progress bar finally works on NVIDIA, thanks to Robert Crovella@NVIDIA …

…http://bit.ly/2XY0IVb
fangq · Jul 22, 2019 · 2894ea1 · 2894ea1
1 parent 2863ae3
commit 2894ea1
Show file tree

Hide file tree

Showing 2 changed files with 15 additions and 10 deletions.
diff --git a/src/mcx_host.cpp b/src/mcx_host.cpp
@@ -467,10 +467,15 @@ void mcx_run_simulation(Config *cfg,float *fluence,float *totalenergy){
  OCL_ASSERT(((gmedia=clCreateBuffer(mcxcontext,RO_MEM, sizeof(cl_uint)*(dimxyz),media,&status),status)));
  OCL_ASSERT(((gproperty=clCreateBuffer(mcxcontext,RO_MEM, cfg->medianum*sizeof(Medium),cfg->prop,&status),status)));
  OCL_ASSERT(((gparam=clCreateBuffer(mcxcontext,RO_MEM, sizeof(MCXParam),&param,&status),status)));
- OCL_ASSERT(((gprogress[0]=clCreateBuffer(mcxcontext,RW_PTR, sizeof(cl_uint),NULL,&status),status)));
- progress = (cl_uint *)clEnqueueMapBuffer(mcxqueue[0], gprogress[0], CL_TRUE, CL_MAP_WRITE, 0, sizeof(cl_uint), 0, NULL, NULL, NULL);
+
+ cl_mem (*clCreateBufferNV)(cl_context,cl_mem_flags, cl_mem_flags_NV, size_t, void*, cl_int*) = (cl_mem (*)(cl_context,cl_mem_flags, cl_mem_flags_NV, size_t, void*, cl_int*)) clGetExtensionFunctionAddressForPlatform(platform, "clCreateBufferNV");
+ if (clCreateBufferNV == NULL)
+ OCL_ASSERT(((gprogress[0]=clCreateBuffer(mcxcontext,RW_PTR, sizeof(cl_uint),NULL,&status),status)));
+ else
+ OCL_ASSERT(((gprogress[0]=clCreateBufferNV(mcxcontext,CL_MEM_READ_WRITE, NV_PIN, sizeof(cl_uint),NULL,&status),status)));
+
+ progress = (cl_uint *)clEnqueueMapBuffer(mcxqueue[0], gprogress[0], CL_TRUE, CL_MAP_READ | CL_MAP_WRITE, 0, sizeof(cl_uint), 0, NULL, NULL, NULL);
  *progress=0;
- clEnqueueUnmapMemObject(mcxqueue[0], gprogress[0], progress, 0, NULL, NULL);
 
  if(cfg->seed==SEED_FROM_FILE){
  // replay should only work with a single device
@@ -673,13 +678,8 @@ void mcx_run_simulation(Config *cfg,float *fluence,float *totalenergy){
 
  mcx_progressbar(-0.f,cfg);
 
- progress = (cl_uint *)clEnqueueMapBuffer(mcxqueue[0], gprogress[0], CL_FALSE, CL_MAP_READ, 0, sizeof(cl_uint), 0, NULL, NULL, NULL);
-
  do{
  ndone = *progress;
-
- MCX_FPRINTF(cfg->flog,"progress=%d\n",ndone); // debug progress bar, will remove
-
  if (ndone > p0){
  mcx_progressbar(ndone/(threadphoton*1.45f),cfg);
  p0 = ndone;
@@ -688,9 +688,8 @@ void mcx_run_simulation(Config *cfg,float *fluence,float *totalenergy){
  }while (p0 < (param.threadphoton*1.45f));
  mcx_progressbar(1.0f,cfg);
  MCX_FPRINTF(cfg->flog,"\n");
-
- clEnqueueUnmapMemObject(mcxqueue[0], gprogress[0], progress, 0, NULL, NULL);
  }
+ clEnqueueUnmapMemObject(mcxqueue[0], gprogress[0], progress, 0, NULL, NULL);
 
  //clWaitForEvents(workdev,waittoread);
  for(devid=0;devid<workdev;devid++)

diff --git a/src/mcx_host.hpp b/src/mcx_host.hpp
@@ -35,10 +35,16 @@ extern "C" {
  #define RAND_BUF_LEN 2 //register arrays
 #endif
 
+#ifndef CL_MEM_LOCATION_HOST_NV
+ #define CL_MEM_LOCATION_HOST_NV (1 << 0)
+ typedef cl_bitfield cl_mem_flags_NV;
+#endif
+
 #define RO_MEM (CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR)
 #define WO_MEM (CL_MEM_WRITE_ONLY | CL_MEM_COPY_HOST_PTR)
 #define RW_MEM (CL_MEM_READ_WRITE | CL_MEM_COPY_HOST_PTR)
 #define RW_PTR (CL_MEM_READ_ONLY | CL_MEM_ALLOC_HOST_PTR)
+#define NV_PIN CL_MEM_LOCATION_HOST_NV
 
 #define OCL_ASSERT(x) ocl_assess((x),__FILE__,__LINE__)