[FRONTEND] Update PTX/SM support for LLVM14 (PR #1038 redux) (#1039)

=

python/triton/compiler.py

@@ -963,23 +963,12 @@ def ptx_get_version(cuda_version) -> int:
     '''
     assert isinstance(cuda_version, str)
     major, minor = map(int, cuda_version.split('.'))
-    version = major * 1000 + minor * 10
-    if version >= 11040:
-        return 74
-    if version >= 11030:
-        return 73
-    if version >= 11020:
-        return 72
-    if version >= 11010:
-        return 71
-    if version >= 11000:
-        return 70
-    if version >= 10020:
-        return 65
-    if version >= 10010:
-        return 64
-    if version >= 10000:
-        return 63
+    if major == 12:
+        return 80 + minor
+    if major == 11:
+        return 70 + minor
+    if major == 10:
+        return 63 + minor
     raise RuntimeError("Triton only support CUDA 10.0 or higher")