    

 ## On this page

  

 

 # api\_cuda\_sph

 Last update: 16.07.2025 

<a id="l00001" name="l00001"></a> 1// (C) 2024 ANSYS, Inc. Unauthorized use, distribution, or duplication is prohibited.

<a id="l00002" name="l00002"></a> 2\#pragma once

<a id="l00003" name="l00003"></a> 3 

<a id="l00004" name="l00004"></a> 4// Includes ---------------------------------------------------------------------------------------

<a id="l00005" name="l00005"></a> 5\#include &lt;rocky20/api/device/api\_cuda.hpp&gt;

<a id="l00006" name="l00006"></a> 6\#include &lt;rocky20/api/sph/api\_device\_sph.hpp&gt;

<a id="l00007" name="l00007"></a> 7\#include &lt;rocky20/sph/sph\_model.h&gt;

<a id="l00008" name="l00008"></a> 8 

<a id="l00009" name="l00009"></a> 9namespace rocky::cuda {

<a id="l00010" name="l00010"></a> 10 

<a id="l00011" name="l00011"></a> 11template &lt;typename Operator, typename Functor&gt;

<a id="l00012" name="l00012"></a> 12__global__ void execute_sph_particles_function(

<a id="l00013" name="l00013"></a> 13 SDeviceModel* device_data,

<a id="l00014" name="l00014"></a> 14 SPHDeviceModel* sph_device,

<a id="l00015" name="l00015"></a> 15 double current_time,

<a id="l00016" name="l00016"></a> 16 Functor sph_particles_f,

<a id="l00017" name="l00017"></a> 17 void* udf_device_data)

<a id="l00018" name="l00018"></a> 18{

<a id="l00019" name="l00019"></a> 19 device::launch_sph_particles_function&lt;CudaBackend, Operator&gt;(

<a id="l00020" name="l00020"></a> 20 device_data, sph_device, current_time, sph_particles_f, udf_device_data);

<a id="l00021" name="l00021"></a> 21}

<a id="l00022" name="l00022"></a> 22 

<a id="l00023" name="l00023"></a> 23template &lt;typename Operator, typename Functor, class SPHNeighborsHandler&gt;

<a id="l00024" name="l00024"></a> 24__global__ void execute_sph_interactions_function(

<a id="l00025" name="l00025"></a> 25 SDeviceModel* device_data,

<a id="l00026" name="l00026"></a> 26 SPHDeviceModel* sph_device,

<a id="l00027" name="l00027"></a> 27 double current_time,

<a id="l00028" name="l00028"></a> 28 Functor sph_particle_interaction_f,

<a id="l00029" name="l00029"></a> 29 void* udf_device_data,

<a id="l00030" name="l00030"></a> 30 IterationType it_type)

<a id="l00031" name="l00031"></a> 31{

<a id="l00032" name="l00032"></a> 32 // SPH -&gt; SPH

<a id="l00033" name="l00033"></a> 33 SPHNeighborsHandler handler { device_data, sph_device, (float) current_time, it_type };

<a id="l00034" name="l00034"></a> 34 device::launch_sph_interactions_particles&lt;CudaBackend, Operator, Functor&gt;(

<a id="l00035" name="l00035"></a> 35 device_data,

<a id="l00036" name="l00036"></a> 36 sph_device,

<a id="l00037" name="l00037"></a> 37 current_time,

<a id="l00038" name="l00038"></a> 38 sph_particle_interaction_f,

<a id="l00039" name="l00039"></a> 39 udf_device_data,

<a id="l00040" name="l00040"></a> 40 handler);

<a id="l00041" name="l00041"></a> 41}

<a id="l00042" name="l00042"></a> 42 

<a id="l00043" name="l00043"></a> 43template &lt;typename Operator, typename Functor&gt;

<a id="l00044" name="l00044"></a> 44__global__ void execute_sph_interactions_triangle_function(

<a id="l00045" name="l00045"></a> 45 SDeviceModel* device_data,

<a id="l00046" name="l00046"></a> 46 SPHDeviceModel* sph_device,

<a id="l00047" name="l00047"></a> 47 double current_time,

<a id="l00048" name="l00048"></a> 48 Functor sph_triangle_interaction_f,

<a id="l00049" name="l00049"></a> 49 void* udf_device_data)

