diff options
author | Hieu Hoang <hieuhoang@gmail.com> | 2017-12-07 04:00:08 +0300 |
---|---|---|
committer | Hieu Hoang <hieuhoang@gmail.com> | 2017-12-07 04:00:08 +0300 |
commit | 60d4266822ca228199bc3027fc82589c83a384a6 (patch) | |
tree | 5662c5ae8ec617818d266d317c27f3373117fde2 | |
parent | 55ac6ae8fb11041c945b70bf27247c31e6665b49 (diff) |
delete all mention of half2halfC
-rw-r--r-- | src/amun/half/mblas/matrix_functions.h | 7 | ||||
-rw-r--r-- | src/amun/half/mblas/thrust_functions.h | 19 |
2 files changed, 4 insertions, 22 deletions
diff --git a/src/amun/half/mblas/matrix_functions.h b/src/amun/half/mblas/matrix_functions.h index 8a7a3037..fdf92ccb 100644 --- a/src/amun/half/mblas/matrix_functions.h +++ b/src/amun/half/mblas/matrix_functions.h @@ -216,17 +216,18 @@ __global__ void gBroadcast(Functor functor, uint batchIdx = batchMappingWrap[ beamIdx ]; - //outWrap[id] = functor(in1Wrap[(batchIdx * srcSize + srcId) * cols + stateIdx], - // in2Wrap[beamIdx * cols + stateIdx]); + outWrap[id] = functor(in1Wrap[(batchIdx * srcSize + srcId) * cols + stateIdx], + in2Wrap[beamIdx * cols + stateIdx]); //outWrap[id] = functor(in1Wrap(indices[0], indices[1], 0, batchIdx), // in2Wrap(indices[2], indices[1], 0, 0)); //outWrap(srcId, stateIdx, beamIdx, 0) = functor(in1Wrap(srcId, stateIdx, 0, batchIdx), // in2Wrap(beamIdx, stateIdx, 0, 0)); + /* const half *in1 = &in1Wrap(srcId, stateIdx, 0, batchIdx); const half *in2 = &in2Wrap(beamIdx, stateIdx, 0, 0); half *out = &outWrap(srcId, stateIdx, beamIdx, 0); *out = functor(*in1, *in2); - + */ } } diff --git a/src/amun/half/mblas/thrust_functions.h b/src/amun/half/mblas/thrust_functions.h index 3373b1e5..ba36f662 100644 --- a/src/amun/half/mblas/thrust_functions.h +++ b/src/amun/half/mblas/thrust_functions.h @@ -20,19 +20,6 @@ inline half htanh(const half x) return ret; } -__device__ -inline half2 htanh(const half2 x) -{ - half2 one = __float2half2_rn(1.0f); - half2 t1 = h2exp(__hmul2(__float2half2_rn(2.0f), x)); - half2 t2 = __hsub2(one, t1); - half2 t3 = __hadd2(one, t1); - t3 = h2rcp(t3); - half2 ret = __hmul2(t1, t3); - - return ret; -} - namespace thrust { namespace detail @@ -75,12 +62,6 @@ namespace thrust T operator()(const T &x, const T &y) const { return x + y; } }; - template<> - struct half_binary_add<half2> : public thrust::binary_function<half2,half2,half2> { - __device__ - half2 operator()(const half2 &x, const half2 &y) const { return __hadd2(x, y); } - }; - template<typename Eval1, typename Eval2> __host__ __device__ actor<composite<binary_operator<half_binary_add>, actor<Eval1>, actor<Eval2>>> |