Add thread fence in kernel to fix Volta indeterminacy (4a51e166) · Commits · 郑智淋 / lammps

lib/gpu/lal_lj_tip4p_long.cu

+3 −0

Original line number	Diff line number	Diff line
		@@ -271,6 +271,7 @@ __kernel void k_lj_tip4p_long(const __global numtyp4 *restrict x_,
		iH2 = hneigh[i*4+1];
		if(fabs(m[iO].w) <= eq_zero) {
		compute_newsite(iO,iH1,iH2, &m[iO], alpha, x_);
		__threadfence();
		m[iO].w = qtmp;
		}
		x1 = m[iO];
		@@ -283,6 +284,7 @@ __kernel void k_lj_tip4p_long(const __global numtyp4 *restrict x_,
		if(fabs(m[iO].w) <= eq_zero) {
		compute_newsite(iO,iH1,iH2, &m[iO], alpha, x_);
		numtyp qO; fetch(qO,iO,q_tex);
		__threadfence();
		m[iO].w = qO;
		}
		}
		@@ -341,6 +343,7 @@ __kernel void k_lj_tip4p_long(const __global numtyp4 *restrict x_,
		jH2 = hneigh[j*4+1];
		if (fabs(m[j].w) <= eq_zero) {
		compute_newsite(j, jH1, jH2, &m[j], alpha, x_);
		__threadfence();
		m[j].w = qj;
		}
		x2 = m[j];