diff --git a/Makefile.system b/Makefile.system
index 2a0694d01..1b473c59d 100644
--- a/Makefile.system
+++ b/Makefile.system
@@ -286,8 +286,15 @@ GCCVERSIONEQ5 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` = 5)
 GCCVERSIONGT5 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \> 5)
 GCCVERSIONGTEQ7 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 7)
 GCCVERSIONGTEQ9 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 9)
-GCCMINORVERSIONGTEQ2 := $(shell expr `$(CC) -dumpversion | cut -f2 -d.` \>= 2)
-GCCMINORVERSIONGTEQ7 := $(shell expr `$(CC) -dumpversion | cut -f2 -d.` \>= 7)
+# Note that the behavior of -dumpversion is compile-time-configurable for
+# gcc-7.x and newer. Use -dumpfullversion there
+ifeq ($(GCCVERSIONGTEQ7),1)
+	GCCDUMPVERSION_PARAM := -dumpfullversion
+else
+	GCCDUMPVERSION_PARAM := -dumpversion
+endif
+GCCMINORVERSIONGTEQ2 := $(shell expr `$(CC) $(GCCDUMPVERSION_PARAM) | cut -f2 -d.` \>= 2)
+GCCMINORVERSIONGTEQ7 := $(shell expr `$(CC) $(GCCDUMPVERSION_PARAM) | cut -f2 -d.` \>= 7)
 endif
 
 #
diff --git a/Makefile.x86_64 b/Makefile.x86_64
index f2de51ef4..2676bd258 100644
--- a/Makefile.x86_64
+++ b/Makefile.x86_64
@@ -31,14 +31,24 @@ ifeq ($(CORE), HASWELL)
 ifndef DYNAMIC_ARCH
 ifndef NO_AVX2
 ifeq ($(C_COMPILER), GCC)
+# AVX2 support was added in 4.7.0
+GCCVERSIONGTEQ4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 4)
+GCCMINORVERSIONGTEQ7 := $(shell expr `$(CC) -dumpversion | cut -f2 -d.` \>= 7)
+ifeq ($(GCCVERSIONGTEQ4)$(GCCMINORVERSIONGTEQ7), 11)
 CCOMMON_OPT += -mavx2
 endif
+endif
 ifeq ($(F_COMPILER), GFORTRAN)
+# AVX2 support was added in 4.7.0
+GCCVERSIONGTEQ4 := $(shell expr `$(FC) -dumpversion | cut -f1 -d.` \>= 4)
+GCCMINORVERSIONGTEQ7 := $(shell expr `$(FC) -dumpversion | cut -f2 -d.` \>= 7)
+ifeq ($(GCCVERSIONGTEQ4)$(GCCMINORVERSIONGTEQ7), 11)
 FCOMMON_OPT += -mavx2
 endif
 endif
 endif
 endif
+endif
 
 
 
diff --git a/cpuid_x86.c b/cpuid_x86.c
index e29adecae..356800b78 100644
--- a/cpuid_x86.c
+++ b/cpuid_x86.c
@@ -1406,6 +1406,17 @@ int get_cpuname(void){
 	    return CPUTYPE_SANDYBRIDGE;
           else
 	    return CPUTYPE_NEHALEM;
+    }
+      case 10: //family 6 exmodel 10
+        switch (model) {
+    case 5: // Comet Lake H and S
+    case 6: // Comet Lake U
+          if(support_avx2())
+            return CPUTYPE_HASWELL;
+          if(support_avx())
+        return CPUTYPE_SANDYBRIDGE;
+          else
+        return CPUTYPE_NEHALEM;
 	}
 	break;    
       }
@@ -1955,6 +1966,19 @@ int get_coretype(void){
 	    return CORE_NEHALEM;
         }
         break;
+      case 10:
+        switch (model) {
+	  case 5: // Comet Lake H and S
+    	  case 6: // Comet Lake U
+            if(support_avx())
+  #ifndef NO_AVX2
+              return CORE_HASWELL;
+  #else
+              return CORE_SANDYBRIDGE;
+  #endif
+            else
+              return CORE_NEHALEM;
+        }
       case 5:
         switch (model) {
 	case 6:
diff --git a/driver/others/dynamic.c b/driver/others/dynamic.c
index 38eb76643..c03b0b21d 100644
--- a/driver/others/dynamic.c
+++ b/driver/others/dynamic.c
@@ -618,6 +618,18 @@ static gotoblas_t *get_coretype(void){
 	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
 	  }
 	}
+      case 10:
+    if (model == 5 || model == 6) {
+	  if(support_avx2())
+	    return &gotoblas_HASWELL;
+	  if(support_avx()) {
+	    openblas_warning(FALLBACK_VERBOSE, SANDYBRIDGE_FALLBACK);
+	    return &gotoblas_SANDYBRIDGE;
+	  } else {
+	    openblas_warning(FALLBACK_VERBOSE, NEHALEM_FALLBACK);
+	    return &gotoblas_NEHALEM; //OS doesn't support AVX. Use old kernels.
+	  }
+    }
 	return NULL;
       }
       case 0xf:
diff --git a/kernel/Makefile.L3 b/kernel/Makefile.L3
index 3d63ff861..dfdaf5cf4 100644
--- a/kernel/Makefile.L3
+++ b/kernel/Makefile.L3
@@ -482,8 +482,8 @@ $(KDIR)$(SHGEMMONCOPYOBJ) : $(KERNELDIR)/$(SHGEMMONCOPY)
 
 $(KDIR)$(SHGEMMOTCOPYOBJ) : $(KERNELDIR)/$(SHGEMMOTCOPY)
 
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DHALF -UDOUBLE -UCOMPLEX $< -o shgemmotcopy.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DHALF -UDOUBLE -UCOMPLEX $< -o - > shgemmotcopy.s
 	m4 shgemmotcopy.s > shgemmotcopy_nomacros.s
 	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX shgemmotcopy_nomacros.s -o $@
 	rm shgemmotcopy.s shgemmotcopy_nomacros.s
@@ -497,8 +497,8 @@ $(KDIR)$(SHGEMMINCOPYOBJ) : $(KERNELDIR)/$(SHGEMMINCOPY)
 	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)$(SHGEMMITCOPYOBJ) : $(KERNELDIR)/$(SHGEMMITCOPY)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DHALF -UDOUBLE -UCOMPLEX $< -o shgemmitcopy.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DHALF -UDOUBLE -UCOMPLEX $< -o - > shgemmitcopy.s
 	m4 shgemmitcopy.s > shgemmitcopy_nomacros.s
 	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX shgemmitcopy_nomacros.s -o $@
 	rm shgemmitcopy.s shgemmitcopy_nomacros.s
@@ -513,8 +513,8 @@ $(KDIR)$(SGEMMONCOPYOBJ) : $(KERNELDIR)/$(SGEMMONCOPY)
 	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)$(SGEMMOTCOPYOBJ) : $(KERNELDIR)/$(SGEMMOTCOPY)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -UDOUBLE -UCOMPLEX $< -o sgemmotcopy.s	
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -UDOUBLE -UCOMPLEX $< -o - > sgemmotcopy.s
 	m4 sgemmotcopy.s > sgemmotcopy_nomacros.s
 	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX sgemmotcopy_nomacros.s -o $@
 	rm sgemmotcopy.s sgemmotcopy_nomacros.s
@@ -529,8 +529,8 @@ $(KDIR)$(SGEMMINCOPYOBJ) : $(KERNELDIR)/$(SGEMMINCOPY)
 	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)$(SGEMMITCOPYOBJ) : $(KERNELDIR)/$(SGEMMITCOPY)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -UDOUBLE -UCOMPLEX $< -o sgemmitcopy.s	
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -UDOUBLE -UCOMPLEX $< -o - > sgemmitcopy.s
 	m4 sgemmitcopy.s > sgemmitcopy_nomacros.s
 	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX sgemmitcopy_nomacros.s -o $@
 	rm sgemmitcopy.s sgemmitcopy_nomacros.s
@@ -541,8 +541,8 @@ endif
 endif
 
