[libspirv] Restore missing builtin removed in 1b6070c

intel · Feb 27, 2025 · e40b162 · e40b162
1 parent 2b1e0ae
commit e40b162
Showing 1 changed file with 16 additions and 0 deletions.
diff --git a/libclc/libspirv/lib/ptx-nvidiacl/group/group_non_uniform.cl b/libclc/libspirv/lib/ptx-nvidiacl/group/group_non_uniform.cl
@@ -8,6 +8,7 @@
 
 #include "membermask.h"
 
+#include <clc/integer/clc_popcount.h>
 #include <libspirv/spirv.h>
 #include <libspirv/spirv_types.h>
 
@@ -33,4 +34,19 @@ _Z29__spirv_GroupNonUniformBallotjb(unsigned flag, bool predicate) {
   res[0] = __nvvm_vote_ballot_sync(threads, predicate);
 
   return res;
+}
+
+_CLC_DEF _CLC_CONVERGENT uint
+_Z37__spirv_GroupNonUniformBallotBitCountN5__spv5Scope4FlagEiDv4_j(
+    uint scope, uint flag, __clc_vec4_uint32_t mask) {
+  // here we assume scope == __spv::Scope::Subgroup
+  // flag == InclusiveScan is not yet implemented
+  if (flag == Reduce) {
+    return __clc_popcount(mask[0]);
+  } else if (flag == ExclusiveScan) {
+    return __clc_popcount(__nvvm_read_ptx_sreg_lanemask_lt() & mask[0]);
+  } else {
+    __builtin_trap();
+    __builtin_unreachable();
+  }
 }