cwpearson/astaroth
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Enabled the generation of API hooks for calling DSL functions (was messing up with compilation earlier)

Browse Source
This commit is contained in:
jpekkila
2019-12-03 15:17:27 +02:00
parent 5a6a3110df
commit 7e4212ddd9
1 changed files with 8 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
src/core/kernels/integration.cu
View File

@@ -136,12 +136,10 @@ read_out(const int idx, AcReal* __restrict__ field[], const int3 handle)
\
const int idx = IDX(vertexIdx.x, vertexIdx.y, vertexIdx.z);
// clang-format off
/*
#define GEN_DEVICE_FUNC_HOOK(identifier) \
template <int step_number> \
AcResult acDeviceKernel_##identifier(const cudaStream_t stream, \
const int3 start, const int3 end, VertexBufferArray vba) \
AcResult acDeviceKernel_##identifier(const cudaStream_t stream, const int3 start, \
const int3 end, VertexBufferArray vba) \
{ \
\
const dim3 tpb(32, 1, 4); \
@@ -151,14 +149,11 @@ read_out(const int idx, AcReal* __restrict__ field[], const int3 handle)
(unsigned int)ceil(n.y / AcReal(tpb.y)), \
(unsigned int)ceil(n.z / AcReal(tpb.z))); \
\
identifier<step_number> \
<<<bpg, tpb, 0, stream>>>(start, end, vba); \
identifier<step_number><<<bpg, tpb, 0, stream>>>(start, end, vba); \
ERRCHK_CUDA_KERNEL(); \
\
return AC_SUCCESS; \
}
*/
#define GEN_DEVICE_FUNC_HOOK(identifier)
#include "user_kernels.h"
@@ -204,7 +199,8 @@ acKernelAutoOptimizeIntegration(const int3 start, const int3 end, VertexBufferAr
cudaEventCreate(&tstop);
// #ifdef AC_dt
//acDeviceLoadScalarUniform(device, STREAM_DEFAULT, AC_dt, FLT_EPSILON); // TODO note, temporarily disabled
// acDeviceLoadScalarUniform(device, STREAM_DEFAULT, AC_dt, FLT_EPSILON); // TODO
// note, temporarily disabled
/*#else
ERROR("FATAL ERROR: acDeviceAutoOptimize() or
acDeviceIntegrateSubstep() was " "called, but AC_dt was not defined. Either define
@@ -245,7 +241,8 @@ acKernelAutoOptimizeIntegration(const int3 start, const int3 end, VertexBufferAr
}
AcResult
acKernelIntegrateSubstep(const cudaStream_t stream, const int step_number, const int3 start, const int3 end, VertexBufferArray vba)
acKernelIntegrateSubstep(const cudaStream_t stream, const int step_number, const int3 start,
const int3 end, VertexBufferArray vba)
{
const dim3 tpb = rk3_tpb;