<a id="l00050" name="l00050"></a> 50{

<a id="l00051" name="l00051"></a> 51 // SPH -&gt; triangles

<a id="l00052" name="l00052"></a> 52 device::launch_sph_interactions_triangles&lt;CudaBackend, Operator&gt;(

<a id="l00053" name="l00053"></a> 53 device_data, sph_device, current_time, sph_triangle_interaction_f, udf_device_data);

<a id="l00054" name="l00054"></a> 54}

<a id="l00055" name="l00055"></a> 55 

<a id="l00056" name="l00056"></a> 56// Dispatchers ------------------------------------------------------------------------------------

<a id="l00057" name="l00057"></a> 57template &lt;typename Operator = device::DefaultOperator&lt;CudaBackend&gt;, typename Functor&gt;

<a id="l00058" name="l00058"></a> 58void dispatch_sph_elements_kernel(

<a id="l00059" name="l00059"></a> 59 [IRockyModel](structIRockyModel.xhtml)&amp; model, Functor&amp; sph_particles_f, void** udf_devices_data)

<a id="l00060" name="l00060"></a> 60{

<a id="l00061" name="l00061"></a> 61 SPHModel* sph_model = &amp;(*model._impl-&gt;sph_model);

<a id="l00062" name="l00062"></a> 62 cuda_host&amp; cu_hos = *model._impl;

<a id="l00063" name="l00063"></a> 63 

<a id="l00064" name="l00064"></a> 64 if (sph_model-&gt;gen_sph.nprt_active &lt;= 0)

<a id="l00065" name="l00065"></a> 65 return;

<a id="l00066" name="l00066"></a> 66 

<a id="l00067" name="l00067"></a> 67 void* udf_device_data = nullptr;

<a id="l00068" name="l00068"></a> 68 

<a id="l00069" name="l00069"></a> 69 // Get the number of active particles to launch kernel

<a id="l00070" name="l00070"></a> 70 std::vector&lt;int&gt; n_active_particles(model.[get\_number\_of\_devices](structIRockyModel.xhtml#a1fd1011aa809fe64647e9cacf1956a11)(), 0);

<a id="l00071" name="l00071"></a> 71 #pragma omp parallel for num\_threads(model.get\_number\_of\_devices()) if (model.\_impl-&gt;gpu\_parallel)

<a id="l00072" name="l00072"></a> 72 for (int i = 0; i &lt; model.[get\_number\_of\_devices](structIRockyModel.xhtml#a1fd1011aa809fe64647e9cacf1956a11)(); ++i)

<a id="l00073" name="l00073"></a> 73 {

<a id="l00074" name="l00074"></a> 74 cuda_general_sph gen_sph_tmp;

<a id="l00075" name="l00075"></a> 75 CUDA_SET_DEVICE(model.[get\_device\_id](structIRockyModel.xhtml#ae8f8865c809a31988291f9180d7cda2b)(i));

<a id="l00076" name="l00076"></a> 76 CUDA_COPY_D2H(&amp;gen_sph_tmp, sph_model-&gt;get_device_general(i), 1);

<a id="l00077" name="l00077"></a> 77 n_active_particles[i] = gen_sph_tmp.nprt_active;

<a id="l00078" name="l00078"></a> 78 }

<a id="l00079" name="l00079"></a> 79 

<a id="l00080" name="l00080"></a> 80 // Launching user kernel forces

<a id="l00081" name="l00081"></a> 81 for (int i = 0; i &lt; model.[get\_number\_of\_devices](structIRockyModel.xhtml#a1fd1011aa809fe64647e9cacf1956a11)(); ++i)

<a id="l00082" name="l00082"></a> 82 {

<a id="l00083" name="l00083"></a> 83 CUDA_SET_DEVICE(model.[get\_device\_id](structIRockyModel.xhtml#ae8f8865c809a31988291f9180d7cda2b)(i));

<a id="l00084" name="l00084"></a> 84 

<a id="l00085" name="l00085"></a> 85 SPHDeviceModel* sph_device = model._impl-&gt;sph_model-&gt;get_device_data(i);

<a id="l00086" name="l00086"></a> 86 SDeviceModel* device_data = model.get_device_data(i);

<a id="l00087" name="l00087"></a> 87 if (udf_devices_data != nullptr)

<a id="l00088" name="l00088"></a> 88 {

<a id="l00089" name="l00089"></a> 89 udf_device_data = udf_devices_data[i];

