From 75f72b7f6ea87ffafce315e456d5d9cecd4e07bf Mon Sep 17 00:00:00 2001
From: Nallani Bhaskar <Nallani.Bhaskar@amd.com>
Date: Thu, 17 Jun 2021 23:53:12 +0530
Subject: [PATCH] Added aocl dynamic feature for dtrsm for small sizes

Details:
1. Added aocl-dynamic for dtrsm native path
   When (m,n)<512 better performance observed for nthreads=4
2. Updated trsm_small threshold such that when (m+n)<320
   trsm_small is doing better than native irrespective of
   number of threads

Change-Id: Ic2c50f14db257a05e323cc97c5d1c9b73b68f487
---
 frame/3/trsm/bli_trsm_front.c |  5 +++++
 frame/base/bli_rntm.c         | 10 ++++++++++
 frame/compat/bla_trsm.c       |  2 +-
 3 files changed, 16 insertions(+), 1 deletion(-)

diff --git a/frame/3/trsm/bli_trsm_front.c b/frame/3/trsm/bli_trsm_front.c
index 0c3bb11d2..f964faf0d 100644
--- a/frame/3/trsm/bli_trsm_front.c
+++ b/frame/3/trsm/bli_trsm_front.c
@@ -126,6 +126,11 @@ void bli_trsm_front
 	bli_obj_set_as_root( &b_local );
 	bli_obj_set_as_root( &c_local );
 
+#ifdef AOCL_DYNAMIC
+	// If dynamic-threading is enabled, calculate optimum number
+	//  of threads and update in rntm
+	bli_nthreads_optimum(a, b, b, BLIS_TRSM, rntm );
+#endif
 	// Parse and interpret the contents of the rntm_t object to properly
 	// set the ways of parallelism for each loop, and then make any
 	// additional modifications necessary for the current operation.
diff --git a/frame/base/bli_rntm.c b/frame/base/bli_rntm.c
index 09cd45ad1..419c65dcb 100644
--- a/frame/base/bli_rntm.c
+++ b/frame/base/bli_rntm.c
@@ -556,6 +556,16 @@ void bli_nthreads_optimum(
 		else
 			n_threads_ideal = n_threads;
 	}
+        else if( family == BLIS_TRSM && bli_obj_is_double(c))
+        {
+                dim_t m = bli_obj_length(c);
+                dim_t n = bli_obj_width(c);
+
+                if(m<=512 && n<=512)
+                    n_threads_ideal = 4;
+
+        }
+
 	dim_t n_threads_opt = bli_min(n_threads, n_threads_ideal);
 
 	bli_pthread_mutex_lock( &global_rntm_mutex );
diff --git a/frame/compat/bla_trsm.c b/frame/compat/bla_trsm.c
index 5a839485b..95ae079cc 100644
--- a/frame/compat/bla_trsm.c
+++ b/frame/compat/bla_trsm.c
@@ -602,7 +602,7 @@ void dtrsm_
      * is doing better than native multithread */
     bool nt = bli_thread_get_is_parallel();
     if((nt==0 && m0<=1000 && n0<=1000) ||
-       (nt && m0<=128  && n0<=128 ) )
+       (nt && (m0+n0)<320) )
     {
         err_t status;
         status = bli_trsm_small