From 5cf58f7c2d5bc0d2d94d9576f7158d8f133b7aac Mon Sep 17 00:00:00 2001
From: "Field G. Van Zee" <field@cs.utexas.edu>
Date: Tue, 4 Feb 2014 09:15:19 -0600
Subject: [PATCH] Added early returns for "object is zeros" case.

Details:
- Added some logic to packm_init(), pack_int() and gemm_int() so that
  (a) objects marked as BLIS_ZEROS are not packed, and (b) those
  objects are not computed with. This functionality is not currently
  needed by any existing implementations, but may be used in the
  future.
---
 frame/1m/packm/bli_packm_init.c |  8 ++++++++
 frame/1m/packm/bli_packm_int.c  | 30 ++++--------------------------
 frame/3/gemm/bli_gemm_int.c     |  9 +++++++++
 3 files changed, 21 insertions(+), 26 deletions(-)

diff --git a/frame/1m/packm/bli_packm_init.c b/frame/1m/packm/bli_packm_init.c
index 2e17632d9..8441fe6fd 100644
--- a/frame/1m/packm/bli_packm_init.c
+++ b/frame/1m/packm/bli_packm_init.c
@@ -95,6 +95,14 @@ void bli_packm_init( obj_t*   a,
 		return;
 	}
 
+	// If the object is marked as being filled with zeros, then we can skip
+	// the packm operation entirely and alias.
+	if ( bli_obj_is_zeros( *a ) )
+	{
+		bli_obj_alias_to( *a, *p );
+		return;
+	}
+
 	// Now, if we are not skipping the pack operation, then the only question
 	// left is whether we are to typecast matrix a before packing.
 	if ( bli_obj_datatype( *a ) != bli_obj_target_datatype( *a ) )
diff --git a/frame/1m/packm/bli_packm_int.c b/frame/1m/packm/bli_packm_int.c
index bd6b8ef73..f048b5fa0 100644
--- a/frame/1m/packm/bli_packm_int.c
+++ b/frame/1m/packm/bli_packm_int.c
@@ -102,34 +102,12 @@ void bli_packm_int( obj_t*   a,
 		return;
 	}
 
-/*
-	// The value for kappa we use will depend on whether the scalar
-	// attached to A has a nonzero imaginary component. If it does,
-	// then we will apply the scalar during packing to facilitate
-	// implementing complex domain micro-kernels in terms of their
-	// real domain counterparts. (In the aforementioned situation,
-	// applying a real scalar is easy, but applying a complex one is
-	// harder, so we avoid the need altogether with the code below.)
-	if ( bli_obj_scalar_has_nonzero_imag( a ) )
+	// If the object is marked as being filled with zeros, then we can skip
+	// the packm operation entirely.
+	if ( bli_obj_is_zeros( *a ) )
 	{
-		bli_check_error_code( BLIS_NOT_YET_IMPLEMENTED );
-
-		// Detach the scalar.
-		bli_obj_scalar_detach( a, &kappa );
-
-		// Reset the attached scalar (to 1.0).
-		bli_obj_scalar_reset( a );
-
-		kappa_p = &kappa;
+		return;
 	}
-	else
-	{
-		// If the internal scalar of A has only a real component, then
-		// we will apply it later (in the micro-kernel), and so we will
-		// use BLIS_ONE to indicate no scaling during packing.
-		kappa_p = &BLIS_ONE;
-	}
-*/
 
 
 	// Extract the variant number and implementation type.
diff --git a/frame/3/gemm/bli_gemm_int.c b/frame/3/gemm/bli_gemm_int.c
index bc0a35472..f6fc6d284 100644
--- a/frame/3/gemm/bli_gemm_int.c
+++ b/frame/3/gemm/bli_gemm_int.c
@@ -81,6 +81,15 @@ void bli_gemm_int( obj_t*  alpha,
 		return;
 	}
 
+	// If A or B is marked as being filled with zeros, scale C by beta and
+	// return early.
+	if ( bli_obj_is_zeros( *a ) ||
+	     bli_obj_is_zeros( *b ) )
+	{
+		bli_scalm( beta, c );
+		return;
+	}
+
 	// Alias A and B in case we need to update attached scalars.
 	bli_obj_alias_to( *a, a_local );
 	bli_obj_alias_to( *b, b_local );