<a id="l00090" name="l00090"></a> 90 }

<a id="l00091" name="l00091"></a> 91 

<a id="l00092" name="l00092"></a> 92 auto [blocks, threads] = CudaBackend::get_launch_params(n_active_particles[i]);

<a id="l00093" name="l00093"></a> 93 rocky::cuda::execute_sph_particles_function&lt;Operator&gt;&lt;&lt;&lt;blocks, threads&gt;&gt;&gt;(

<a id="l00094" name="l00094"></a> 94 device_data,

<a id="l00095" name="l00095"></a> 95 sph_device,

<a id="l00096" name="l00096"></a> 96 cu_hos.iteration.current_time,

<a id="l00097" name="l00097"></a> 97 sph_particles_f,

<a id="l00098" name="l00098"></a> 98 udf_device_data);

<a id="l00099" name="l00099"></a> 99 }

<a id="l00100" name="l00100"></a> 100 

<a id="l00101" name="l00101"></a> 101 // Syncing all GPUs

<a id="l00102" name="l00102"></a> 102 sync_devices(model);

<a id="l00103" name="l00103"></a> 103}

<a id="l00104" name="l00104"></a> 104 

<a id="l00105" name="l00105"></a> 105template &lt;typename Operator = device::DefaultElementsOperator&lt;CudaBackend&gt;, typename Functor&gt;

<a id="l00106" name="l00106"></a> 106void dispatch_sph_interactions_elements_kernel(

<a id="l00107" name="l00107"></a> 107 [IRockyModel](structIRockyModel.xhtml)&amp; model,

<a id="l00108" name="l00108"></a> 108 Functor&amp; sph_particle_interaction_f,

<a id="l00109" name="l00109"></a> 109 void** udf_devices_data,

<a id="l00110" name="l00110"></a> 110 IterationType it_type = itAll)

<a id="l00111" name="l00111"></a> 111{

<a id="l00112" name="l00112"></a> 112 SPHModel* sph_model = &amp;(*model._impl-&gt;sph_model);

<a id="l00113" name="l00113"></a> 113 cuda_host&amp; cu_hos = *model._impl;

<a id="l00114" name="l00114"></a> 114 

<a id="l00115" name="l00115"></a> 115 if (sph_model-&gt;gen_sph.nprt_active &lt;= 0)

<a id="l00116" name="l00116"></a> 116 return;

<a id="l00117" name="l00117"></a> 117 

<a id="l00118" name="l00118"></a> 118 void* udf_device_data = nullptr;

<a id="l00119" name="l00119"></a> 119 

<a id="l00120" name="l00120"></a> 120 // Get the number of interacting particle x triangle pairs to launch kernel

<a id="l00121" name="l00121"></a> 121 std::vector&lt;size_t&gt; n_particle_interactions(model.[get\_number\_of\_devices](structIRockyModel.xhtml#a1fd1011aa809fe64647e9cacf1956a11)(), 0);

<a id="l00122" name="l00122"></a> 122 if (sph_model-&gt;gen_sph.use_particles_neighbors_list)

<a id="l00123" name="l00123"></a> 123 {

<a id="l00124" name="l00124"></a> 124 #pragma omp parallel for num\_threads(model.get\_number\_of\_devices()) if (model.\_impl-&gt;gpu\_parallel)

<a id="l00125" name="l00125"></a> 125 for (int i = 0; i &lt; model.[get\_number\_of\_devices](structIRockyModel.xhtml#a1fd1011aa809fe64647e9cacf1956a11)(); ++i)

<a id="l00126" name="l00126"></a> 126 {

<a id="l00127" name="l00127"></a> 127 cuda_general_sph gen_sph_tmp;

<a id="l00128" name="l00128"></a> 128 CUDA_SET_DEVICE(model.[get\_device\_id](structIRockyModel.xhtml#ae8f8865c809a31988291f9180d7cda2b)(i));

<a id="l00129" name="l00129"></a> 129 CUDA_COPY_D2H(&amp;gen_sph_tmp, sph_model-&gt;get_device_general(i), 1);

<a id="l00130" name="l00130"></a> 130 if (it_type == itFreeAndFreeCoupled)

<a id="l00131" name="l00131"></a> 131 n_particle_interactions[i] =

<a id="l00132" name="l00132"></a> 132 gen_sph_tmp.ncont_ff_p_int + gen_sph_tmp.ncont_fc_p_int;

