
+    false,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
+
+extern template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
     false,
     false,
     false,
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..76f797baec
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..d7f3519fee
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..2f2899c9fd
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_has_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..3e6a2c1664
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..1a228db8cf
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..a1b326c469
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_has_bias_no_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    true,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..9e44077967
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..b58b3a7ada
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..57f1337ae2
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_has_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..36a80da483
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..b7cce60aaf
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..e03e1e272a
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_false_no_bias_no_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    false,
+    false,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..1e7b7d9633
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..1f8a1d7d4e
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..1e4f4b4b65
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_has_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..043d693856
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..487786dd6a
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..f433cbba4d
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_has_bias_no_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    true,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..6138bce84f
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    true,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..ff3759616f
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    true,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..60fd4c1e06
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_has_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    true,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
new file mode 100644
index 0000000000..5b0fb56a3f
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_128.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    false,
+    128>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
new file mode 100644
index 0000000000..e46694aaab
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_256.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    false,
+    256>(HstuAttentionFwdParams& param, hipStream_t stream);
diff --git a/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
new file mode 100644
index 0000000000..0d7a4320b3
--- /dev/null
+++ b/example/ck_tile/18_hstu_attention/instances/hstu_attention_jagged_forward_fp16_no_causal_softmax_true_no_bias_no_dropout_maxk_64.cpp
@@ -0,0 +1,18 @@
+
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
+
+// The file is automatically generated, don't modify!
+// See the generator script
+// `composable_kernel/example/ck_tile/18_hstu_attention/generate_instances.py`
+
+#include <ck_tile/core/numeric/bfloat16.hpp>
+#include "hstu_attention_jagged_forward_dispatch.hpp"
+
+template void run_jagged_forward_causal_softmax_bias_dropout_dispatch<
+    ck_tile::fp16_t,
+    false,
+    true,
+    false,
+    false,
+    64>(HstuAttentionFwdParams& param, hipStream_t stream);