 $(KDIR)$(DGEMMONCOPYOBJ) : $(KERNELDIR)/$(DGEMMONCOPY)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DDOUBLE -UCOMPLEX $< -o dgemm_ncopy.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DDOUBLE -UCOMPLEX $< -o - > dgemm_ncopy.s
 	m4 dgemm_ncopy.s > dgemm_ncopy_nomacros.s
 	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX dgemm_ncopy_nomacros.s -o $@
 	rm dgemm_ncopy.s dgemm_ncopy_nomacros.s
@@ -559,8 +559,8 @@ $(KDIR)$(DGEMMINCOPYOBJ) : $(KERNELDIR)/$(DGEMMINCOPY)
 	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)$(DGEMMITCOPYOBJ) : $(KERNELDIR)/$(DGEMMITCOPY)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DDOUBLE -UCOMPLEX $< -o dgemm_itcopy.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DDOUBLE -UCOMPLEX $< -o - > dgemm_itcopy.s
 	m4 dgemm_itcopy.s > dgemm_itcopy_nomacros.s
 	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX dgemm_itcopy_nomacros.s -o $@
 	rm dgemm_itcopy.s dgemm_itcopy_nomacros.s
@@ -602,8 +602,8 @@ $(KDIR)$(CGEMMINCOPYOBJ) : $(KERNELDIR)/$(CGEMMINCOPY)
 	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)$(CGEMMITCOPYOBJ) : $(KERNELDIR)/$(CGEMMITCOPY)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -UDOUBLE -UCOMPLEX -E $< -o cgemm_itcopy.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -UDOUBLE -UCOMPLEX -S $< -o - > cgemm_itcopy.s
 	m4 cgemm_itcopy.s > cgemm_itcopy_nomacros.s
 	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX cgemm_itcopy_nomacros.s -o $@
 	rm cgemm_itcopy.s cgemm_itcopy_nomacros.s
@@ -625,8 +625,8 @@ $(KDIR)$(ZGEMMINCOPYOBJ) : $(KERNELDIR)/$(ZGEMMINCOPY)
 	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)$(ZGEMMITCOPYOBJ) : $(KERNELDIR)/$(ZGEMMITCOPY)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DDOUBLE -UCOMPLEX $< -o zgemm_itcopy.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DDOUBLE -UCOMPLEX $< -o - > zgemm_itcopy.s
 	m4 zgemm_itcopy.s > zgemm_itcopy_nomacros.s
 	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX zgemm_itcopy_nomacros.s -o $@
 	rm zgemm_itcopy.s zgemm_itcopy_nomacros.s
@@ -657,8 +657,8 @@ endif
 endif
 
 $(KDIR)sgemm_kernel$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL) $(SGEMMDEPEND)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -UDOUBLE -UCOMPLEX  $< -o sgemm_kernel$(TSUFFIX).s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -UDOUBLE -UCOMPLEX  $< -o - > sgemm_kernel$(TSUFFIX).s
 	m4 sgemm_kernel$(TSUFFIX).s > sgemm_kernel$(TSUFFIX)_nomacros.s
 	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX sgemm_kernel$(TSUFFIX)_nomacros.s -o $@
 	rm sgemm_kernel$(TSUFFIX).s sgemm_kernel$(TSUFFIX)_nomacros.s
@@ -669,8 +669,8 @@ endif
 ifeq ($(BUILD_HALF), 1)
 
 $(KDIR)shgemm_kernel$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SHGEMMKERNEL) $(SHGEMMDEPEND)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DHALF -UDOUBLE -UCOMPLEX  $< -o shgemm_kernel$(TSUFFIX).s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DHALF -UDOUBLE -UCOMPLEX  $< -o - > shgemm_kernel$(TSUFFIX).s
 	m4 shgemm_kernel$(TSUFFIX).s > shgemm_kernel$(TSUFFIX)_nomacros.s
 	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX shgemm_kernel$(TSUFFIX)_nomacros.s -o $@
 	rm shgemm_kernel$(TSUFFIX).s shgemm_kernel$(TSUFFIX)_nomacros.s
@@ -680,8 +680,8 @@ endif
 endif
 
 $(KDIR)dgemm_kernel$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMMKERNEL) $(DGEMMDEPEND)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DDOUBLE -UCOMPLEX $< -o dgemm_kernel$(TSUFFIX).s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DDOUBLE -UCOMPLEX $< -o - > dgemm_kernel$(TSUFFIX).s
 	m4 dgemm_kernel$(TSUFFIX).s > dgemm_kernel$(TSUFFIX)_nomacros.s
 	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX dgemm_kernel$(TSUFFIX)_nomacros.s -o $@
 	rm dgemm_kernel$(TSUFFIX).s dgemm_kernel$(TSUFFIX)_nomacros.s
@@ -693,9 +693,9 @@ $(KDIR)qgemm_kernel$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL) $(QGEMMDEP
 	$(CC) $(CFLAGS) -c -DXDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)cgemm_kernel_n$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL) $(CGEMMDEPEND)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -UDOUBLE -DCOMPLEX -DNN $< -o cgemm_kernel_n.s
-	m4 -B 16384 cgemm_kernel_n.s > cgemm_kernel_n_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -UDOUBLE -DCOMPLEX -DNN $< -o - > cgemm_kernel_n.s
+	m4 cgemm_kernel_n.s > cgemm_kernel_n_nomacros.s
 	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DNN cgemm_kernel_n_nomacros.s -o $@
 	rm cgemm_kernel_n.s cgemm_kernel_n_nomacros.s
 else
@@ -703,9 +703,9 @@ else
 endif
 
 $(KDIR)cgemm_kernel_l$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL) $(CGEMMDEPEND)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -UDOUBLE -DCOMPLEX -DCN $< -o cgemm_kernel_l.s
-	m4 -B 16384 cgemm_kernel_l.s > cgemm_kernel_l_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -UDOUBLE -DCOMPLEX -DCN $< -o - > cgemm_kernel_l.s
+	m4 cgemm_kernel_l.s > cgemm_kernel_l_nomacros.s
 	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DCN cgemm_kernel_l_nomacros.s -o $@
 	rm cgemm_kernel_l.s cgemm_kernel_l_nomacros.s
 else
@@ -713,9 +713,9 @@ else
 endif
 
