rebase from main

Author: lanluo-nvidia

.github/workflows/mlir-tensorrt/ci-utils.sh

@@ -77,7 +77,7 @@ cmd_lint_check() {
   local range
   range="$(compute_range)"
 
-  uv tool install black
+  uv tool install 'black<26.0'
   uvx black --check --extend-exclude='.*\.pyi' mlir-tensorrt/compiler/ mlir-tensorrt/integrations
 
   local clang_format_diff

.github/workflows/tripy-l0.yml

@@ -11,6 +11,7 @@ env:
   REGISTRY: ghcr.io
   DEFAULT_IMAGE: ghcr.io/nvidia/tensorrt-incubator/tripy:latest
   NEW_TEST_IMAGE: test-image:latest
+  HF_TOKEN: ${{ secrets.HF_TOKEN }}
 
 
 jobs:
@@ -58,7 +59,7 @@ jobs:
       uses: addnab/docker-run-action@v3
       with:
         image: ${{ env.l0_image }}
-        options: --gpus all -v ${{ github.workspace }}/tripy:/tripy
+        options: --gpus all -v ${{ github.workspace }}/tripy:/tripy -e HF_TOKEN=${{ env.HF_TOKEN }}
         run: |
           python3 docs/generate_rsts.py
           sphinx-build build/doc_sources build/docs -c docs/ -j 4 -W -n
@@ -67,15 +68,15 @@ jobs:
       uses: addnab/docker-run-action@v3
       with:
         image: ${{ env.l0_image }}
-        options: --gpus all -v ${{ github.workspace }}/tripy:/tripy
+        options: --gpus all -v ${{ github.workspace }}/tripy:/tripy -e HF_TOKEN=${{ env.HF_TOKEN }}
         run: |
           pytest --cov=nvtripy/ --cov-config=.coveragerc tests/ -v -m "not l1" -n 4 --durations=15 --ignore tests/performance
 
     - name: Run performance benchmarks
       uses: addnab/docker-run-action@v3
       with:
         image: ${{ env.l0_image }}
-        options: --gpus all -v ${{ github.workspace }}/tripy:/tripy
+        options: --gpus all -v ${{ github.workspace }}/tripy:/tripy -e HF_TOKEN=${{ env.HF_TOKEN }}
         run: |
           pytest tests/performance -v -m "not l1" --benchmark-warmup=on --benchmark-json benchmark.json
 

.github/workflows/tripy-l1.yml

@@ -18,6 +18,8 @@ concurrency:
 jobs:
   l1-test:
     runs-on: tripy-self-hosted
+    env:
+      HF_TOKEN: ${{ secrets.HF_TOKEN }}
     container:
       image: ghcr.io/nvidia/tensorrt-incubator/tripy:latest
       volumes:

mlir-tensorrt/.pre-commit-config.yaml

@@ -30,7 +30,7 @@ repos:
       rev: 6.0.0
       hooks:
           - id: isort
-            name: isort (python)
+            args: ["--profile", "black"]
     - repo: https://github.com/pre-commit/mirrors-mypy
       rev: "v1.15.0"
       hooks:

mlir-tensorrt/CMakeLists.txt

@@ -166,9 +166,13 @@ include(MLIRDetectPythonEnv)
 mlir_configure_python_dev_packages()
 
 find_package(TVMFFI REQUIRED)
-message(STATUS "LAN ADDED BEFORE FINDING GTEST PACKAGE ")
+
 find_package(GTest REQUIRED)
-message(STATUS "LAN ADDED AFTER FINDING GTEST PACKAGE ")
+
+if(MLIR_TRT_ENABLE_CUDATILE)
+  find_package(CUDATile REQUIRED)
+endif()
+
 #--------------------------------------------------
 # Diagnostics
 #--------------------------------------------------
@@ -204,7 +208,9 @@ endforeach()
 include_directories(${LLVM_INCLUDE_DIRS})
 include_directories(${MLIR_INCLUDE_DIRS})
 link_directories(${LLVM_BUILD_LIBRARY_DIR})
-add_definitions(${LLVM_DEFINITIONS} ${MLIR_TRT_DEFINITIONS})
+separate_arguments(LLVM_DEFINITIONS_LIST NATIVE_COMMAND ${LLVM_DEFINITIONS})
+add_definitions(${LLVM_DEFINITIONS_LIST})
+add_definitions(${LLVM_DEFINITIONS_LIST} ${MLIR_TRT_DEFINITIONS})
 
 mtrt_write_feature_flags_header()
 include_directories("${CMAKE_CURRENT_BINARY_DIR}/include")

