Fixes to HPC runtime code path. (#773)

srinivasyadav18 · web-flow · commit a4a63295b96e · 2023-09-26T17:58:38.000-05:00
Details:
- Fixed hpx::for_each invocation and replace with hpx::for_loop. The HPX 
  runtime was initialized using hpx::start, but the hpx::for_each 
  function was being called on a non-hpx runtime (i.e standard BLIS 
  runtime - single main thread). To run hpx::for_each on HPX runtime 
  correctly, the code now uses hpx::run_as_hpx_thread(func, args...). 
- Replaced hpx::for_each with hpx::for_loop, which eliminates use of 
  hpx::util::counting_iterator.
- Employ hpx::execution::chunk_size(1) to make sure that a thread 
  resides on a particular core.
- Replaced hpx::apply() with updated version hpx::post().
- Initialize tdata-&gt;id = 0 in libblis.c to 0, as it is the main thread 
  and is needed for writing results to output file.
- By default, if not specified, the HPX runtime uses all N threads/cores 
  available in the system. But, if we want to only specify n_threads out 
  N threads, we use hpx::execution::experimental::num_cores(n_threads).
diff --git a/CREDITS b/CREDITS
@@ -130,6 +130,7 @@ but many others have contributed code, ideas, and feedback, including
   Benda Xu                 @heroxbd
   Guodong Xu               @docularxu                 (Linaro.org)
   RuQing Xu                @xrq-phys                  (The University of Tokyo)
+  Srinivas Yadav           @srinivasyadav18
   Costas Yamin             @cosstas
   Chenhan Yu               @ChenhanYu                 (The University of Texas at Austin)
   Roman Yurchak            @rth                       (Symerio)
diff --git a/frame/thread/bli_thread_hpx.cpp b/frame/thread/bli_thread_hpx.cpp
@@ -36,9 +36,10 @@
 
 #ifdef BLIS_ENABLE_HPX
 
-#include <hpx/local/execution.hpp>
-#include <hpx/parallel/algorithms/for_each.hpp>
+#include <hpx/execution.hpp>
 #include <hpx/hpx_start.hpp>
+#include <hpx/parallel/algorithms/for_loop.hpp>
+#include <hpx/runtime_local/run_as_hpx_thread.hpp>
 
 extern "C"
 {
@@ -55,13 +56,16 @@ void bli_thread_launch_hpx
 	// Allocate a global communicator for the root thrinfo_t structures.
 	pool_t*    gl_comm_pool = nullptr;
 	thrcomm_t* gl_comm      = bli_thrcomm_create( ti, gl_comm_pool, n_threads );
-
-	auto irange = hpx::util::counting_shape(n_threads);
-
-	hpx::for_each(hpx::execution::par, hpx::util::begin(irange), hpx::util::end(irange),
-	[&gl_comm, &func, &params](const dim_t tid)
+	hpx::threads::run_as_hpx_thread([&]()
 	{
-		func( gl_comm, tid, params );
+		hpx::execution::experimental::num_cores num_cores_(n_threads);
+		hpx::execution::static_chunk_size chunk_size_(1);
+		hpx::experimental::for_loop(
+		hpx::execution::par.with(num_cores_).with(chunk_size_), 0, n_threads,
+		[&gl_comm, &func, &params](const dim_t tid)
+		{
+			func( gl_comm, tid, params );
+		});
 	});
 
 	// Free the global communicator, because the root thrinfo_t node
@@ -76,7 +80,7 @@ void bli_thread_initialize_hpx( int argc, char** argv )
 
 int bli_thread_finalize_hpx()
 {
-	hpx::apply([]() { hpx::finalize(); });
+	hpx::post([]() { hpx::finalize(); });
 	return hpx::stop();
 }
 
diff --git a/testsuite/src/test_libblis.c b/testsuite/src/test_libblis.c
@@ -135,15 +135,16 @@ void libblis_test_thread_decorator( test_params_t* params, test_ops_t* ops )
 	err_t r_val;
 
 #ifdef BLIS_ENABLE_HPX
+	size_t nt = ( size_t )params->n_app_threads;
 
-	size_t tdata_size = ( size_t )params->n_app_threads *
+	size_t tdata_size = ( size_t )nt *
 	                    ( size_t )sizeof( thread_data_t );
 	thread_data_t* tdata = bli_malloc_user( tdata_size, &r_val );
 
 	tdata->params  = params;
 	tdata->ops     = ops;
 	tdata->nt      = nt;
-	tdata->id      = 1;
+	tdata->id      = 0;
 	tdata->xc      = 0;
 
 	// Walk through all test modules.