<a id="l00133" name="l00133"></a> 133 else if (it_type == itFreeCoupledOnly)

<a id="l00134" name="l00134"></a> 134 n_particle_interactions[i] = gen_sph_tmp.ncont_fc_p_int;

<a id="l00135" name="l00135"></a> 135 else // itAll

<a id="l00136" name="l00136"></a> 136 n_particle_interactions[i] = gen_sph_tmp.ncont_ff_p_int + gen_sph_tmp.ncont_fc_p_int

<a id="l00137" name="l00137"></a> 137 + gen_sph_tmp.ncont_cc_p_int;

<a id="l00138" name="l00138"></a> 138 }

<a id="l00139" name="l00139"></a> 139 }

<a id="l00140" name="l00140"></a> 140 

<a id="l00141" name="l00141"></a> 141 // Launching user kernel forces

<a id="l00142" name="l00142"></a> 142 for (int i = 0; i &lt; model.[get\_number\_of\_devices](structIRockyModel.xhtml#a1fd1011aa809fe64647e9cacf1956a11)(); ++i)

<a id="l00143" name="l00143"></a> 143 {

<a id="l00144" name="l00144"></a> 144 CUDA_SET_DEVICE(model.[get\_device\_id](structIRockyModel.xhtml#ae8f8865c809a31988291f9180d7cda2b)(i));

<a id="l00145" name="l00145"></a> 145 

<a id="l00146" name="l00146"></a> 146 if (udf_devices_data != nullptr)

<a id="l00147" name="l00147"></a> 147 {

<a id="l00148" name="l00148"></a> 148 udf_device_data = udf_devices_data[i];

<a id="l00149" name="l00149"></a> 149 }

<a id="l00150" name="l00150"></a> 150 

<a id="l00151" name="l00151"></a> 151 const auto [blocks, threads] = CudaBackend::get_launch_params(

<a id="l00152" name="l00152"></a> 152 sph_model-&gt;gen_sph.use_particles_neighbors_list

<a id="l00153" name="l00153"></a> 153 ? n_particle_interactions[i]

<a id="l00154" name="l00154"></a> 154 : model._impl-&gt;sph_model-&gt;gen_sph.nprt_active / model.get_number_of_devices());

<a id="l00155" name="l00155"></a> 155 

<a id="l00156" name="l00156"></a> 156 // Dispatch the SPH elements interaction function using the SPH Iterators.

<a id="l00157" name="l00157"></a> 157 if (cu_hos.sph_model-&gt;gen_sph.use_particles_neighbors_list)

<a id="l00158" name="l00158"></a> 158 {

<a id="l00159" name="l00159"></a> 159 if (cu_hos.gen.periodic_domain.enabled())

<a id="l00160" name="l00160"></a> 160 {

<a id="l00161" name="l00161"></a> 161 rocky::cuda::execute_sph_interactions_function&lt;

<a id="l00162" name="l00162"></a> 162 Operator,

<a id="l00163" name="l00163"></a> 163 decltype(sph_particle_interaction_f),

<a id="l00164" name="l00164"></a> 164 SPHExplicitHandler&lt;CudaBackend, true&gt;&gt;&lt;&lt;&lt;blocks, threads&gt;&gt;&gt;(

<a id="l00165" name="l00165"></a> 165 cu_hos.GetDeviceData(i),

<a id="l00166" name="l00166"></a> 166 cu_hos.sph_model-&gt;get_device_data(i),

<a id="l00167" name="l00167"></a> 167 cu_hos.iteration.current_time,

<a id="l00168" name="l00168"></a> 168 sph_particle_interaction_f,

<a id="l00169" name="l00169"></a> 169 udf_device_data,

<a id="l00170" name="l00170"></a> 170 it_type);

<a id="l00171" name="l00171"></a> 171 }

<a id="l00172" name="l00172"></a> 172 else

<a id="l00173" name="l00173"></a> 173 {

<a id="l00174" name="l00174"></a> 174 rocky::cuda::execute_sph_interactions_function&lt;

<a id="l00175" name="l00175"></a> 175 Operator,

<a id="l00176" name="l00176"></a> 176 decltype(sph_particle_interaction_f),

<a id="l00177" name="l00177"></a> 177 SPHExplicitHandler&lt;CudaBackend&gt;&gt;&lt;&lt;&lt;blocks, threads&gt;&gt;&gt;(