 $(KDIR)cgemm_kernel_r$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL) $(CGEMMDEPEND)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -UDOUBLE -DCOMPLEX -DNC  $< -o cgemm_kernel_r.s
-	m4 -B 16384 cgemm_kernel_r.s > cgemm_kernel_r_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -UDOUBLE -DCOMPLEX -DNC  $< -o - > cgemm_kernel_r.s
+	m4 cgemm_kernel_r.s > cgemm_kernel_r_nomacros.s
 	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DNC cgemm_kernel_r_nomacros.s -o $@
 	rm cgemm_kernel_r.s cgemm_kernel_r_nomacros.s
 else
@@ -723,9 +723,9 @@ else
 endif
 
 $(KDIR)cgemm_kernel_b$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL) $(CGEMMDEPEND)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -UDOUBLE -DCOMPLEX -DCC $< -o cgemm_kernel_b.s
-	m4 -B 16384 cgemm_kernel_b.s > cgemm_kernel_b_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -UDOUBLE -DCOMPLEX -DCC $< -o - > cgemm_kernel_b.s
+	m4 cgemm_kernel_b.s > cgemm_kernel_b_nomacros.s
 	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DCC cgemm_kernel_b_nomacros.s -o $@
 	rm cgemm_kernel_b.s cgemm_kernel_b_nomacros.s
 else
@@ -733,9 +733,9 @@ else
 endif
 
 $(KDIR)zgemm_kernel_n$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL) $(ZGEMMDEPEND)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DDOUBLE -DCOMPLEX -DNN $< -o zgemm_kernel_n.s
-	m4 -B 16384 zgemm_kernel_n.s > zgemm_kernel_n_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DDOUBLE -DCOMPLEX -DNN $< -o - > zgemm_kernel_n.s
+	m4 zgemm_kernel_n.s > zgemm_kernel_n_nomacros.s
 	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DNN zgemm_kernel_n_nomacros.s -o $@
 	rm zgemm_kernel_n.s zgemm_kernel_n_nomacros.s
 else
@@ -743,9 +743,9 @@ else
 endif
 
 $(KDIR)zgemm_kernel_l$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL) $(ZGEMMDEPEND)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DDOUBLE -DCOMPLEX -DCN $< -o zgemm_kernel_l.s
-	m4 -B 16384 zgemm_kernel_l.s > zgemm_kernel_l_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DDOUBLE -DCOMPLEX -DCN $< -o - > zgemm_kernel_l.s
+	m4 zgemm_kernel_l.s > zgemm_kernel_l_nomacros.s
 	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DCN zgemm_kernel_l_nomacros.s -o $@
 	rm zgemm_kernel_l.s zgemm_kernel_l_nomacros.s
 else
@@ -753,9 +753,9 @@ else
 endif
 
 $(KDIR)zgemm_kernel_r$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL) $(ZGEMMDEPEND)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DDOUBLE -DCOMPLEX -DNC $< -o zgemm_kernel_r.s
-	m4 -B 16384 zgemm_kernel_r.s > zgemm_kernel_r_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DDOUBLE -DCOMPLEX -DNC $< -o - > zgemm_kernel_r.s
+	m4 zgemm_kernel_r.s > zgemm_kernel_r_nomacros.s
 	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DNC zgemm_kernel_r_nomacros.s -o $@
 	rm zgemm_kernel_r.s zgemm_kernel_r_nomacros.s
 else
@@ -763,9 +763,9 @@ else
 endif
 
 $(KDIR)zgemm_kernel_b$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL) $(ZGEMMDEPEND)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DDOUBLE -DCOMPLEX -DCC $< -o zgemm_kernel_b.s
-	m4 -B 16384 zgemm_kernel_b.s > zgemm_kernel_b_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DDOUBLE -DCOMPLEX -DCC $< -o - > zgemm_kernel_b.s
+	m4 zgemm_kernel_b.s > zgemm_kernel_b_nomacros.s
 	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DCC zgemm_kernel_b_nomacros.s -o $@
 	rm zgemm_kernel_b.s zgemm_kernel_b_nomacros.s
 else
@@ -787,8 +787,8 @@ $(KDIR)xgemm_kernel_b$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL) $(XGEMMD
 
 ifdef USE_TRMM
 $(KDIR)strmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o strmmkernel_ln.s	
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o - > strmmkernel_ln.s	
 	m4 strmmkernel_ln.s > strmmkernel_ln_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -UTRANSA strmmkernel_ln_nomacros.s -o $@
 	rm strmmkernel_ln.s strmmkernel_ln_nomacros.s
@@ -797,8 +797,8 @@ else
 endif
 
 $(KDIR)strmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o strmmkernel_lt.s	
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o - > strmmkernel_lt.s	
 	m4 strmmkernel_lt.s > strmmkernel_lt_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -DTRANSA strmmkernel_lt_nomacros.s -o $@
 	rm strmmkernel_lt.s strmmkernel_lt_nomacros.s
@@ -807,8 +807,8 @@ else
 endif
 
 $(KDIR)strmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o strmmkernel_rn.s	
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o - > strmmkernel_rn.s	
 	m4 strmmkernel_rn.s > strmmkernel_rn_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -UTRANSA strmmkernel_rn_nomacros.s -o $@
 	rm strmmkernel_rn.s strmmkernel_rn_nomacros.s
@@ -817,8 +817,8 @@ else
 endif
 
 $(KDIR)strmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o strmm_kernel_rt.s	
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o - > strmm_kernel_rt.s	
 	m4 strmm_kernel_rt.s > strmm_kernel_rt_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA strmm_kernel_rt_nomacros.s -o $@
 	rm strmm_kernel_rt.s strmm_kernel_rt_nomacros.s
@@ -827,8 +827,8 @@ else
 endif
 
 $(KDIR)dtrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o dtrmm_kernel_ln.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o - > dtrmm_kernel_ln.s
 	m4 dtrmm_kernel_ln.s > dtrmm_kernel_ln_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -UTRANSA dtrmm_kernel_ln_nomacros.s -o $@
 	rm dtrmm_kernel_ln.s dtrmm_kernel_ln_nomacros.s
@@ -837,8 +837,8 @@ else
 endif
 
 $(KDIR)dtrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o dtrmm_kernel_lt.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o - > dtrmm_kernel_lt.s
 	m4 dtrmm_kernel_lt.s > dtrmm_kernel_lt_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -DTRANSA dtrmm_kernel_lt_nomacros.s -o $@
 	rm dtrmm_kernel_lt.s dtrmm_kernel_lt_nomacros.s
@@ -847,8 +847,8 @@ else
 endif
 
 $(KDIR)dtrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o dtrmm_kernel_rn.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o - > dtrmm_kernel_rn.s
 	m4 dtrmm_kernel_rn.s > dtrmm_kernel_rn_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -UTRANSA dtrmm_kernel_rn_nomacros.s -o $@
 	rm dtrmm_kernel_rn.s dtrmm_kernel_rn_nomacros.s
@@ -857,8 +857,8 @@ else
 endif
 
 $(KDIR)dtrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o dtrmm_kernel_rt.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o - > dtrmm_kernel_rt.s
 	m4 dtrmm_kernel_rt.s > dtrmm_kernel_rt_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -DTRANSA dtrmm_kernel_rt_nomacros.s -o $@
 	rm dtrmm_kernel_rt.s dtrmm_kernel_rt_nomacros.s
@@ -879,9 +879,9 @@ $(KDIR)qtrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
 
