update cuda archCoresPerSM

ginkgo-project · Nov 3, 2022 · 2ce0fb8 · 2ce0fb8
1 parent 1fd207a
commit 2ce0fb8
Showing 1 changed file with 4 additions and 1 deletion.
diff --git a/common/cuda_hip/base/executor.hpp.inc b/common/cuda_hip/base/executor.hpp.inc
@@ -34,7 +34,7 @@ namespace {
 
 
 // The function is copied from _ConvertSMVer2Cores of
-// cuda-11.0/samples/common/inc/helper_cuda.h
+// cuda-samples:Common/helper_cuda.h
 inline int convert_sm_ver_to_cores(int major, int minor)
 {
     // Defines for GPU Architecture types (using the SM version to determine
@@ -60,6 +60,9 @@ inline int convert_sm_ver_to_cores(int major, int minor)
         {0x72, 64},   // Volta Generation (SM 7.2) GV11b class
         {0x75, 64},   // Turing Generation (SM 7.5) TU1xx class
         {0x80, 64},   // Ampere Generation (SM 8.0) GA100 class
+        {0x86, 128},  // Ampere Generation (SM 8.6)
+        {0x87, 128},  // Ampere Generation (SM 8.7)
+        {0x90, 128},  // Hopper Generation (SM 9.0)
         {-1, -1}};
 
     int index = 0;