v4.3 update. (NVIDIA#2709)

* v4.3 update.

* Update the cute_dsl_api changelog's doc link

* Update version to 4.3.0

* Update the example link

* Update doc to encourage user to install DSL from requirements.txt

---------

Co-authored-by: Larry Wu <larwu@nvidia.com>

Author: Junkai-Wu

CHANGELOG.md

@@ -73,6 +73,16 @@ endif()
 
 include(${CMAKE_CURRENT_SOURCE_DIR}/CUDA.cmake)
 
+# nvcc supports response files with --options-file but some tools like clangd
+# might choke on it. Thus provide a way to control the use of this feature.
+set(CUTLASS_CUDA_USE_RESPONSE_FILE ON CACHE BOOL "Enable CUDA response files for includes, libraries, and objects")
+
+if(NOT CUTLASS_CUDA_USE_RESPONSE_FILE)
+  set(CMAKE_CUDA_USE_RESPONSE_FILE_FOR_INCLUDES 0)
+  set(CMAKE_CUDA_USE_RESPONSE_FILE_FOR_LIBRARIES 0)
+  set(CMAKE_CUDA_USE_RESPONSE_FILE_FOR_OBJECTS 0)
+endif()
+
 if (CUDA_VERSION VERSION_LESS 11.3)
   message(WARNING "CUTLASS ${CUTLASS_VERSION} requires CUDA 11.4 or higher, and strongly recommends CUDA 11.8 or higher.")
 elseif (CUDA_VERSION VERSION_LESS 11.4)
@@ -804,9 +814,9 @@ if(NOT WIN32)
   # Add common library search paths so executables and libraries can load and run
   # without LD_LIBRARY_PATH being set.
   link_libraries(
-    "-Wl,-rpath,'$ORIGIN'"
-    "-Wl,-rpath,'$ORIGIN/../lib64'"
-    "-Wl,-rpath,'$ORIGIN/../lib'"
+    "-Wl,-rpath,'$$ORIGIN'"
+    "-Wl,-rpath,'$$ORIGIN/../lib64'"
+    "-Wl,-rpath,'$$ORIGIN/../lib'"
     "-Wl,-rpath,'${CUDA_TOOLKIT_ROOT_DIR}/lib64'"
     "-Wl,-rpath,'${CUDA_TOOLKIT_ROOT_DIR}/lib'"
     ${CMAKE_DL_LIBS}
@@ -934,7 +944,7 @@ function(cutlass_add_executable_tests NAME TARGET)
 
      install(
        FILES ${__RESULT_CACHE_FILE}
-       DESTINATION ${CUTLASS_TEST_INSTALL_BINDIR}/
+       DESTINATION ${CUTLASS_TEST_INSTALL_BINDIR}
        )
 
     endif()
@@ -1062,7 +1072,7 @@ function(cutlass_generate_profiler_tests NAME)
 
   install(
     FILES ${CUTLASS_PROFILER_REGRESSION_LIST_FILE}
-    DESTINATION ${CMAKE_INSTALL_INFODIR}/cutlass/
+    DESTINATION ${CMAKE_INSTALL_INFODIR}/cutlass
     RENAME profiler_regressions.csv
     )
 

CMakeLists.txt

@@ -36,6 +36,7 @@ set(CUTLASS_PROFILER_REGRESSION_TEST_LEVEL  ${CUTLASS_TEST_LEVEL} CACHE STRING "
 
 find_package(Python3 3.5 COMPONENTS Interpreter REQUIRED)
 
+
 function(cutlass_generate_kernel_filter_and_testlist_files)
 
   set(options)

README.md

@@ -33,7 +33,7 @@ set(TEST_SWIZZLE_2 --swizzle=2)
 set(TEST_SWIZZLE_5 --swizzle=5)
 set(TEST_SWIZZLE_5_UNEVEN --swizzle=5 --m=4096 --n=16384)
 