<a id="l00178" name="l00178"></a> 178 cu_hos.GetDeviceData(i),

<a id="l00179" name="l00179"></a> 179 cu_hos.sph_model-&gt;get_device_data(i),

<a id="l00180" name="l00180"></a> 180 cu_hos.iteration.current_time,

<a id="l00181" name="l00181"></a> 181 sph_particle_interaction_f,

<a id="l00182" name="l00182"></a> 182 udf_device_data,

<a id="l00183" name="l00183"></a> 183 it_type);

<a id="l00184" name="l00184"></a> 184 }

<a id="l00185" name="l00185"></a> 185 }

<a id="l00186" name="l00186"></a> 186 else

<a id="l00187" name="l00187"></a> 187 {

<a id="l00188" name="l00188"></a> 188 if (cu_hos.gen.nprt_active &gt; 0)

<a id="l00189" name="l00189"></a> 189 {

<a id="l00190" name="l00190"></a> 190 if (cu_hos.gen.periodic_domain.enabled())

<a id="l00191" name="l00191"></a> 191 {

<a id="l00192" name="l00192"></a> 192 rocky::cuda::execute_sph_interactions_function&lt;

<a id="l00193" name="l00193"></a> 193 Operator,

<a id="l00194" name="l00194"></a> 194 decltype(sph_particle_interaction_f),

<a id="l00195" name="l00195"></a> 195 SPHApplyHandler&lt;CudaBackend, true, true&gt;&gt;&lt;&lt;&lt;blocks, threads&gt;&gt;&gt;(

<a id="l00196" name="l00196"></a> 196 cu_hos.GetDeviceData(i),

<a id="l00197" name="l00197"></a> 197 cu_hos.sph_model-&gt;get_device_data(i),

<a id="l00198" name="l00198"></a> 198 cu_hos.iteration.current_time,

<a id="l00199" name="l00199"></a> 199 sph_particle_interaction_f,

<a id="l00200" name="l00200"></a> 200 udf_device_data,

<a id="l00201" name="l00201"></a> 201 it_type);

<a id="l00202" name="l00202"></a> 202 if (i &lt; model._impl-&gt;get_n_gpus() - 1)

<a id="l00203" name="l00203"></a> 203 {

<a id="l00204" name="l00204"></a> 204 rocky::cuda::execute_sph_interactions_function&lt;

<a id="l00205" name="l00205"></a> 205 Operator,

<a id="l00206" name="l00206"></a> 206 decltype(sph_particle_interaction_f),

<a id="l00207" name="l00207"></a> 207 SPHApplyHandler&lt;CudaBackend, true, true, true&gt;&gt;

<a id="l00208" name="l00208"></a> 208 &lt;&lt;&lt;sph::iterator::BLOCKS, sph::iterator::THREADS&gt;&gt;&gt;(

<a id="l00209" name="l00209"></a> 209 model._impl-&gt;GetDeviceData(i),

<a id="l00210" name="l00210"></a> 210 model._impl-&gt;sph_model-&gt;get_device_data(i),

<a id="l00211" name="l00211"></a> 211 cu_hos.iteration.current_time,

<a id="l00212" name="l00212"></a> 212 sph_particle_interaction_f,

<a id="l00213" name="l00213"></a> 213 udf_device_data,

<a id="l00214" name="l00214"></a> 214 it_type);

<a id="l00215" name="l00215"></a> 215 }

<a id="l00216" name="l00216"></a> 216 }

<a id="l00217" name="l00217"></a> 217 else

<a id="l00218" name="l00218"></a> 218 {

<a id="l00219" name="l00219"></a> 219 rocky::cuda::execute_sph_interactions_function&lt;

<a id="l00220" name="l00220"></a> 220 Operator,

<a id="l00221" name="l00221"></a> 221 decltype(sph_particle_interaction_f),

<a id="l00222" name="l00222"></a> 222 SPHApplyHandler&lt;CudaBackend, true&gt;&gt;&lt;&lt;&lt;blocks, threads&gt;&gt;&gt;(

<a id="l00223" name="l00223"></a> 223 cu_hos.GetDeviceData(i),

<a id="l00224" name="l00224"></a> 224 cu_hos.sph_model-&gt;get_device_data(i),

<a id="l00225" name="l00225"></a> 225 cu_hos.iteration.current_time,

<a id="l00226" name="l00226"></a> 226 sph_particle_interaction_f,

<a id="l00227" name="l00227"></a> 227 udf_device_data,

<a id="l00228" name="l00228"></a> 228 it_type);

