[Common] NVFP4 kernels

build_tools/utils.py

@@ -237,9 +237,9 @@ def cuda_archs() -> str:
     version = cuda_version()
     if os.getenv("NVTE_CUDA_ARCHS") is None:
         if version >= (13, 0):
-            os.environ["NVTE_CUDA_ARCHS"] = "75;80;89;90;100;120"
+            os.environ["NVTE_CUDA_ARCHS"] = "100a"
         elif version >= (12, 8):
-            os.environ["NVTE_CUDA_ARCHS"] = "70;80;89;90;100;120"
+            os.environ["NVTE_CUDA_ARCHS"] = "100a"
         else:
             os.environ["NVTE_CUDA_ARCHS"] = "70;80;89;90"
     return os.getenv("NVTE_CUDA_ARCHS")

tests/cpp/CMakeLists.txt

@@ -6,7 +6,7 @@ cmake_minimum_required(VERSION 3.18)
 
 if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
   if (CUDAToolkit_VERSION VERSION_GREATER_EQUAL 12.8)
-    set(CMAKE_CUDA_ARCHITECTURES 75 80 89 90 100 120)
+    set(CMAKE_CUDA_ARCHITECTURES 100)
   else ()
     set(CMAKE_CUDA_ARCHITECTURES 75 80 89 90)
   endif()
@@ -26,7 +26,7 @@ enable_testing()
 include_directories(${gtest_SOURCE_DIR}/include ${gtest_SOURCE_DIR})
 
 if(NOT DEFINED TE_LIB_PATH)
-    execute_process(COMMAND bash -c "pip3 show transformer-engine | grep Location | cut -d ' ' -f 2 | tr -d '\n'"
+    execute_process(COMMAND bash -c "pip3 show transformer-engine | grep 'Editable project location' | cut -d ' ' -f 4 | tr -d '\n'"
                     OUTPUT_VARIABLE TE_LIB_PATH)
 endif()
 

tests/cpp/operator/CMakeLists.txt

@@ -3,29 +3,30 @@
 # See LICENSE for license information.
 
 add_executable(test_operator
-               test_cast.cu
-               test_cast_current_scaling.cu
-               test_cast_dbias.cu
-               test_cast_dbias_dgelu.cu
-               test_cast_gated_swiglu.cu
-               test_cast_mxfp8_gated_swiglu.cu
-               test_qdq.cu
-               test_cast_mxfp8.cu
-               test_cast_float8blockwise.cu
-               test_dequantize_mxfp8.cu
-               test_transpose.cu
-               test_cast_transpose.cu
-               test_cast_transpose_current_scaling.cu
-               test_cast_transpose_dbias.cu
-               test_cast_transpose_dbias_dgelu.cu
-               test_cast_transpose_dgeglu.cu
-               test_act.cu
-               test_normalization.cu
-               test_normalization_mxfp8.cu
-               test_multi_cast_transpose.cu
-               test_multi_padding.cu
-               test_causal_softmax.cu
-               test_swizzle.cu
+            #    test_cast.cu
+            #    test_cast_current_scaling.cu
+            #    test_cast_dbias.cu
+            #    test_cast_dbias_dgelu.cu
+            #    test_cast_gated_swiglu.cu
+               # test_cast_mxfp8_gated_swiglu.cu
+            #    test_qdq.cu
+               # test_cast_mxfp8.cu
+               test_cast_nvfp4.cu
+            #    test_cast_float8blockwise.cu
+            #    test_dequantize_mxfp8.cu
+            #    test_transpose.cu
+            #    test_cast_transpose.cu
+            #    test_cast_transpose_current_scaling.cu
+            #    test_cast_transpose_dbias.cu
+            #    test_cast_transpose_dbias_dgelu.cu
+            #    test_cast_transpose_dgeglu.cu
+            #    test_act.cu
+            #    test_normalization.cu
+            #    test_normalization_mxfp8.cu
+            #    test_multi_cast_transpose.cu
+            #    test_multi_padding.cu
+            #    test_causal_softmax.cu
+            #    test_swizzle.cu
                ../test_common.cu)
 
 find_package(OpenMP REQUIRED)