 $(KDIR)ctrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN  $< -o ctrmm_kernel_ln.s
-	m4 -B 16384 ctrmm_kernel_ln.s > ctrmm_kernel_ln_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN  $< -o - > ctrmm_kernel_ln.s
+	m4 ctrmm_kernel_ln.s > ctrmm_kernel_ln_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN ctrmm_kernel_ln_nomacros.s -o $@
 	rm ctrmm_kernel_ln.s ctrmm_kernel_ln_nomacros.s
 else
@@ -889,9 +889,9 @@ else
 endif
 
 $(KDIR)ctrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o ctrmm_kernel_lt.s
-	m4 -B 16384 ctrmm_kernel_lt.s > ctrmm_kernel_lt_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o - > ctrmm_kernel_lt.s
+	m4 ctrmm_kernel_lt.s > ctrmm_kernel_lt_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN ctrmm_kernel_lt_nomacros.s -o $@
 	rm ctrmm_kernel_lt.s ctrmm_kernel_lt_nomacros.s
 else
@@ -899,9 +899,9 @@ else
 endif
 
 $(KDIR)ctrmm_kernel_LR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o ctrmm_kernel_lr.s
-	m4 -B 16384 ctrmm_kernel_lr.s > ctrmm_kernel_lr_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o - > ctrmm_kernel_lr.s
+	m4 ctrmm_kernel_lr.s > ctrmm_kernel_lr_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN  ctrmm_kernel_lr_nomacros.s -o $@
 	rm ctrmm_kernel_lr.s ctrmm_kernel_lr_nomacros.s
 else
@@ -909,9 +909,9 @@ else
 endif
 
 $(KDIR)ctrmm_kernel_LC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o ctrmm_kernel_lc.s
-	m4 -B 16384 ctrmm_kernel_lc.s > ctrmm_kernel_lc_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o - > ctrmm_kernel_lc.s
+	m4 ctrmm_kernel_lc.s > ctrmm_kernel_lc_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN ctrmm_kernel_lc_nomacros.s -o $@
 	rm ctrmm_kernel_lc_nomacros.s ctrmm_kernel_lc.s
 else
@@ -919,9 +919,9 @@ else
 endif
 
 $(KDIR)ctrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o ctrmm_kernel_rn.s
-	m4 -B 16384 ctrmm_kernel_rn.s > ctrmm_kernel_rn_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o - > ctrmm_kernel_rn.s
+	m4 ctrmm_kernel_rn.s > ctrmm_kernel_rn_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN ctrmm_kernel_rn_nomacros.s -o $@
 	rm ctrmm_kernel_rn.s ctrmm_kernel_rn_nomacros.s
 else
@@ -929,9 +929,9 @@ else
 endif
 
 $(KDIR)ctrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o ctrmm_kernel_rt.s
-	m4 -B 16384 ctrmm_kernel_rt.s > ctrmm_kernel_rt_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o - > ctrmm_kernel_rt.s
+	m4 ctrmm_kernel_rt.s > ctrmm_kernel_rt_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN ctrmm_kernel_rt_nomacros.s -o $@
 	rm ctrmm_kernel_rt.s ctrmm_kernel_rt_nomacros.s
 else
@@ -939,9 +939,9 @@ else
 endif
 
 $(KDIR)ctrmm_kernel_RR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o ctrmm_kernel_rr.s
-	m4 -B 16384 ctrmm_kernel_rr.s > ctrmm_kernel_rr_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o - > ctrmm_kernel_rr.s
+	m4 ctrmm_kernel_rr.s > ctrmm_kernel_rr_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC ctrmm_kernel_rr_nomacros.s -o $@
 	rm ctrmm_kernel_rr.s ctrmm_kernel_rr_nomacros.s
 else
@@ -949,9 +949,9 @@ else
 endif
 
 $(KDIR)ctrmm_kernel_RC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o ctrmm_kernel_RC.s
-	m4 -B 16384 ctrmm_kernel_RC.s > ctrmm_kernel_RC_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o - > ctrmm_kernel_RC.s
+	m4 ctrmm_kernel_RC.s > ctrmm_kernel_RC_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC ctrmm_kernel_RC_nomacros.s -o $@
 	rm ctrmm_kernel_RC.s ctrmm_kernel_RC_nomacros.s
 else
@@ -959,9 +959,9 @@ else
 endif
 
 $(KDIR)ztrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o ztrmm_kernel_ln.s
-	m4 -B 16384 ztrmm_kernel_ln.s > ztrmm_kernel_ln_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o - > ztrmm_kernel_ln.s
+	m4 ztrmm_kernel_ln.s > ztrmm_kernel_ln_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN ztrmm_kernel_ln_nomacros.s -o $@
 	rm ztrmm_kernel_ln.s ztrmm_kernel_ln_nomacros.s
 else
@@ -969,9 +969,9 @@ else
 endif
 
 $(KDIR)ztrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o ztrmm_kernel_lt.s
-	m4 -B 16384 ztrmm_kernel_lt.s > ztrmm_kernel_lt_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o - > ztrmm_kernel_lt.s
+	m4 ztrmm_kernel_lt.s > ztrmm_kernel_lt_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN ztrmm_kernel_lt_nomacros.s -o $@
 	rm ztrmm_kernel_lt.s ztrmm_kernel_lt_nomacros.s
 else
@@ -979,9 +979,9 @@ else
 endif
 
 $(KDIR)ztrmm_kernel_LR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o ztrmm_kernel_lr.s
-	m4 -B 16384 ztrmm_kernel_lr.s > ztrmm_kernel_lr_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o - > ztrmm_kernel_lr.s
+	m4 ztrmm_kernel_lr.s > ztrmm_kernel_lr_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN ztrmm_kernel_lr_nomacros.s -o $@
 	rm ztrmm_kernel_lr.s ztrmm_kernel_lr_nomacros.s
 else
@@ -989,9 +989,9 @@ else
 endif
 
 $(KDIR)ztrmm_kernel_LC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o ztrmm_kernel_lc.s
-	m4 -B 16384 ztrmm_kernel_lc.s >ztrmm_kernel_lc_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o - > ztrmm_kernel_lc.s
+	m4 ztrmm_kernel_lc.s >ztrmm_kernel_lc_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN ztrmm_kernel_lc_nomacros.s -o $@
 	rm ztrmm_kernel_lc.s ztrmm_kernel_lc_nomacros.s 
 else
@@ -999,9 +999,9 @@ else
 endif
 
 $(KDIR)ztrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o ztrmm_kernel_rn.s
-	m4 -B 16384 ztrmm_kernel_rn.s > ztrmm_kernel_rn_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o - > ztrmm_kernel_rn.s
+	m4 ztrmm_kernel_rn.s > ztrmm_kernel_rn_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN ztrmm_kernel_rn_nomacros.s -o $@
 	rm ztrmm_kernel_rn.s ztrmm_kernel_rn_nomacros.s
 else
@@ -1009,9 +1009,9 @@ else
 endif
 
 $(KDIR)ztrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o ztrmm_kernel_rt.s
-	m4 -B 16384 ztrmm_kernel_rt.s > ztrmm_kernel_rt_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o - > ztrmm_kernel_rt.s
+	m4 ztrmm_kernel_rt.s > ztrmm_kernel_rt_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN ztrmm_kernel_rt_nomacros.s -o $@
 	rm ztrmm_kernel_rt.s ztrmm_kernel_rt_nomacros.s
 else
@@ -1019,8 +1019,8 @@ else
 endif
 
