diff --git a/test/sup/Makefile b/test/sup/Makefile
index d2b3c7170..7934d7052 100644
--- a/test/sup/Makefile
+++ b/test/sup/Makefile
@@ -103,6 +103,9 @@ OPENBLASP_LIB  := $(HOME_LIB_PATH)/libopenblasp.a
 # BLASFEO
 BLASFEO_LIB    := $(HOME_LIB_PATH)/libblasfeo.a
 
+# libxsmm
+LIBXSMM_LIB    := $(HOME_LIB_PATH)/libxsmm.a
+
 # ATLAS
 ATLAS_LIB      := $(HOME_LIB_PATH)/libf77blas.a \
                   $(HOME_LIB_PATH)/libatlas.a
@@ -306,14 +309,18 @@ get-imp-defs = $(strip $(subst  blissup,-DSTR=\"$(1)\" -DBLIS -DSUP, \
                        $(subst    eigen,-DSTR=\"$(1)\" -DEIGEN, \
                        $(subst openblas,-DSTR=\"$(1)\" -DCBLAS, \
                        $(subst  blasfeo,-DSTR=\"$(1)\" -DCBLAS, \
-                       $(subst   vendor,-DSTR=\"$(1)\" -DCBLAS,$(1))))))))
+                       $(subst  libxsmm,-DSTR=\"$(1)\" -DBLAS, \
+                       $(subst   vendor,-DSTR=\"$(1)\" -DCBLAS,$(1)))))))))
 
 TRANS0  = $(call stripu,$(TRANS))
 STORS0  = $(call stripu,$(STORS))
 
 # Limit BLAS and Eigen to only using all row-stored, or all column-stored matrices.
+# Also, limit libxsmm to using all column-stored matrices since it does not offer
+# CBLAS interfaces.
 BSTORS0 = rrr ccc
 ESTORS0 = rrr ccc
+XSTORS0 = ccc
 
 
 #
@@ -339,6 +346,9 @@ OPENBLAS_ST_BINS := $(patsubst %.o,%.x,$(OPENBLAS_ST_OBJS))
 BLASFEO_ST_OBJS  := $(call get-st-objs,$(DTS),$(TRANS0),$(BSTORS0),$(SHAPES),$(SMS),$(SNS),$(SKS),blasfeo)
 BLASFEO_ST_BINS  := $(patsubst %.o,%.x,$(BLASFEO_ST_OBJS))
 
+LIBXSMM_ST_OBJS  := $(call get-st-objs,$(DTS),$(TRANS0),$(XSTORS0),$(SHAPES),$(SMS),$(SNS),$(SKS),libxsmm)
+LIBXSMM_ST_BINS  := $(patsubst %.o,%.x,$(LIBXSMM_ST_OBJS))
+
 VENDOR_ST_OBJS   := $(call get-st-objs,$(DTS),$(TRANS0),$(BSTORS0),$(SHAPES),$(SMS),$(SNS),$(SKS),vendor)
 VENDOR_ST_BINS   := $(patsubst %.o,%.x,$(VENDOR_ST_OBJS))
 
@@ -351,6 +361,7 @@ VENDOR_ST_BINS   := $(patsubst %.o,%.x,$(VENDOR_ST_OBJS))
                $(EIGEN_ST_OBJS) \
                $(OPENBLAS_ST_OBJS) \
                $(BLASFEO_ST_OBJS) \
+               $(LIBXSMM_ST_OBJS) \
                $(VENDOR_ST_OBJS)
 
 
@@ -365,9 +376,11 @@ blislpab:    blislpab-st
 eigen:       eigen-st
 openblas:    openblas-st
 blasfeo:     blasfeo-st
+libxsmm:     libxsmm-st
 vendor:      vendor-st
 
-st:          blissup-st blislpab-st eigen-st openblas-st blasfeo-st vendor-st
+st:          blissup-st blislpab-st \
+             eigen-st openblas-st blasfeo-st libxsmm-st vendor-st
 blis:        blissup-st blislpab-st
 
 blissup-st:  $(BLISSUP_ST_BINS)
@@ -375,13 +388,14 @@ blislpab-st: $(BLISLPAB_ST_BINS)
 eigen-st:    $(EIGEN_ST_BINS)
 openblas-st: $(OPENBLAS_ST_BINS)
 blasfeo-st:  $(BLASFEO_ST_BINS)
+libxsmm-st:  $(LIBXSMM_ST_BINS)
 vendor-st:   $(VENDOR_ST_BINS)
 
 
 # --Object file rules --
 
 # Define the implementations for which we will instantiate compilation rules.
-BIMPLS := blissup blislpab openblas blasfeo vendor
+BIMPLS := blissup blislpab openblas blasfeo libxsmm vendor
 EIMPLS := eigen
 
 #      1     2  3   4 567  8
@@ -447,6 +461,9 @@ test_%_openblas_st.x: test_%_openblas_st.o $(LIBBLIS_LINK)
 test_%_blasfeo_st.x:  test_%_blasfeo_st.o  $(LIBBLIS_LINK)
 	$(CC) $(strip $<  $(BLASFEO_LIB)       $(LIBBLIS_LINK) $(LDFLAGS) -o $@)
 
+test_%_libxsmm_st.x:  test_%_libxsmm_st.o  $(LIBBLIS_LINK)
+	$(CC) $(strip $<  $(LIBXSMM_LIB)       $(LIBBLIS_LINK) $(LDFLAGS) -o $@)
+
 test_%_vendor_st.x:   test_%_vendor_st.o   $(LIBBLIS_LINK)
 	$(CC) $(strip $<  $(VENDOR_LIB)        $(LIBBLIS_LINK) $(LDFLAGS) -o $@)
 
diff --git a/test/sup/runme.sh b/test/sup/runme.sh
index 9646e3ccc..912c50dc0 100755
--- a/test/sup/runme.sh
+++ b/test/sup/runme.sh
@@ -37,13 +37,14 @@ sns="8"
 sks="4"
 
 # Implementations to test.
-impls="vendor blissup blislpab openblas eigen"
+impls="vendor blissup blislpab openblas eigen libxsmm blasfeo"
 #impls="vendor openblas eigen"
 #impls="blislpab blissup"
-#mpls="openblas eigen vendor"
-#mpls="eigen"
+#impls="openblas eigen vendor"
+#impls="eigen"
 #impls="blissup"
 #impls="blasfeo"
+#impls="libxsmm"
 
 # Example: test_dgemm_nn_rrc_m6npkp_blissup_st.x
 
@@ -75,6 +76,13 @@ for th in ${threads}; do
 											continue;
 										fi
 
+										# Further limit execution of libxsmm to
+										# ccc storage cases.
+										if [ "${im:0:7}" = "libxsmm" ] && \
+										   [ "${st}" != "ccc" ]; then
+											continue;
+										fi
+
 										# Extract the shape chars for m, n, k.
 										chm=${sh:0:1}
 										chn=${sh:1:1}
diff --git a/test/sup/test_gemm.c b/test/sup/test_gemm.c
index 311e8552a..8c9f24c8a 100644
--- a/test/sup/test_gemm.c
+++ b/test/sup/test_gemm.c
@@ -158,7 +158,8 @@ int main( int argc, char** argv )
 	        ( unsigned long )0, 0.0 );
 
 
-	for ( p = p_begin; p <= p_max; p += p_inc )
+	//for ( p = p_begin; p <= p_max; p += p_inc )
+	for ( p = p_max; p_begin <= p; p -= p_inc )
 	{
 		obj_t  a, b, c;
 		obj_t  c_save;