[CUDA] Blackwell Support Codegen (#3103)

* Blackwell Support Codegen

10.0 blackwell b100/b200
12.0 blackwell rtx50

* update

---------

Co-authored-by: Ruihang Lai <[email protected]>

Author: johnnynunez

.github/workflows/documentation.yaml

@@ -11,7 +11,7 @@ jobs:
     runs-on: ubuntu-latest
 
     steps:
-    - uses: actions/checkout@v2
+    - uses: actions/checkout@v4
       with:
         submodules: recursive
 

.github/workflows/update-relax.yaml

@@ -14,7 +14,7 @@ jobs:
 
     steps:
     - name: Checkout
-      uses: actions/checkout@v3
+      uses: actions/checkout@v4
       with:
         submodules: true
 

.github/workflows/windows-build.yaml

@@ -25,7 +25,7 @@ jobs:
     - uses: actions/checkout@v3
       with:
         submodules: 'recursive'
-    - uses: conda-incubator/setup-miniconda@v2
+    - uses: conda-incubator/setup-miniconda@v3
       with:
         activate-environment: mlc-llm-build
         channel-priority: strict

ci/task/build_lib.sh

@@ -24,7 +24,7 @@ if [[ ${GPU} == rocm* ]]; then
 elif [[ ${GPU} == cuda* ]]; then
     echo set\(USE_VULKAN ON\) >>config.cmake
     echo set\(CMAKE_CUDA_COMPILER_LAUNCHER ccache\) >>config.cmake
-    echo set\(CMAKE_CUDA_ARCHITECTURES "80;90"\) >>config.cmake
+    echo set\(CMAKE_CUDA_ARCHITECTURES "80;90;100;120"\) >>config.cmake
     echo set\(CMAKE_CUDA_FLAGS \"\$\{CMAKE_CUDA_FLAGS\} -t $NUM_THREADS\"\) >>config.cmake
     echo set\(USE_CUDA ON\) >>config.cmake
     echo set\(USE_CUBLAS ON\) >>config.cmake

cmake/gen_cmake_config.py

@@ -83,12 +83,14 @@
     if use_flashInfer:
         while True:
             user_input = input("Enter your CUDA compute capability: ")
-            if user_input in ["80", "86", "89", "90"]:
+            if user_input in ["80", "86", "89", "90", "100", "120"]:
                 cmake_config_str += f"set(FLASHINFER_CUDA_ARCHITECTURES {user_input})\n"
                 cmake_config_str += f"set(CMAKE_CUDA_ARCHITECTURES {user_input})\n"
                 break
             else:
-                print(f"Invalid input: {user_input}. FlashInfer requires 80, 86, 89, or 90.")
+                print(
+                    f"Invalid input: {user_input}. FlashInfer requires 80, 86, 89, 90, 100 or 120"
+                )
 
     print("\nWriting the following configuration to config.cmake...")
     print(cmake_config_str)