CK Tile FA Training kernels (#1286)

* FA fwd dropout * FA bwd * epilogue reuse * CMakeLists update * [CK_TILE] support alibi (#1269) * add alibi support * fix code * update code based on comment * Support more hdim * fix fp8 bias * support seqlen_k=0 case * remove unused printf * fix format --------- Co-authored-by: rocking <ChunYu.Lai@amd.com> * now fwd/bwd can build * bwd alibi * add bwd validation stream_config * update generated filenames * update bwd kernel launch * CK_TILE_HOST_DEVICE in philox * Transpose -> transpose * format * format * format * Generate the instance for FA required * format * fix error in WarpGemm --------- Co-authored-by: danyao12 <danyao12> Co-authored-by: carlushuang <carlus.huang@amd.com> Co-authored-by: rocking <ChunYu.Lai@amd.com> Co-authored-by: Po Yen Chen <PoYen.Chen@amd.com> Co-authored-by: Jing Zhang <jizhan@amd.com>
2026-05-01 20:21:23 +00:00 · 2024-06-05 02:12:45 +08:00
parent 76827d82ca
commit 2cab8d39e3
70 changed files with 9506 additions and 482 deletions
--- a/include/ck_tile/host/reference/reference_batched_dropout.hpp
+++ b/include/ck_tile/host/reference/reference_batched_dropout.hpp
@@ -0,0 +1,33 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck_tile/core.hpp"
+#include "ck_tile/host/host_tensor.hpp"
+#include <thread>
+
+namespace ck_tile {
+
+template <typename DataType, typename RandValOutputDataType>
+CK_TILE_HOST void reference_batched_dropout(HostTensor<DataType>& in_out_b_m_n,
+                                            const HostTensor<RandValOutputDataType>& randval_b_m_n,
+                                            const uint8_t& p_undrop_in_uint8_t,
+                                            const float scale)
+{
+    const int N = in_out_b_m_n.mDesc.get_lengths()[2];
+    auto f      = [&](auto batch, auto m) {
+        for(int n = 0; n < N; ++n)
+        {
+            float tmp = ck_tile::type_convert<float>(in_out_b_m_n(batch, m, n)) * scale;
+            in_out_b_m_n(batch, m, n) = randval_b_m_n(batch, m, n) <= p_undrop_in_uint8_t
+                                                 ? ck_tile::type_convert<DataType>(tmp)
+                                                 : DataType(0);
+        }
+    };
+
+    make_ParallelTensorFunctor(
+        f, randval_b_m_n.mDesc.get_lengths()[0], randval_b_m_n.mDesc.get_lengths()[1])(
+        std::thread::hardware_concurrency());
+}
+} // namespace ck_tile