AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

SWDEV-548892 - Stop using __ockl_lane_id (#2186)

Browse Source 
__lane_id already exists and is identical.
This commit is contained in:
Matt Arsenault
2025-12-19 20:34:55 +01:00
committed by GitHub
parent 7c989ac022
commit 0c0d8dc974
3 changed files with 9 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/clr/hipamd/include/hip/amd_detail/amd_device_functions.h
+3 -3
View File
@@ -640,7 +640,7 @@ __device__ inline void __named_sync() { __builtin_amdgcn_s_barrier(); }
// hip.amdgcn.bc - lanemask
__device__ inline __hip_uint64_t __lanemask_gt() {
__hip_uint32_t lane = __ockl_lane_u32();
__hip_uint32_t lane = __lane_id();
if (lane == 63) return 0;
__hip_uint64_t ballot = __ballot64(1);
__hip_uint64_t mask = (~((__hip_uint64_t)0)) << (lane + 1);
@@ -648,14 +648,14 @@ __device__ inline __hip_uint64_t __lanemask_gt() {
}
__device__ inline __hip_uint64_t __lanemask_lt() {
__hip_uint32_t lane = __ockl_lane_u32();
__hip_uint32_t lane = __lane_id();
__hip_int64_t ballot = __ballot64(1);
__hip_uint64_t mask = ((__hip_uint64_t)1 << lane) - (__hip_uint64_t)1;
return mask & ballot;
}
__device__ inline __hip_uint64_t __lanemask_eq() {
__hip_uint32_t lane = __ockl_lane_u32();
__hip_uint32_t lane = __lane_id();
__hip_int64_t mask = ((__hip_uint64_t)1 << lane);
return mask;
}
projects/clr/hipamd/include/hip/amd_detail/amd_warp_sync_functions.h
+1 -1
View File
@@ -357,7 +357,7 @@ __device__ inline T __reduce_op_sync(MaskT mask, T val, BinaryOp op, WfReduce wf
#endif
firstLane = __builtin_ctzll(mask);
laneId = __ockl_lane_u32();
laneId = __lane_id();
nextBit = laneId;
// the number of iterations needs to be at least log2(number of bits on)
numIterations = sizeof(int) * 8 - __clz(maskNumBits);