-if(CUTLASS_NVCC_ARCHS STREQUAL "100a" OR CUTLASS_NVCC_ARCHS STREQUAL "100f" OR CUTLASS_NVCC_ARCHS STREQUAL "101a" OR CUTLASS_NVCC_ARCHS STREQUAL "101f" OR CUTLASS_NVCC_ARCHS STREQUAL "103a" OR CUTLASS_NVCC_ARCHS STREQUAL "103f")
+if(CUTLASS_NVCC_ARCHS MATCHES "100a|100f|101a|101f|103a|103f")
 cutlass_example_add_executable(
   70_blackwell_fp16_gemm
   70_blackwell_fp16_gemm.cu

customConfigs.cmake

@@ -27,7 +27,7 @@
 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 # Both filenames are shorter to avoid MAX_PATH issues on Windows.
-if(CUTLASS_NVCC_ARCHS STREQUAL "100a" OR CUTLASS_NVCC_ARCHS STREQUAL "100f" OR CUTLASS_NVCC_ARCHS STREQUAL "101a" OR CUTLASS_NVCC_ARCHS STREQUAL "101f" OR CUTLASS_NVCC_ARCHS STREQUAL "103a" OR CUTLASS_NVCC_ARCHS STREQUAL "103f")
+if(CUTLASS_NVCC_ARCHS MATCHES "100a|100f|101a|101f|103a|103f")
 cutlass_example_add_executable(
   71_blackwell_gemm_with_collective_builder
   71_blackwell_gemm_with_collective_builder.cu

examples/70_blackwell_gemm/CMakeLists.txt

@@ -28,7 +28,7 @@
 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
-if(CUTLASS_NVCC_ARCHS STREQUAL "100a" OR CUTLASS_NVCC_ARCHS STREQUAL "100f" OR CUTLASS_NVCC_ARCHS STREQUAL "101a" OR CUTLASS_NVCC_ARCHS STREQUAL "101f" OR CUTLASS_NVCC_ARCHS STREQUAL "103a" OR CUTLASS_NVCC_ARCHS STREQUAL "103f")
+if(CUTLASS_NVCC_ARCHS MATCHES "100a|100f|101a|101f|103a|103f")
 cutlass_example_add_executable(
   72a_blackwell_nvfp4_bf16_gemm
   72a_blackwell_nvfp4_bf16_gemm.cu

examples/71_blackwell_gemm_with_collective_builder/CMakeLists.txt

@@ -28,7 +28,7 @@
 
 
 
-if(CUTLASS_NVCC_ARCHS STREQUAL "100a" OR CUTLASS_NVCC_ARCHS STREQUAL "100f" OR CUTLASS_NVCC_ARCHS STREQUAL "101a" OR CUTLASS_NVCC_ARCHS STREQUAL "101f" OR CUTLASS_NVCC_ARCHS STREQUAL "103a" OR CUTLASS_NVCC_ARCHS STREQUAL "103f")
+if(CUTLASS_NVCC_ARCHS MATCHES "100a|100f|101a|101f|103a|103f")
 cutlass_example_add_executable(
   73_blackwell_gemm_preferred_cluster
   blackwell_gemm_preferred_cluster.cu

examples/72_blackwell_narrow_precision_gemm/CMakeLists.txt

@@ -29,7 +29,7 @@
 
 
 
-if(CUTLASS_NVCC_ARCHS STREQUAL "100a" OR CUTLASS_NVCC_ARCHS STREQUAL "100f" OR CUTLASS_NVCC_ARCHS STREQUAL "101a" OR CUTLASS_NVCC_ARCHS STREQUAL "101f" OR CUTLASS_NVCC_ARCHS STREQUAL "103a" OR CUTLASS_NVCC_ARCHS STREQUAL "103f")
+if(CUTLASS_NVCC_ARCHS MATCHES "100a|100f|101a|101f|103a|103f")
   cutlass_example_add_executable(
     74_blackwell_gemm_streamk
     blackwell_gemm_streamk.cu

examples/73_blackwell_gemm_preferred_cluster/CMakeLists.txt

@@ -49,7 +49,7 @@ set(TEST_SMALL_LARGE_GROUP --m=128 --n=128 --groups=50 --iterations=0)
 set(TEST_RANDOM_PERF --iterations=10)                                               # Random problem sizes
 set(TEST_RANDOM_PERF_LARGE_GROUP --groups=50 --iterations=10)                       # Random problem sizes
 
-if(CUTLASS_NVCC_ARCHS STREQUAL "100a")
+if("100a" IN_LIST CUTLASS_NVCC_ARCHS)
 cutlass_example_add_executable(
   75_blackwell_grouped_gemm
   75_blackwell_grouped_gemm.cu