diff --git a/csrc/kernels.cu b/csrc/kernels.cu index df8488389..6d15dbe64 100644 --- 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;