Benchmarks: Micro benchmarks - add support for NVIDIA L4/L40/L40s GPUs in gemm-flops (#634)

**Description**
Add support GPU ARCH 8.9 for NVIDIA L4/L40/L40s GPUs in gemm-flops.

superbench/benchmarks/micro_benchmarks/cuda_common.cmake

@@ -33,6 +33,6 @@ if(NOT DEFINED NVCC_ARCHS_SUPPORTED)
       list(APPEND NVCC_ARCHS_SUPPORTED 86)
     endif()
     if (NOT CMAKE_CUDA_COMPILER_VERSION VERSION_LESS 11.8)
-      list(APPEND NVCC_ARCHS_SUPPORTED 90)
+      list(APPEND NVCC_ARCHS_SUPPORTED 89 90)
     endif()
 endif()

superbench/benchmarks/micro_benchmarks/cuda_gemm_flops_performance.py

@@ -47,6 +47,8 @@ class CudaGemmFlopsBenchmark(GemmFlopsBenchmark):
         # Skip FP64 for RTX Turing/Ampere and Tesla T4/GA10x due to very limited FP64 TFLOP rate
         self.__kernel_map[7.5] = {k: self.__kernel_map[7.0][k] for k in self.__kernel_map[7.0] if 'fp64' not in k}
         self.__kernel_map[8.6] = {k: self.__kernel_map[8.0][k] for k in self.__kernel_map[8.0] if 'fp64' not in k}
+        # Skip FP64 for Ada Lovelace L4/L40 due to no native CUDA/Tensor Cores
+        self.__kernel_map[8.9] = {k: self.__kernel_map[8.0][k] for k in self.__kernel_map[8.0] if 'fp64' not in k}
         # Skip INT4 for Hopper due to no native CUDA/Tensor Cores
         self.__kernel_map[9.0] = {k: self.__kernel_map[8.0][k] for k in self.__kernel_map[8.0] if 'int4_tc' not in k}
         self.__parse_logline = [

third_party/Makefile

@@ -33,7 +33,7 @@ sb_micro_path:
 # Build cutlass.
 cuda_cutlass:
 ifeq ($(shell echo $(CUDA_VER)">=11.8" | bc -l), 1)
-	$(eval ARCHS := "70;75;80;86;90")
+	$(eval ARCHS := "70;75;80;86;89;90")
 else
 	$(eval ARCHS := "70;75;80;86")
 endif