 $(KDIR)ztrmm_kernel_RR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o ztrmm_kernel_rr.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o - > ztrmm_kernel_rr.s
 	m4 ztrmm_kernel_rr.s > ztrmm_kernel_rr_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC ztrmm_kernel_rr_nomacros.s -o $@
 	rm ztrmm_kernel_rr.s ztrmm_kernel_rr_nomacros.s
@@ -1029,9 +1029,9 @@ else
 endif
 
 $(KDIR)ztrmm_kernel_RC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o ztrmm_kernel_rc.s
-	m4 -B 16384 ztrmm_kernel_rc.s > ztrmm_kernel_rc_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o - > ztrmm_kernel_rc.s
+	m4 ztrmm_kernel_rc.s > ztrmm_kernel_rc_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC ztrmm_kernel_rc_nomacros.s -o $@
 	rm ztrmm_kernel_rc.s ztrmm_kernel_rc_nomacros.s
 else
@@ -1049,9 +1049,9 @@ $(KDIR)strmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL)
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
 
 $(KDIR)strmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o strmm_kernel_rt.s	
-	m4 -B 16384 strmm_kernel_rt.s > strmm_kernel_rt_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o - > strmm_kernel_rt.s	
+	m4 strmm_kernel_rt.s > strmm_kernel_rt_nomacros.s
 	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA strmm_kernel_rt_nomacros.s -o $@
 	rm strmm_kernel_rt.s strmm_kernel_rt_nomacros.s
 else
@@ -1183,9 +1183,9 @@ $(KDIR)dtrsm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRSMKERNEL_LN) $(DT
 	$(CC) -c $(CFLAGS) -DTRSMKERNEL -UCOMPLEX -DDOUBLE -DUPPER -DLN -UCONJ $< -o $@
 
 $(KDIR)dtrsm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRSMKERNEL_LT) $(DTRSMDEPEND)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRSMKERNEL -UCOMPLEX -DDOUBLE -UUPPER -DLT -UCONJ $< -o dtrsm_kernel_lt.s
-	m4 -B 16384 dtrsm_kernel_lt.s > dtrsm_kernel_lt_nomacros.s
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRSMKERNEL -UCOMPLEX -DDOUBLE -UUPPER -DLT -UCONJ $< -o - > dtrsm_kernel_lt.s
+	m4 dtrsm_kernel_lt.s > dtrsm_kernel_lt_nomacros.s
 	$(CC) -c $(CFLAGS) -DTRSMKERNEL -UCOMPLEX -DDOUBLE -UUPPER -DLT -UCONJ dtrsm_kernel_lt_nomacros.s -o $@
 	rm dtrsm_kernel_lt.s dtrsm_kernel_lt_nomacros.s
 else
@@ -2459,8 +2459,8 @@ $(KDIR)cgemm_kernel_l$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL) $(CGEMM
 	$(CC) $(PFLAGS) -c -UDOUBLE -DCOMPLEX -DCN $< -o $@
 
 $(KDIR)cgemm_kernel_r$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL) $(CGEMMDEPEND)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(PFLAGS) -E -UDOUBLE -DCOMPLEX -DNC $< -o cgemm_kernel_r.s
+ifeq ($(OS), AIX)
+	$(CC) $(PFLAGS) -S -UDOUBLE -DCOMPLEX -DNC $< -o - > cgemm_kernel_r.s
 	m4 cgemm_kernel_r.s > cgemm_kernel_r_nomacros.s
 	$(CC) $(PFLAGS) -c -UDOUBLE -DCOMPLEX -DNC cgemm_kernel_r_nomacros.s -o $@
 	rm cgemm_kernel_r.s cgemm_kernel_r_nomacros.s 
@@ -2505,8 +2505,8 @@ $(KDIR)strmm_kernel_RN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL)
 	$(CC) $(PFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
 
 $(KDIR)strmm_kernel_RT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL)
-ifeq ($(findstring AIXPOW, $(OS)$(filter $(TARGET), POWER8 POWER9 POWER10)),AIXPOW)
-	$(CC) $(CFLAGS) -E -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o strmm_kernel_rt.s	
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o - > strmm_kernel_rt.s	
 	m4 strmmkernel_rn.s > strmm_kernel_rt_nomacros.s
 	$(CC) $(PFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA strmm_kernel_rt_nomacros.s -o $@
 	rm strmm_kernel_rt.s strmm_kernel_rt_nomacros.s
diff --git a/kernel/generic/gemm_ncopy_16.c b/kernel/generic/gemm_ncopy_16.c
index 5f91d0dbe..d3ab46472 100644
--- a/kernel/generic/gemm_ncopy_16.c
+++ b/kernel/generic/gemm_ncopy_16.c
@@ -39,24 +39,24 @@
 #include <stdio.h>
 #include "common.h"
 
-int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b){
+int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
   BLASLONG i, j;
 
-  FLOAT *aoffset;
-  FLOAT *aoffset1, *aoffset2, *aoffset3, *aoffset4;
-  FLOAT *aoffset5, *aoffset6, *aoffset7, *aoffset8;
-  FLOAT *aoffset9, *aoffset10, *aoffset11, *aoffset12;
-  FLOAT *aoffset13, *aoffset14, *aoffset15, *aoffset16;
+  IFLOAT *aoffset;
+  IFLOAT *aoffset1, *aoffset2, *aoffset3, *aoffset4;
+  IFLOAT *aoffset5, *aoffset6, *aoffset7, *aoffset8;
+  IFLOAT *aoffset9, *aoffset10, *aoffset11, *aoffset12;
+  IFLOAT *aoffset13, *aoffset14, *aoffset15, *aoffset16;
 
-  FLOAT *boffset;
-  FLOAT ctemp01, ctemp02, ctemp03, ctemp04;
-  FLOAT ctemp05, ctemp06, ctemp07, ctemp08;
-  FLOAT ctemp09, ctemp10, ctemp11, ctemp12;
-  FLOAT ctemp13, ctemp14, ctemp15, ctemp16;
-  FLOAT ctemp17, ctemp18, ctemp19, ctemp20;
-  FLOAT ctemp21, ctemp22, ctemp23, ctemp24;
-  FLOAT ctemp25, ctemp26, ctemp27, ctemp28;
-  FLOAT ctemp29, ctemp30, ctemp31, ctemp32;
+  IFLOAT *boffset;
+  IFLOAT ctemp01, ctemp02, ctemp03, ctemp04;
+  IFLOAT ctemp05, ctemp06, ctemp07, ctemp08;
+  IFLOAT ctemp09, ctemp10, ctemp11, ctemp12;
+  IFLOAT ctemp13, ctemp14, ctemp15, ctemp16;
+  IFLOAT ctemp17, ctemp18, ctemp19, ctemp20;
+  IFLOAT ctemp21, ctemp22, ctemp23, ctemp24;
+  IFLOAT ctemp25, ctemp26, ctemp27, ctemp28;
+  IFLOAT ctemp29, ctemp30, ctemp31, ctemp32;
 
   aoffset = a;
   boffset = b;
diff --git a/kernel/generic/gemm_ncopy_8.c b/kernel/generic/gemm_ncopy_8.c
index a49a778e6..aaf9c8917 100644
--- a/kernel/generic/gemm_ncopy_8.c
+++ b/kernel/generic/gemm_ncopy_8.c
@@ -39,30 +39,30 @@
 #include <stdio.h>
 #include "common.h"
 
