diff --git a/transformer_engine/common/common.cu b/transformer_engine/common/common.cu index 192c915..e94a92b 100644 --- a/transformer_engine/common/common.cu +++ b/transformer_engine/common/common.cu @@ -6,7 +6,6 @@ #include -#include #include "./common.h" #include "./utils.cuh" diff --git a/transformer_engine/common/cudnn_utils.h b/transformer_engine/common/cudnn_utils.h index 0016ad7..985a9ba 100644 --- a/transformer_engine/common/cudnn_utils.h +++ b/transformer_engine/common/cudnn_utils.h @@ -8,8 +8,8 @@ #define TRANSFORMER_ENGINE_CUDNN_UTILS_H_ #include -#include -#include +#include "third_party/cudnn_frontend/include/cudnn_frontend.h" +#include "third_party/cudnn_frontend/include/cudnn_frontend_utils.h" #include #include "transformer_engine/transformer_engine.h" diff --git a/transformer_engine/common/fused_attn/fused_attn_f16_arbitrary_seqlen.cu b/transformer_engine/common/fused_attn/fused_attn_f16_arbitrary_seqlen.cu index 0932b2c..7fd2a94 100644 --- a/transformer_engine/common/fused_attn/fused_attn_f16_arbitrary_seqlen.cu +++ b/transformer_engine/common/fused_attn/fused_attn_f16_arbitrary_seqlen.cu @@ -6,8 +6,8 @@ #include #include -#include -#include +#include "third_party/cudnn_frontend/include/cudnn_frontend.h" +#include "third_party/cudnn_frontend/include/cudnn_frontend_utils.h" #include #include diff --git a/transformer_engine/common/fused_attn/fused_attn_f16_max512_seqlen.cu b/transformer_engine/common/fused_attn/fused_attn_f16_max512_seqlen.cu index 89528fa..9f53123 100644 --- a/transformer_engine/common/fused_attn/fused_attn_f16_max512_seqlen.cu +++ b/transformer_engine/common/fused_attn/fused_attn_f16_max512_seqlen.cu @@ -6,7 +6,7 @@ #include #include -#include +#include "third_party/cudnn_frontend/include/cudnn_frontend.h" #include #include diff --git a/transformer_engine/common/fused_attn/utils.h b/transformer_engine/common/fused_attn/utils.h index 678b636..58bcec4 100644 --- a/transformer_engine/common/fused_attn/utils.h +++ b/transformer_engine/common/fused_attn/utils.h @@ -8,8 +8,8 @@ #define TRANSFORMER_ENGINE_FUSED_ATTN_UTILS_H_ #include -#include -#include +#include "third_party/cudnn_frontend/include/cudnn_frontend.h" +#include "third_party/cudnn_frontend/include/cudnn_frontend_utils.h" #include #include diff --git a/transformer_engine/common/normalization/common.h b/transformer_engine/common/normalization/common.h index 0ec1604..aef319d 100644 --- a/transformer_engine/common/normalization/common.h +++ b/transformer_engine/common/normalization/common.h @@ -8,8 +8,8 @@ #define TRANSFORMER_ENGINE_COMMON_NORM_COMMON_H_ #include -#include -#include +#include "third_party/cudnn_frontend/include/cudnn_frontend.h" +#include "third_party/cudnn_frontend/include/cudnn_frontend_utils.h" #include #include diff --git a/transformer_engine/common/util/cuda_runtime.cpp b/transformer_engine/common/util/cuda_runtime.cpp index ac1bb1b..febb9a1 100644 --- a/transformer_engine/common/util/cuda_runtime.cpp +++ b/transformer_engine/common/util/cuda_runtime.cpp @@ -13,18 +13,12 @@ #include "../util/cuda_driver.h" #include "../util/system.h" #include "common/util/cuda_runtime.h" +#include "tsl/platform/cuda_root_path.h" namespace transformer_engine { namespace cuda { -namespace { - -// String with build-time CUDA include path -#include "string_path_cuda_include.h" - -} // namespace - int num_devices() { auto query_num_devices = []() -> int { int count; @@ -146,8 +140,10 @@ const std::string &include_directory(bool required) { std::vector> search_paths = {{"NVTE_CUDA_INCLUDE_DIR", ""}, {"CUDA_HOME", ""}, {"CUDA_DIR", ""}, - {"", string_path_cuda_include}, {"", "/usr/local/cuda"}}; + for (auto &candidate : tsl::CandidateCudaRoots()) { + search_paths.push_back({"", candidate}); + } for (auto &[env, p] : search_paths) { if (p.empty()) { p = getenv(env.c_str()); diff --git a/transformer_engine/jax/csrc/extensions/attention.cpp b/transformer_engine/jax/csrc/extensions/attention.cpp index 40089dc..cf256e3 100644 --- a/transformer_engine/jax/csrc/extensions/attention.cpp +++ b/transformer_engine/jax/csrc/extensions/attention.cpp @@ -5,6 +5,7 @@ ************************************************************************/ #include "../extensions.h" +#include #include "transformer_engine/fused_attn.h" #include "transformer_engine/transformer_engine.h" diff --git a/transformer_engine/jax/csrc/extensions/ffi.h b/transformer_engine/jax/csrc/extensions/ffi.h index 852a67c..92df9de 100644 --- a/transformer_engine/jax/csrc/extensions/ffi.h +++ b/transformer_engine/jax/csrc/extensions/ffi.h @@ -5,7 +5,7 @@ ************************************************************************/ #include -#include +#include "xla/ffi/api/ffi.h" #include diff --git a/transformer_engine/jax/csrc/extensions/utils.cpp b/transformer_engine/jax/csrc/extensions/utils.cpp index 3ba0737..10b72c8 100644 --- a/transformer_engine/jax/csrc/extensions/utils.cpp +++ b/transformer_engine/jax/csrc/extensions/utils.cpp @@ -6,6 +6,7 @@ #include "utils.h" #include +#include #include