mlir-tensorrt/CMakeOptions.cmake

@@ -8,6 +8,7 @@ mtrt_option(MLIR_TRT_ENABLE_NCCL "Enable the NCCL runtime module" OFF)
 mtrt_option(MLIR_TRT_ENABLE_CUBLAS "Enable CUBLAS in the executor" ON)
 mtrt_option(MLIR_TRT_ENABLE_SHARDY "Enable compiler build with Shardy" OFF)
 mtrt_option(MLIR_TRT_ENABLE_CUDA "Enable the use of CUDA runtime" ON)
+mtrt_option(MLIR_TRT_ENABLE_CUDATILE "Enable the use of CUDA Tile IR" ON)
 mtrt_option(MLIR_TRT_ENABLE_MPI "Enable use of MPI in the runtime" ${MLIR_TRT_ENABLE_NCCL})
 mtrt_option(MLIR_TRT_LINK_MTRT_DYLIB "Link all tools against libMTRT dylib" OFF)
 mtrt_option(MLIR_TRT_LINK_MLIR_DYLIB "Use the libMLIR dylib to provide MLIR-TensorRT's MLIR dependencies" OFF)
@@ -65,8 +66,10 @@ set(MLIR_TRT_CUDA_ARCHITECTURES "detect" CACHE STRING
 # TODO: consider adding an option to link cudart statically.
 if(MLIR_TRT_ENABLE_CUDA)
   set(MLIR_TRT_CUDA_TARGET "CUDA::cudart" CACHE INTERNAL "")
+  set(MLIR_TRT_CUDA_DRIVER_TARGET "CUDA::cuda_driver" CACHE INTERNAL "")
 else()
   set(MLIR_TRT_CUDA_TARGET "" CACHE INTERNAL "")
+  set(MLIR_TRT_CUDA_DRIVER_TARGET "" CACHE INTERNAL "")
 endif()
 
 # Set `MLIR_TRT_MPI_TARGET` to `MPI::MPI_C` if MPI is enabled, otherwise set it to an empty string.

mlir-tensorrt/DependencyProvider.cmake

@@ -71,6 +71,27 @@ function(mtrt_find_tvm_ffi)
     CACHE INTERNAL "")
 endfunction()
 
+#-------------------------------------------------------------------------------------
+# CUDA Tile
+#-------------------------------------------------------------------------------------
+
+nv_register_package(
+  NAME CUDATile
+  GIT_REPOSITORY https://github.com/NVIDIA/cuda-tile.git
+  GIT_TAG 802d9378800a3b7c9f88875206e84b2746d6991b
+  EXCLUDE_FROM_ALL TRUE
+  OPTIONS
+    "CUDA_TILE_ENABLE_TESTING ON"
+  PRE_ADD_HOOK [[
+    nv_pkg_append_options("CUDA_TILE_USE_LLVM_INSTALL_DIR ${LLVM_BINARY_DIR}")
+    if(NOT CPM_CUDATile_SOURCE)
+      set(patch_dir "${MTRT_TOP_LEVEL_DIR}/build_tools/patches/CUDATile")
+      nv_update_append_pkg_args(PATCHES
+        "${patch_dir}/0001-Fix-build-with-BUILD_SHARED_LIBS-ON-and-prevent-auto.patch")
+    endif()
+  ]]
+)
+
 #-------------------------------------------------------------------------------------
 # Declare the LLVM dependency.
 #-------------------------------------------------------------------------------------

mlir-tensorrt/Version.cmake

@@ -1,6 +1,6 @@
 set(MLIR_TENSORRT_VERSION_MAJOR "0")
 set(MLIR_TENSORRT_VERSION_MINOR "4")
-set(MLIR_TENSORRT_VERSION_PATCH "4")
+set(MLIR_TENSORRT_VERSION_PATCH "6")
 set(MLIR_TENSORRT_VERSION
   "${MLIR_TENSORRT_VERSION_MAJOR}.${MLIR_TENSORRT_VERSION_MINOR}.${MLIR_TENSORRT_VERSION_PATCH}")
 

mlir-tensorrt/build_tools/cmake/MTRTCompilationOptions.cmake

@@ -113,9 +113,10 @@ function(mtrt_apply_relative_path_options)
   # The following checks if the source directory is a prefix of the binary directory.
   cmake_path(IS_PREFIX CMAKE_SOURCE_DIR "${CMAKE_BINARY_DIR}" src_dir_is_prefix_of_bin_dir)
 
