[Openmp-commits] [openmp] 9f69d3c - [Libomptarget] Use NVPTX lane id intrinsic in DeviceRTL (#84928)

[via Openmp-commits]

Author: Joseph Huber
Date: 2024-03-12T10:39:40-05:00
New Revision: 9f69d3cf88905df5006f93dce536b7e73c0b1735

URL: https://github.com/llvm/llvm-project/commit/9f69d3cf88905df5006f93dce536b7e73c0b1735
DIFF: https://github.com/llvm/llvm-project/commit/9f69d3cf88905df5006f93dce536b7e73c0b1735.diff

LOG: [Libomptarget] Use NVPTX lane id intrinsic in DeviceRTL (#84928)

Summary:
We are currently taking the lower 5 bites of the thread ID as the warp
ID. This doesn't work in non-1D grids and is also slower than just using
the dedicated hardware register.

Added: 
    

Modified: 
    openmp/libomptarget/DeviceRTL/src/Mapping.cpp

Removed: 
    


################################################################################
diff  --git a/openmp/libomptarget/DeviceRTL/src/Mapping.cpp b/openmp/libomptarget/DeviceRTL/src/Mapping.cpp
index 31dd8054dec33e..b2028a8fb4f506 100644
--- a/openmp/libomptarget/DeviceRTL/src/Mapping.cpp
+++ b/openmp/libomptarget/DeviceRTL/src/Mapping.cpp
@@ -172,10 +172,7 @@ uint32_t getThreadIdInBlock(int32_t Dim) {
   UNREACHABLE("Dim outside range!");
 }
 
-uint32_t getThreadIdInWarp() {
-  return impl::getThreadIdInBlock(mapping::DIM_X) &
-         (mapping::getWarpSize() - 1);
-}
+uint32_t getThreadIdInWarp() { return __nvvm_read_ptx_sreg_laneid(); }
 
 uint32_t getBlockIdInKernel(int32_t Dim) {
   switch (Dim) {