<a id="l00229" name="l00229"></a> 229 if (i &lt; model._impl-&gt;get_n_gpus() - 1)

<a id="l00230" name="l00230"></a> 230 {

<a id="l00231" name="l00231"></a> 231 rocky::cuda::execute_sph_interactions_function&lt;

<a id="l00232" name="l00232"></a> 232 Operator,

<a id="l00233" name="l00233"></a> 233 decltype(sph_particle_interaction_f),

<a id="l00234" name="l00234"></a> 234 SPHApplyHandler&lt;CudaBackend, true, false, true&gt;&gt;

<a id="l00235" name="l00235"></a> 235 &lt;&lt;&lt;sph::iterator::BLOCKS, sph::iterator::THREADS&gt;&gt;&gt;(

<a id="l00236" name="l00236"></a> 236 model._impl-&gt;GetDeviceData(i),

<a id="l00237" name="l00237"></a> 237 model._impl-&gt;sph_model-&gt;get_device_data(i),

<a id="l00238" name="l00238"></a> 238 cu_hos.iteration.current_time,

<a id="l00239" name="l00239"></a> 239 sph_particle_interaction_f,

<a id="l00240" name="l00240"></a> 240 udf_device_data,

<a id="l00241" name="l00241"></a> 241 it_type);

<a id="l00242" name="l00242"></a> 242 }

<a id="l00243" name="l00243"></a> 243 }

<a id="l00244" name="l00244"></a> 244 }

<a id="l00245" name="l00245"></a> 245 else

<a id="l00246" name="l00246"></a> 246 {

<a id="l00247" name="l00247"></a> 247 if (cu_hos.gen.periodic_domain.enabled())

<a id="l00248" name="l00248"></a> 248 {

<a id="l00249" name="l00249"></a> 249 rocky::cuda::execute_sph_interactions_function&lt;

<a id="l00250" name="l00250"></a> 250 Operator,

<a id="l00251" name="l00251"></a> 251 decltype(sph_particle_interaction_f),

<a id="l00252" name="l00252"></a> 252 SPHApplyHandler&lt;CudaBackend, false, true&gt;&gt;&lt;&lt;&lt;blocks, threads&gt;&gt;&gt;(

<a id="l00253" name="l00253"></a> 253 cu_hos.GetDeviceData(i),

<a id="l00254" name="l00254"></a> 254 cu_hos.sph_model-&gt;get_device_data(i),

<a id="l00255" name="l00255"></a> 255 cu_hos.iteration.current_time,

<a id="l00256" name="l00256"></a> 256 sph_particle_interaction_f,

<a id="l00257" name="l00257"></a> 257 udf_device_data,

<a id="l00258" name="l00258"></a> 258 it_type);

<a id="l00259" name="l00259"></a> 259 if (i &lt; model._impl-&gt;get_n_gpus() - 1)

<a id="l00260" name="l00260"></a> 260 {

<a id="l00261" name="l00261"></a> 261 rocky::cuda::execute_sph_interactions_function&lt;

<a id="l00262" name="l00262"></a> 262 Operator,

<a id="l00263" name="l00263"></a> 263 decltype(sph_particle_interaction_f),

<a id="l00264" name="l00264"></a> 264 SPHApplyHandler&lt;CudaBackend, false, true, true&gt;&gt;

<a id="l00265" name="l00265"></a> 265 &lt;&lt;&lt;sph::iterator::BLOCKS, sph::iterator::THREADS&gt;&gt;&gt;(

<a id="l00266" name="l00266"></a> 266 model._impl-&gt;GetDeviceData(i),

<a id="l00267" name="l00267"></a> 267 model._impl-&gt;sph_model-&gt;get_device_data(i),

<a id="l00268" name="l00268"></a> 268 cu_hos.iteration.current_time,

<a id="l00269" name="l00269"></a> 269 sph_particle_interaction_f,

<a id="l00270" name="l00270"></a> 270 udf_device_data,

<a id="l00271" name="l00271"></a> 271 it_type);

