From bdbdd209967ab721a9ecd86736ac0cdb1e5ea8e7 Mon Sep 17 00:00:00 2001
From: Mangala V <mangala.v@amd.com>
Date: Fri, 7 Oct 2022 07:29:09 -0400
Subject: [PATCH] Enabled MT path in DTRSM Small

1. Fixed accuracy issues in dtrsm small multi-thread.
2. Fixed out of bound memory accesses in the patch
   "Fixes to avoid Out of Bound Memory Access in TRSM small algorithm"
3. Re-enabled DTRSM small MT with above fixes.

AMD-Internal: [CPUPL-2567]
Change-Id: Ibf2949b25fde4007a92cc635526bed0e0d897800
---
 frame/compat/bla_trsm_amd.c | 32 ++++++++++++++++++++++++++++++++
 1 file changed, 32 insertions(+)

diff --git a/frame/compat/bla_trsm_amd.c b/frame/compat/bla_trsm_amd.c
index 1fc8899ce..9e4d35dfd 100644
--- a/frame/compat/bla_trsm_amd.c
+++ b/frame/compat/bla_trsm_amd.c
@@ -967,9 +967,41 @@ void dtrsm_blis_impl
                 return;
             }
         }
+        // bli_trsm_small_mt is performing better than native multithread
+        // for certain sizes of m & n.
+#ifdef BLIS_ENABLE_OPENMP
+        rntm_t rntm;
+        bli_rntm_init_from_global( &rntm );
+        // Query the total number of threads from the rntm_t object.
+        dim_t n_threads = bli_rntm_num_threads( &rntm );
+        if ( ( (n_threads  >  1) && (m0 <= 1500) && (n0 <= 1500) ) ||
+            ( (n_threads == 32) && (m0 <= 2300) && (n0 <= 2300) ) ||
+            ( (n_threads == 16) && (m0 <= 3800) && (n0 <= 3800) ) ||
+            ( (n_threads ==  8) && (m0 <= 2800) && (n0 <= 2800) ) ||
+            ( (n_threads ==  4) && (m0 <= 2000) && (n0 <= 2000) ) ||
+            ( (n_threads ==  2) && (m0 <= 2000) && (n0 <= 2000) ) )
+        {
+            err_t status;
+            status = bli_trsm_small_mt(
+                    blis_side,
+                    &alphao,
+                    &ao,
+                    &bo,
+                    NULL,
+                    NULL);
+            if ( status == BLIS_SUCCESS )
+            {
+                AOCL_DTL_TRACE_EXIT(AOCL_DTL_LEVEL_INFO);
+                /* Finalize BLIS. */
+                bli_finalize_auto();
+                return;
+            }
+        }
+#endif// BLIS_ENABLE_OPENMP
     } // bli_cpuid_is_avx_supported
 #endif// END of BLIS_ENABLE_SMALL_MATRIX_TRSM
 
+
     bli_trsmnat
     (
         blis_side,