From dc5024e88c0516217db916f2ba87f5ebff2697f6 Mon Sep 17 00:00:00 2001 From: Peter Boyle Date: Sat, 8 Jun 2019 13:39:11 +0100 Subject: [PATCH] The GPU reduction was not working for me and causing errors. Need to revisit. Gianluca is working on deterministic reduction/ --- Grid/lattice/Lattice_reduction.h | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-) diff --git a/Grid/lattice/Lattice_reduction.h b/Grid/lattice/Lattice_reduction.h index ba871d1f..94128ed5 100644 --- a/Grid/lattice/Lattice_reduction.h +++ b/Grid/lattice/Lattice_reduction.h @@ -59,8 +59,7 @@ inline ComplexD innerProduct(const Lattice &left,const Lattice &righ auto left_v = left.View(); auto right_v=right.View(); -#ifdef GRID_NVCC - +#if 0 typedef decltype(innerProduct(left_v[0],right_v[0])) inner_t; thrust::plus binary_sum; innerProductFunctor binary_inner_p; @@ -69,7 +68,6 @@ inline ComplexD innerProduct(const Lattice &left,const Lattice &righ // is there a way of using the efficient thrust reduction while maintaining memory coalescing? inner_t vnrm = thrust::inner_product(thrust::device, &left_v[0], &left_v[sN], &right_v[0], zero, binary_sum, binary_inner_p); nrm = Reduce(TensorRemove(vnrm));// sum across simd - #else thread_loop( (int thr=0;thrSumArraySize();thr++),{ int mywork, myoff;