<a id="l00272" name="l00272"></a> 272 }

<a id="l00273" name="l00273"></a> 273 }

<a id="l00274" name="l00274"></a> 274 else

<a id="l00275" name="l00275"></a> 275 {

<a id="l00276" name="l00276"></a> 276 rocky::cuda::execute_sph_interactions_function&lt;

<a id="l00277" name="l00277"></a> 277 Operator,

<a id="l00278" name="l00278"></a> 278 decltype(sph_particle_interaction_f),

<a id="l00279" name="l00279"></a> 279 SPHApplyHandler&lt;CudaBackend, false, false&gt;&gt;&lt;&lt;&lt;blocks, threads&gt;&gt;&gt;(

<a id="l00280" name="l00280"></a> 280 cu_hos.GetDeviceData(i),

<a id="l00281" name="l00281"></a> 281 cu_hos.sph_model-&gt;get_device_data(i),

<a id="l00282" name="l00282"></a> 282 cu_hos.iteration.current_time,

<a id="l00283" name="l00283"></a> 283 sph_particle_interaction_f,

<a id="l00284" name="l00284"></a> 284 udf_device_data,

<a id="l00285" name="l00285"></a> 285 it_type);

<a id="l00286" name="l00286"></a> 286 if (i &lt; model._impl-&gt;get_n_gpus() - 1)

<a id="l00287" name="l00287"></a> 287 {

<a id="l00288" name="l00288"></a> 288 rocky::cuda::execute_sph_interactions_function&lt;

<a id="l00289" name="l00289"></a> 289 Operator,

<a id="l00290" name="l00290"></a> 290 decltype(sph_particle_interaction_f),

<a id="l00291" name="l00291"></a> 291 SPHApplyHandler&lt;CudaBackend, false, false, true&gt;&gt;

<a id="l00292" name="l00292"></a> 292 &lt;&lt;&lt;sph::iterator::BLOCKS, sph::iterator::THREADS&gt;&gt;&gt;(

<a id="l00293" name="l00293"></a> 293 model._impl-&gt;GetDeviceData(i),

<a id="l00294" name="l00294"></a> 294 model._impl-&gt;sph_model-&gt;get_device_data(i),

<a id="l00295" name="l00295"></a> 295 cu_hos.iteration.current_time,

<a id="l00296" name="l00296"></a> 296 sph_particle_interaction_f,

<a id="l00297" name="l00297"></a> 297 udf_device_data,

<a id="l00298" name="l00298"></a> 298 it_type);

<a id="l00299" name="l00299"></a> 299 }

<a id="l00300" name="l00300"></a> 300 }

<a id="l00301" name="l00301"></a> 301 }

<a id="l00302" name="l00302"></a> 302 }

<a id="l00303" name="l00303"></a> 303 }

<a id="l00304" name="l00304"></a> 304 

<a id="l00305" name="l00305"></a> 305 // Syncing all GPUs

<a id="l00306" name="l00306"></a> 306 sync_devices(model);

<a id="l00307" name="l00307"></a> 307}

<a id="l00308" name="l00308"></a> 308 

<a id="l00309" name="l00309"></a> 309template &lt;typename Operator = device::DefaultTrianglesOperator&lt;CudaBackend&gt;, typename Functor&gt;

<a id="l00310" name="l00310"></a> 310void dispatch_sph_interactions_triangles_kernel(

<a id="l00311" name="l00311"></a> 311 [IRockyModel](structIRockyModel.xhtml)&amp; model, Functor&amp; sph_triangle_interaction_f, void** udf_devices_data)

<a id="l00312" name="l00312"></a> 312{

<a id="l00313" name="l00313"></a> 313 SPHModel* sph_model = &amp;(*model._impl-&gt;sph_model);

<a id="l00314" name="l00314"></a> 314 cuda_host&amp; cu_hos = *model._impl;

<a id="l00315" name="l00315"></a> 315 

<a id="l00316" name="l00316"></a> 316 if (sph_model-&gt;gen_sph.nprt_active &lt;= 0)

<a id="l00317" name="l00317"></a> 317 return;

<a id="l00318" name="l00318"></a> 318 

<a id="l00319" name="l00319"></a> 319 void* udf_device_data = nullptr;

<a id="l00320" name="l00320"></a> 320 

<a id="l00321" name="l00321"></a> 321 // Get the number of interacting particle x triangle pairs to launch kernel