-  set(debug_remaps "-fdebug-prefix-map=${CMAKE_SOURCE_DIR}=.")
-  set(file_remaps "-ffile-prefix-map=${CMAKE_SOURCE_DIR}=.")
-  if(src_dir_is_prefix_of_bin_dir)
+  set(debug_remaps "-fdebug-prefix-map=${CMAKE_SOURCE_DIR}/=")
+  set(file_remaps "-ffile-prefix-map=${CMAKE_SOURCE_DIR}/=")
+
+  if(NOT src_dir_is_prefix_of_bin_dir)
     set(debug_remaps "${debug_remaps} -fdebug-prefix-map=${CMAKE_BINARY_DIR}=build")
     set(file_remaps "${file_remaps} -ffile-prefix-map=${CMAKE_BINARY_DIR}=build")
   endif()
@@ -133,6 +134,9 @@ function(mtrt_apply_relative_path_options)
     CHECK "-no-canonical-prefixes"
     APPEND "-no-canonical-prefixes"
   )
+
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}" PARENT_SCOPE)
+  set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS}" PARENT_SCOPE)
 endfunction()
 
 if(PROJECT_IS_TOP_LEVEL AND MLIR_TRT_RELATIVE_DEBUG_PATHS)

mlir-tensorrt/build_tools/patches/CUDATile/0001-Fix-build-with-BUILD_SHARED_LIBS-ON-and-prevent-auto.patch

@@ -0,0 +1,53 @@
+From dd0b0215100ee530ee58dec18726cff70c0b17a1 Mon Sep 17 00:00:00 2001
+From: Christopher Bate <cbate@nvidia.com>
+Date: Sun, 11 Jan 2026 13:40:52 -0700
+Subject: [PATCH 1/1] Fix build with BUILD_SHARED_LIBS=ON and prevent automatic
+ creation of empty media directory
+
+---
+ include/cuda_tile/Dialect/CudaTile/IR/CMakeLists.txt | 2 +-
+ lib/Bytecode/Translation/CMakeLists.txt              | 1 +
+ lib/Dialect/CudaTile/IR/CMakeLists.txt               | 3 +++
+ 3 files changed, 5 insertions(+), 1 deletion(-)
+
+diff --git a/include/cuda_tile/Dialect/CudaTile/IR/CMakeLists.txt b/include/cuda_tile/Dialect/CudaTile/IR/CMakeLists.txt
+index 93b9c7f..414db08 100644
+--- a/include/cuda_tile/Dialect/CudaTile/IR/CMakeLists.txt
++++ b/include/cuda_tile/Dialect/CudaTile/IR/CMakeLists.txt
+@@ -43,10 +43,10 @@ set(LLVM_TARGET_DEFINITIONS Types.td)
+ tablegen(MLIR CudaTileTypes.md -gen-typedef-doc -allow-hugo-specific-features)
+ 
+ set(GEN_DOC_DIR ${CMAKE_SOURCE_DIR}/media/docs/Dialects/CudaTile)
+-file(MAKE_DIRECTORY ${GEN_DOC_DIR})
+ 
+ add_custom_command(
+ 	OUTPUT ${GEN_DOC_DIR}/CudaTile.md
++        COMMAND ${CMAKE_COMMAND} -E make_directory ${GEN_DOC_DIR}
+         COMMAND ${CMAKE_COMMAND} -E cat
+         ${CMAKE_CURRENT_BINARY_DIR}/CudaTile.md
+         ${CMAKE_CURRENT_BINARY_DIR}/CudaTileOps.md
+diff --git a/lib/Bytecode/Translation/CMakeLists.txt b/lib/Bytecode/Translation/CMakeLists.txt
+index e8055b7..b783b4d 100644
+--- a/lib/Bytecode/Translation/CMakeLists.txt
++++ b/lib/Bytecode/Translation/CMakeLists.txt
+@@ -12,4 +12,5 @@ add_mlir_library(CudaTileBytecodeTranslation
+     CudaTileDialect
+     MLIRIR
+     MLIRPass
++    MLIRTranslateLib
+ )
+diff --git a/lib/Dialect/CudaTile/IR/CMakeLists.txt b/lib/Dialect/CudaTile/IR/CMakeLists.txt
+index 822a2c7..3934b30 100644
+--- a/lib/Dialect/CudaTile/IR/CMakeLists.txt
++++ b/lib/Dialect/CudaTile/IR/CMakeLists.txt
+@@ -23,4 +23,7 @@ add_mlir_library(CudaTileDialect
+   LINK_LIBS PUBLIC
+   MLIRIR
+   MLIRSideEffectInterfaces
++  MLIRFunctionInterfaces
++  MLIRInferTypeOpInterface
++  MLIRControlFlowInterfaces
+ )
+-- 
+2.39.3 (Apple Git-146)
+