From 5d57d67cb3f49fe9c6ff673044ed6caf4c55ec3a Mon Sep 17 00:00:00 2001
From: dzambare <DipalMadhukar.Zambare@amd.com>
Date: Wed, 3 Jun 2020 09:58:55 +0530
Subject: [PATCH] Checking for zero dimension is moved to bli_gemm_xx call.

This will ensure early return in case full gemm processing is not needed.

Based on dimension which is found to be zero following actions will be taken:

If 'c' has zero dimension, no further processing is requried
If alpha is zero or if 'a' or 'b' has zero diemension, we
perform scalm operation instead of gemm. (c = alpha*a + beta*b)

Change-Id: Icc031944fc4e80138adf991974547f2d57ab570b
AMD-Internal: [CPUPL-904]
---
 frame/3/bli_l3_oapi.c         | 13 +++++++++++++
 frame/3/gemm/bli_gemm_front.c | 10 +---------
 2 files changed, 14 insertions(+), 9 deletions(-)

diff --git a/frame/3/bli_l3_oapi.c b/frame/3/bli_l3_oapi.c
index 29eca1bf1..d09053fe7 100644
--- a/frame/3/bli_l3_oapi.c
+++ b/frame/3/bli_l3_oapi.c
@@ -57,6 +57,19 @@ void PASTEMAC(opname,EX_SUF) \
 	bli_init_once(); \
 \
 	BLIS_OAPI_EX_DECLS \
+\
+	/* If C has a zero dimension, return early.	*/	\
+	if ( bli_obj_has_zero_dim( c ) ) return; \
+\
+	/* if alpha or A or B has a zero dimension, \
+	   scale C by beta and return early. */ \
+	if ( bli_obj_equals( alpha, &BLIS_ZERO ) || \
+		 bli_obj_has_zero_dim( a ) || \
+		 bli_obj_has_zero_dim( b ) ) \
+	{\
+		bli_scalm( beta, c ); \
+		return;\
+	}\
 \
 	/* If the rntm is non-NULL, it may indicate that we should forgo sup
 	   handling altogether. */ \
diff --git a/frame/3/gemm/bli_gemm_front.c b/frame/3/gemm/bli_gemm_front.c
index f274d51e5..6ac1755f4 100644
--- a/frame/3/gemm/bli_gemm_front.c
+++ b/frame/3/gemm/bli_gemm_front.c
@@ -56,15 +56,7 @@ void bli_gemm_front
 	// Check parameters.
 	if ( bli_error_checking_is_enabled() )
 		bli_gemm_check( alpha, a, b, beta, c, cntx );
-
-	// If alpha is zero, scale by beta and return.
-	if ( bli_obj_equals( alpha, &BLIS_ZERO ) )
-	{
-		bli_scalm( beta, c );
-		return;
-	}
-
-
+	
 #ifdef BLIS_ENABLE_SMALL_MATRIX
 	// Only handle small problems separately for homogeneous datatypes.
 	if ( bli_obj_dt( a ) == bli_obj_dt( b ) &&