[SYCL][NVPTX] Fix NVVM annotation for min_work_groups_per_cu attribute (

#14418)

While the PTX directive is "minnctapersm", the NVVM annotation is
"minctasm". This can be seen in the NVPTX backend, and in the NVVM IR
spec, section 13.3.

clang/lib/CodeGen/Targets/NVPTX.cpp

@@ -275,7 +275,7 @@ void NVPTXTargetCodeGenInfo::setTargetAttributes(
             << MWGPCU << 0;
       } else {
         // The value is guaranteed to be > 0, pass it to the metadata.
-        addNVVMMetadata(F, "minnctapersm", attrValue(MWGPCU->getValue()));
+        addNVVMMetadata(F, "minctasm", attrValue(MWGPCU->getValue()));
         HasMinWorkGroupPerCU = true;
       }
     }

clang/test/CodeGenSYCL/launch_bounds_nvptx.cpp

@@ -4,7 +4,7 @@
 // compute unit and maximum work groups per multi-processor attributes, that
 // correspond to CUDA's launch bounds. Expect max_work_group_size,
 // min_work_groups_per_cu and max_work_groups_per_mp that are mapped to
-// maxntidx, minnctapersm, maxclusterrank PTX directives respectively.
+// maxntidx, minctasm, and maxclusterrank NVVM annotations respectively.
 
 #include "sycl.hpp"
 
@@ -66,34 +66,34 @@ int main() {
 // CHECK: define dso_local void @{{.*}}kernel_name5() #0 {{.*}} !min_work_groups_per_cu ![[MWGPC_MWGPM_2:[0-9]+]] !max_work_groups_per_mp ![[MWGPC_MWGPM_2]] !max_work_group_size ![[MWGS_3:[0-9]+]]
 
 // CHECK: {{.*}}@{{.*}}kernel_name1, !"maxntidx", i32 512}
-// CHECK: {{.*}}@{{.*}}kernel_name1, !"minnctapersm", i32 2}
+// CHECK: {{.*}}@{{.*}}kernel_name1, !"minctasm", i32 2}
 // CHECK: {{.*}}@{{.*}}kernel_name1, !"maxclusterrank", i32 4}
 // CHECK: {{.*}}@{{.*}}Foo{{.*}}, !"maxntidx", i32 512}
-// CHECK: {{.*}}@{{.*}}Foo{{.*}}, !"minnctapersm", i32 2}
+// CHECK: {{.*}}@{{.*}}Foo{{.*}}, !"minctasm", i32 2}
 // CHECK: {{.*}}@{{.*}}Foo{{.*}}, !"maxclusterrank", i32 4}
 // CHECK: {{.*}}@{{.*}}kernel_name2, !"maxntidx", i32 512}
-// CHECK: {{.*}}@{{.*}}kernel_name2, !"minnctapersm", i32 2}
+// CHECK: {{.*}}@{{.*}}kernel_name2, !"minctasm", i32 2}
 // CHECK: {{.*}}@{{.*}}kernel_name2, !"maxclusterrank", i32 4}
 // CHECK: {{.*}}@{{.*}}main{{.*}}, !"maxntidx", i32 512}
-// CHECK: {{.*}}@{{.*}}main{{.*}}, !"minnctapersm", i32 2}
+// CHECK: {{.*}}@{{.*}}main{{.*}}, !"minctasm", i32 2}
 // CHECK: {{.*}}@{{.*}}main{{.*}}, !"maxclusterrank", i32 4}
 // CHECK: {{.*}}@{{.*}}kernel_name3, !"maxntidx", i32 384}
-// CHECK: {{.*}}@{{.*}}kernel_name3, !"minnctapersm", i32 6}
+// CHECK: {{.*}}@{{.*}}kernel_name3, !"minctasm", i32 6}
 // CHECK: {{.*}}@{{.*}}kernel_name3, !"maxclusterrank", i32 6}
 // CHECK: {{.*}}@{{.*}}Functor{{.*}}, !"maxntidx", i32 384}
-// CHECK: {{.*}}@{{.*}}Functor{{.*}}, !"minnctapersm", i32 6}
+// CHECK: {{.*}}@{{.*}}Functor{{.*}}, !"minctasm", i32 6}
 // CHECK: {{.*}}@{{.*}}Functor{{.*}}, !"maxclusterrank", i32 6}
 // CHECK: {{.*}}@{{.*}}kernel_name4, !"maxntidx", i32 512}
-// CHECK: {{.*}}@{{.*}}kernel_name4, !"minnctapersm", i32 2}
+// CHECK: {{.*}}@{{.*}}kernel_name4, !"minctasm", i32 2}
 // CHECK: {{.*}}@{{.*}}kernel_name4, !"maxclusterrank", i32 4}
 // CHECK: {{.*}}@{{.*}}bar{{.*}}, !"maxntidx", i32 512}
-// CHECK: {{.*}}@{{.*}}bar{{.*}}, !"minnctapersm", i32 2}
+// CHECK: {{.*}}@{{.*}}bar{{.*}}, !"minctasm", i32 2}
 // CHECK: {{.*}}@{{.*}}bar{{.*}}, !"maxclusterrank", i32 4}
 // CHECK: {{.*}}@{{.*}}kernel_name5, !"maxntidx", i32 1024}
-// CHECK: {{.*}}@{{.*}}kernel_name5, !"minnctapersm", i32 16}
+// CHECK: {{.*}}@{{.*}}kernel_name5, !"minctasm", i32 16}
 // CHECK: {{.*}}@{{.*}}kernel_name5, !"maxclusterrank", i32 16}
 // CHECK: {{.*}}@{{.*}}zoo{{.*}}, !"maxntidx", i32 1024}
-// CHECK: {{.*}}@{{.*}}zoo{{.*}}, !"minnctapersm", i32 16}
+// CHECK: {{.*}}@{{.*}}zoo{{.*}}, !"minctasm", i32 16}
 // CHECK: {{.*}}@{{.*}}zoo{{.*}}, !"maxclusterrank", i32 16}
 
 // CHECK: ![[MWGPC]] = !{i32 2}