Fix race condition in kEstimateQuantiles (#1061)

5b28fd3f · pnunna93 · GitHub · 344e8516 · 5b28fd3f
Unverified Commit 5b28fd3f authored Feb 14, 2024 by pnunna93 Committed by GitHub Feb 14, 2024
Show whitespace changes
Inline Side-by-side

Showing with 2 additions and 0 deletions

csrc/kernels.cu csrc/kernels.cu +2 -0

No files found.
--- a/csrc/kernels.cu
+++ b/csrc/kernels.cu
@@ -654,6 +654,8 @@ __global__ void kEstimateQuantiles(T *__restrict__ const A, float *code, const f
      for(int j = threadIdx.x; j < BLOCK_ESTIMATE; j+=blockDim.x)
          temp_storage.smem_qidx[j] = -1;

+      __syncthreads();
+
      if(threadIdx.x < 256)
      {
          float q_interval = (1.0f-(2.0f*offset))/255.0f;