-int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b){
+int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
   BLASLONG i, j;
 
-  FLOAT *aoffset;
-  FLOAT *aoffset1, *aoffset2, *aoffset3, *aoffset4;
-  FLOAT *aoffset5, *aoffset6, *aoffset7, *aoffset8;
+  IFLOAT *aoffset;
+  IFLOAT *aoffset1, *aoffset2, *aoffset3, *aoffset4;
+  IFLOAT *aoffset5, *aoffset6, *aoffset7, *aoffset8;
 
-  FLOAT *boffset;
-  FLOAT ctemp01, ctemp02, ctemp03, ctemp04;
-  FLOAT ctemp05, ctemp06, ctemp07, ctemp08;
-  FLOAT ctemp09, ctemp10, ctemp11, ctemp12;
-  FLOAT ctemp13, ctemp14, ctemp15, ctemp16;
-  FLOAT ctemp17, ctemp18, ctemp19, ctemp20;
-  FLOAT ctemp21, ctemp22, ctemp23, ctemp24;
-  FLOAT ctemp25, ctemp26, ctemp27, ctemp28;
-  FLOAT ctemp29, ctemp30, ctemp31, ctemp32;
-  FLOAT ctemp33, ctemp34, ctemp35, ctemp36;
-  FLOAT ctemp37, ctemp38, ctemp39, ctemp40;
-  FLOAT ctemp41, ctemp42, ctemp43, ctemp44;
-  FLOAT ctemp45, ctemp46, ctemp47, ctemp48;
-  FLOAT ctemp49, ctemp50, ctemp51, ctemp52;
-  FLOAT ctemp53, ctemp54, ctemp55, ctemp56;
-  FLOAT ctemp57, ctemp58, ctemp59, ctemp60;
-  FLOAT ctemp61, ctemp62, ctemp63, ctemp64;
+  IFLOAT *boffset;
+  IFLOAT ctemp01, ctemp02, ctemp03, ctemp04;
+  IFLOAT ctemp05, ctemp06, ctemp07, ctemp08;
+  IFLOAT ctemp09, ctemp10, ctemp11, ctemp12;
+  IFLOAT ctemp13, ctemp14, ctemp15, ctemp16;
+  IFLOAT ctemp17, ctemp18, ctemp19, ctemp20;
+  IFLOAT ctemp21, ctemp22, ctemp23, ctemp24;
+  IFLOAT ctemp25, ctemp26, ctemp27, ctemp28;
+  IFLOAT ctemp29, ctemp30, ctemp31, ctemp32;
+  IFLOAT ctemp33, ctemp34, ctemp35, ctemp36;
+  IFLOAT ctemp37, ctemp38, ctemp39, ctemp40;
+  IFLOAT ctemp41, ctemp42, ctemp43, ctemp44;
+  IFLOAT ctemp45, ctemp46, ctemp47, ctemp48;
+  IFLOAT ctemp49, ctemp50, ctemp51, ctemp52;
+  IFLOAT ctemp53, ctemp54, ctemp55, ctemp56;
+  IFLOAT ctemp57, ctemp58, ctemp59, ctemp60;
+  IFLOAT ctemp61, ctemp62, ctemp63, ctemp64;
 
 
   aoffset = a;
diff --git a/kernel/generic/gemm_tcopy_16.c b/kernel/generic/gemm_tcopy_16.c
index 56268ebf2..14252599a 100644
--- a/kernel/generic/gemm_tcopy_16.c
+++ b/kernel/generic/gemm_tcopy_16.c
@@ -39,22 +39,22 @@
 #include <stdio.h>
 #include "common.h"
 
-int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b){
+int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
 
   BLASLONG i, j;
 
-  FLOAT *aoffset;
-  FLOAT *aoffset1, *aoffset2;
-  FLOAT *boffset;
+  IFLOAT *aoffset;
+  IFLOAT *aoffset1, *aoffset2;
+  IFLOAT *boffset;
 
-  FLOAT ctemp01, ctemp02, ctemp03, ctemp04;
-  FLOAT ctemp05, ctemp06, ctemp07, ctemp08;
-  FLOAT ctemp09, ctemp10, ctemp11, ctemp12;
-  FLOAT ctemp13, ctemp14, ctemp15, ctemp16;
-  FLOAT ctemp17, ctemp18, ctemp19, ctemp20;
-  FLOAT ctemp21, ctemp22, ctemp23, ctemp24;
-  FLOAT ctemp25, ctemp26, ctemp27, ctemp28;
-  FLOAT ctemp29, ctemp30, ctemp31, ctemp32;
+  IFLOAT ctemp01, ctemp02, ctemp03, ctemp04;
+  IFLOAT ctemp05, ctemp06, ctemp07, ctemp08;
+  IFLOAT ctemp09, ctemp10, ctemp11, ctemp12;
+  IFLOAT ctemp13, ctemp14, ctemp15, ctemp16;
+  IFLOAT ctemp17, ctemp18, ctemp19, ctemp20;
+  IFLOAT ctemp21, ctemp22, ctemp23, ctemp24;
+  IFLOAT ctemp25, ctemp26, ctemp27, ctemp28;
+  IFLOAT ctemp29, ctemp30, ctemp31, ctemp32;
 
   aoffset   = a;
   boffset   = b;
diff --git a/kernel/generic/gemm_tcopy_8.c b/kernel/generic/gemm_tcopy_8.c
index b28f3d219..3e8a839db 100644
--- a/kernel/generic/gemm_tcopy_8.c
+++ b/kernel/generic/gemm_tcopy_8.c
@@ -39,32 +39,32 @@
 #include <stdio.h>
 #include "common.h"
 
-int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b){
+int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
 
   BLASLONG i, j;
 
-  FLOAT *aoffset;
-  FLOAT *aoffset1, *aoffset2, *aoffset3, *aoffset4;
-  FLOAT *aoffset5, *aoffset6, *aoffset7, *aoffset8;
+  IFLOAT *aoffset;
+  IFLOAT *aoffset1, *aoffset2, *aoffset3, *aoffset4;
+  IFLOAT *aoffset5, *aoffset6, *aoffset7, *aoffset8;
 
-  FLOAT *boffset,  *boffset1, *boffset2, *boffset3, *boffset4;
+  IFLOAT *boffset,  *boffset1, *boffset2, *boffset3, *boffset4;
 