<a id="l00322" name="l00322"></a> 322 std::vector&lt;int&gt; n_triangle_force_pairs(model.[get\_number\_of\_devices](structIRockyModel.xhtml#a1fd1011aa809fe64647e9cacf1956a11)(), 0);

<a id="l00323" name="l00323"></a> 323 #pragma omp parallel for num\_threads(model.get\_number\_of\_devices()) if (model.\_impl-&gt;gpu\_parallel)

<a id="l00324" name="l00324"></a> 324 for (int i = 0; i &lt; model.[get\_number\_of\_devices](structIRockyModel.xhtml#a1fd1011aa809fe64647e9cacf1956a11)(); ++i)

<a id="l00325" name="l00325"></a> 325 {

<a id="l00326" name="l00326"></a> 326 cuda_general_sph gen_sph_tmp;

<a id="l00327" name="l00327"></a> 327 CUDA_SET_DEVICE(model.[get\_device\_id](structIRockyModel.xhtml#ae8f8865c809a31988291f9180d7cda2b)(i));

<a id="l00328" name="l00328"></a> 328 CUDA_COPY_D2H(&amp;gen_sph_tmp, sph_model-&gt;get_device_general(i), 1);

<a id="l00329" name="l00329"></a> 329 n_triangle_force_pairs[i] = gen_sph_tmp.n_triangle_force_pairs;

<a id="l00330" name="l00330"></a> 330 }

<a id="l00331" name="l00331"></a> 331 

<a id="l00332" name="l00332"></a> 332 // Launching user kernel forces

<a id="l00333" name="l00333"></a> 333 for (int i = 0; i &lt; model.[get\_number\_of\_devices](structIRockyModel.xhtml#a1fd1011aa809fe64647e9cacf1956a11)(); ++i)

<a id="l00334" name="l00334"></a> 334 {

<a id="l00335" name="l00335"></a> 335 CUDA_SET_DEVICE(model.[get\_device\_id](structIRockyModel.xhtml#ae8f8865c809a31988291f9180d7cda2b)(i));

<a id="l00336" name="l00336"></a> 336 

<a id="l00337" name="l00337"></a> 337 SPHDeviceModel* sph_device = model._impl-&gt;sph_model-&gt;get_device_data(i);

<a id="l00338" name="l00338"></a> 338 SDeviceModel* device_data = model.get_device_data(i);

<a id="l00339" name="l00339"></a> 339 if (udf_devices_data != nullptr)

<a id="l00340" name="l00340"></a> 340 {

<a id="l00341" name="l00341"></a> 341 udf_device_data = udf_devices_data[i];

<a id="l00342" name="l00342"></a> 342 }

<a id="l00343" name="l00343"></a> 343 

<a id="l00344" name="l00344"></a> 344 auto [blocks, threads] = CudaBackend::get_launch_params(n_triangle_force_pairs[i]);

<a id="l00345" name="l00345"></a> 345 rocky::cuda::execute_sph_interactions_triangle_function&lt;Operator&gt;

<a id="l00346" name="l00346"></a> 346 &lt;&lt;&lt;blocks, threads&gt;&gt;&gt;(

<a id="l00347" name="l00347"></a> 347 device_data,

<a id="l00348" name="l00348"></a> 348 sph_device,

<a id="l00349" name="l00349"></a> 349 cu_hos.iteration.current_time,

<a id="l00350" name="l00350"></a> 350 sph_triangle_interaction_f,

<a id="l00351" name="l00351"></a> 351 udf_device_data);

<a id="l00352" name="l00352"></a> 352 }

<a id="l00353" name="l00353"></a> 353 

<a id="l00354" name="l00354"></a> 354 // Syncing all GPUs

<a id="l00355" name="l00355"></a> 355 sync_devices(model);

<a id="l00356" name="l00356"></a> 356}

<a id="l00357" name="l00357"></a> 357 

<a id="l00358" name="l00358"></a> 358} // namespace rocky::cuda

[IRockyModel](structIRockyModel.xhtml)

**Definition** rocky_api.h:58



[IRockyModel::get\_number\_of\_devices](structIRockyModel.xhtml#a1fd1011aa809fe64647e9cacf1956a11)

int get_number_of_devices() const

**Definition** rocky_api.cpp:173



[IRockyModel::get\_device\_id](structIRockyModel.xhtml#ae8f8865c809a31988291f9180d7cda2b)

int get_device_id(int device_index) const

**Definition** rocky_api.cpp:178