-  FLOAT ctemp01, ctemp02, ctemp03, ctemp04;
-  FLOAT ctemp05, ctemp06, ctemp07, ctemp08;
-  FLOAT ctemp09, ctemp10, ctemp11, ctemp12;
-  FLOAT ctemp13, ctemp14, ctemp15, ctemp16;
-  FLOAT ctemp17, ctemp18, ctemp19, ctemp20;
-  FLOAT ctemp21, ctemp22, ctemp23, ctemp24;
-  FLOAT ctemp25, ctemp26, ctemp27, ctemp28;
-  FLOAT ctemp29, ctemp30, ctemp31, ctemp32;
-  FLOAT ctemp33, ctemp34, ctemp35, ctemp36;
-  FLOAT ctemp37, ctemp38, ctemp39, ctemp40;
-  FLOAT ctemp41, ctemp42, ctemp43, ctemp44;
-  FLOAT ctemp45, ctemp46, ctemp47, ctemp48;
-  FLOAT ctemp49, ctemp50, ctemp51, ctemp52;
-  FLOAT ctemp53, ctemp54, ctemp55, ctemp56;
-  FLOAT ctemp57, ctemp58, ctemp59, ctemp60;
-  FLOAT ctemp61, ctemp62, ctemp63, ctemp64;
+  IFLOAT ctemp01, ctemp02, ctemp03, ctemp04;
+  IFLOAT ctemp05, ctemp06, ctemp07, ctemp08;
+  IFLOAT ctemp09, ctemp10, ctemp11, ctemp12;
+  IFLOAT ctemp13, ctemp14, ctemp15, ctemp16;
+  IFLOAT ctemp17, ctemp18, ctemp19, ctemp20;
+  IFLOAT ctemp21, ctemp22, ctemp23, ctemp24;
+  IFLOAT ctemp25, ctemp26, ctemp27, ctemp28;
+  IFLOAT ctemp29, ctemp30, ctemp31, ctemp32;
+  IFLOAT ctemp33, ctemp34, ctemp35, ctemp36;
+  IFLOAT ctemp37, ctemp38, ctemp39, ctemp40;
+  IFLOAT ctemp41, ctemp42, ctemp43, ctemp44;
+  IFLOAT ctemp45, ctemp46, ctemp47, ctemp48;
+  IFLOAT ctemp49, ctemp50, ctemp51, ctemp52;
+  IFLOAT ctemp53, ctemp54, ctemp55, ctemp56;
+  IFLOAT ctemp57, ctemp58, ctemp59, ctemp60;
+  IFLOAT ctemp61, ctemp62, ctemp63, ctemp64;
 
   aoffset   = a;
   boffset   = b;
diff --git a/kernel/power/KERNEL.POWER10 b/kernel/power/KERNEL.POWER10
index 4fc7190b0..39f5e9414 100644
--- a/kernel/power/KERNEL.POWER10
+++ b/kernel/power/KERNEL.POWER10
@@ -7,6 +7,17 @@ else
 #CGEMM_BETA = ../generic/zgemm_beta.c
 #ZGEMM_BETA = ../generic/zgemm_beta.c
 
+SHGEMM_BETA = ../generic/gemm_beta.c
+SHGEMMKERNEL    = shgemm_kernel_power10.c
+SHGEMMINCOPY    = ../generic/gemm_ncopy_16.c
+SHGEMMITCOPY    = ../generic/gemm_tcopy_16.c
+SHGEMMONCOPY    = ../generic/gemm_ncopy_8.c
+SHGEMMOTCOPY    = ../generic/gemm_tcopy_8.c
+SHGEMMINCOPYOBJ =  shgemm_incopy$(TSUFFIX).$(SUFFIX)
+SHGEMMITCOPYOBJ =  shgemm_itcopy$(TSUFFIX).$(SUFFIX)
+SHGEMMONCOPYOBJ =  shgemm_oncopy$(TSUFFIX).$(SUFFIX)
+SHGEMMOTCOPYOBJ =  shgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
 STRMMKERNEL	= sgemm_kernel_power10.c
 DTRMMKERNEL	= dgemm_kernel_power10.c
 CTRMMKERNEL	= cgemm_kernel_power10.S
diff --git a/kernel/power/cgemm_kernel_8x4_power8.S b/kernel/power/cgemm_kernel_8x4_power8.S
index 2bc99974f..6be8c128c 100644
--- a/kernel/power/cgemm_kernel_8x4_power8.S
+++ b/kernel/power/cgemm_kernel_8x4_power8.S
@@ -424,7 +424,7 @@ L999:
 	lwz	r16,  204(SP)
 	lwz	r15,  208(SP)
 	lwz	r14,  212(SP)
-        addi    r11, 224
+        addi    r11, SP, 224
 #endif
         lvx     v20, r11, r0
         addi    r11, r11, 16
@@ -459,4 +459,4 @@ L999:
 	blr
 
 	EPILOGUE
-#endif^
+#endif
diff --git a/kernel/power/shgemm_kernel_power10.c b/kernel/power/shgemm_kernel_power10.c
new file mode 100644
index 000000000..7455f925c
--- /dev/null
+++ b/kernel/power/shgemm_kernel_power10.c
@@ -0,0 +1,1044 @@
+/*********************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+**********************************************************************************/
+#include "common.h"
+#include <altivec.h>
+#if defined(HALF) && defined(HALFCONVERSION)
+static float
+bfloat16tof32 (bfloat16 f16)
+{
+  float result = 0;
+  unsigned short *q = (unsigned short *) (&result);
+#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  q[0] = f16;
+#else
+  q[1] = f16;
+#endif
+  return result;
+}
+
+#define BF16TOF32(x) (bfloat16tof32(x))
+#else
+#define BF16TOF32(x) x
+#endif
+
+typedef unsigned char vec_t __attribute__ ((vector_size (16)));
+typedef FLOAT v4sf_t __attribute__ ((vector_size (16)));
+typedef FLOAT v2sf_t __attribute__ ((vector_size (8)));
+
+vector char mask =
+  { 0x0, 0x1, 0x8, 0x9, 0x2, 0x3, 0xa, 0xb, 0x4, 0x5, 0xc, 0xd, 0x6, 0x7, 0xe,
+  0xf
+};
+
+/* 
+ * BFLOAT16 xvbf16ger2pp instruction needs 4×2 matrix of
+ * bfloat16 floating-point values as input. Hence this
+ * merging is needed on A and B matrices. 
+ */
+#define MERGE_ROW(x) vec_perm(x, x, mask)
+#define MERGE_HIGH(x, y) (vec_t) vec_mergeh ((vector short)x, (vector short)y)
+#define MERGE_LOW(x, y) (vec_t) vec_mergel ((vector short)x, (vector short)y)
+
+#define SAVE_ACC(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v4sf_t *) &CO[0* ldc+J]; \
+          rowC[0] += result[3] * alpha; \
+          rowC = (v4sf_t *) &CO[1*ldc+J]; \
+          rowC[0] += result[2] * alpha; \
+          rowC = (v4sf_t *) &CO[2*ldc+J]; \
+          rowC[0] += result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[3*ldc+J]; \
+          rowC[0] += result[0] * alpha;
+#define SAVE_ACC1(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v4sf_t *) &CO[4* ldc+J]; \
+          rowC[0] += result[3] * alpha; \
+          rowC = (v4sf_t *) &CO[5*ldc+J]; \
+          rowC[0] += result[2] * alpha; \
+          rowC = (v4sf_t *) &CO[6*ldc+J]; \
+          rowC[0] += result[1] * alpha; \
+          rowC = (v4sf_t *) &CO[7*ldc+J]; \
+          rowC[0] += result[0] * alpha;
+#define  SAVE4x2_ACC(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v2sf_t *) &CO[0* ldc+J]; \
+          rowC[0] += result[6] * alpha; \
+	  rowC = (v2sf_t *) &CO[1* ldc+J]; \
+          rowC[0] += result[4] * alpha; \
+	  rowC = (v2sf_t *) &CO[2* ldc+J]; \
+          rowC[0] += result[2] * alpha; \
+	  rowC = (v2sf_t *) &CO[3* ldc+J]; \
+          rowC[0] += result[0] * alpha;
+#define  SAVE4x2_ACC1(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v2sf_t *) &CO[4* ldc+J]; \
+          rowC[0] += result[6] * alpha; \
+	  rowC = (v2sf_t *) &CO[5* ldc+J]; \
+          rowC[0] += result[4] * alpha; \
+	  rowC = (v2sf_t *) &CO[6* ldc+J]; \
+          rowC[0] += result[2] * alpha; \
+	  rowC = (v2sf_t *) &CO[7* ldc+J]; \
+          rowC[0] += result[0] * alpha;
+
+#define MMA __builtin_mma_xvbf16ger2pp
+
+#define  SAVE2x4_ACC(ACC, J)  \
+	  __builtin_mma_disassemble_acc (result, ACC); \
+	  rowC = (v4sf_t *) &CO[0* ldc+J]; \
+          rowC[0] += result[3] * alpha; \
+	  rowC = (v4sf_t *) &CO[1* ldc+J]; \
+          rowC[0] += result[2] * alpha;
+
+#define SET_ACC_ZERO4() \
+	  __builtin_mma_xxsetaccz (&acc0); \
+	  __builtin_mma_xxsetaccz (&acc1); \
+	  __builtin_mma_xxsetaccz (&acc2); \
+	  __builtin_mma_xxsetaccz (&acc3);
+
+#define SET_ACC_ZERO8() \
+	  __builtin_mma_xxsetaccz (&acc0); \
+	  __builtin_mma_xxsetaccz (&acc1); \
+	  __builtin_mma_xxsetaccz (&acc2); \
+	  __builtin_mma_xxsetaccz (&acc3); \
+	  __builtin_mma_xxsetaccz (&acc4); \
+	  __builtin_mma_xxsetaccz (&acc5); \
+	  __builtin_mma_xxsetaccz (&acc6); \
+	  __builtin_mma_xxsetaccz (&acc7);
+
+#define PREFETCH1(x, y) asm volatile ("dcbt %0, %1" : : "r" (x), "b" (y) : "memory");
+/*************************************************************************************
+* SHGEMM Kernel
+*************************************************************************************/
+int
+CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A,
+       IFLOAT * B, FLOAT * C, BLASLONG ldc)
+{
+  BLASLONG N = n;
+  BLASLONG i1;
+  v4sf_t valpha = { alpha, alpha, alpha, alpha };
+  vector short vzero = { 0, 0, 0, 0, 0, 0, 0, 0 };
+  N = n >> 3;
+  /* Loop for n >= 8. */
+  for (i1 = 0; i1 < N; i1++)
+    {
+      BLASLONG i, j;
+      FLOAT *CO;
+      IFLOAT *AO;
+      CO = C;
+      C += ldc << 3;
+      AO = A;
+      PREFETCH1 (A, 128);
+      PREFETCH1 (A, 256);
+      i = m >> 4;
+      /* Loop for m >= 16. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1, acc2, acc3, acc4, acc5, acc6, acc7;
+	  SET_ACC_ZERO8 ();
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & (AO[l << 5]);
+	      vec_t *rowB = (vec_t *) & (BO[l << 4]);
+	      vec_t rowB_h = MERGE_HIGH (rowB[0], rowB[1]);
+	      vec_t rowB_l = MERGE_LOW (rowB[0], rowB[1]);
+	      vec_t rowA_h = MERGE_HIGH (rowA[0], rowA[2]);
+	      vec_t rowA_l = MERGE_LOW (rowA[0], rowA[2]);
+	      vec_t rowA2_h = MERGE_HIGH (rowA[1], rowA[3]);
+	      vec_t rowA2_l = MERGE_LOW (rowA[1], rowA[3]);
+	      MMA (&acc0, rowB_h, rowA_h);
+	      MMA (&acc1, rowB_l, rowA_h);
+	      MMA (&acc2, rowB_h, rowA_l);
+	      MMA (&acc3, rowB_l, rowA_l);
+	      MMA (&acc4, rowB_h, rowA2_h);
+	      MMA (&acc5, rowB_l, rowA2_h);
+	      MMA (&acc6, rowB_h, rowA2_l);
+	      MMA (&acc7, rowB_l, rowA2_l);
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 4;
+	      vec_t *rowA = (vec_t *) & (AO[l << 1]);
+	      vec_t *rowB = (vec_t *) & (BO[l]);
+	      vec_t rowB_h = MERGE_HIGH (rowB[0], rowB[1]);
+	      vec_t rowB_l = MERGE_LOW (rowB[0], rowB[1]);
+	      vec_t rowA_h = MERGE_HIGH (rowA[0], vzero);
+	      vec_t rowA_l = MERGE_LOW (rowA[0], vzero);
+	      vec_t rowA2_h = MERGE_HIGH (rowA[1], vzero);
+	      vec_t rowA2_l = MERGE_LOW (rowA[1], vzero);
+	      MMA (&acc0, rowB_h, rowA_h);
+	      MMA (&acc1, rowB_l, rowA_h);
+	      MMA (&acc2, rowB_h, rowA_l);
+	      MMA (&acc3, rowB_l, rowA_l);
+	      MMA (&acc4, rowB_h, rowA2_h);
+	      MMA (&acc5, rowB_l, rowA2_h);
+	      MMA (&acc6, rowB_h, rowA2_l);
+	      MMA (&acc7, rowB_l, rowA2_l);
+	    }
+	  SAVE_ACC (&acc0, 0);
+	  SAVE_ACC (&acc2, 4);
+	  SAVE_ACC1 (&acc1, 0);
+	  SAVE_ACC1 (&acc3, 4);
+	  SAVE_ACC (&acc4, 8);
+	  SAVE_ACC (&acc6, 12);
+	  SAVE_ACC1 (&acc5, 8);
+	  SAVE_ACC1 (&acc7, 12);
+	  CO += 16;
+
+	  AO += (k << 4);
+	  BO += (k << 3);
+	}
+      i = (m & 15) >> 3;
+      /* Loop for m >= 8. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1, acc2, acc3;
+	  SET_ACC_ZERO4 ();
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & (AO[l << 4]);
+	      vec_t *rowB = (vec_t *) & (BO[l << 4]);
+	      vec_t rowB_h = MERGE_HIGH (rowB[0], rowB[1]);
+	      vec_t rowB_l = MERGE_LOW (rowB[0], rowB[1]);
+	      vec_t rowA_h = MERGE_HIGH (rowA[0], rowA[1]);
+	      vec_t rowA_l = MERGE_LOW (rowA[0], rowA[1]);
+	      MMA (&acc0, rowB_h, rowA_h);
+	      MMA (&acc1, rowB_l, rowA_h);
+	      MMA (&acc2, rowB_h, rowA_l);
+	      MMA (&acc3, rowB_l, rowA_l);
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 4;
+	      vec_t *rowA = (vec_t *) & (AO[l]);
+	      vec_t *rowB = (vec_t *) & (BO[l]);
+	      vec_t rowB_h = MERGE_HIGH (rowB[0], rowB[1]);
+	      vec_t rowB_l = MERGE_LOW (rowB[0], rowB[1]);
+	      vec_t rowA_h = MERGE_HIGH (rowA[0], vzero);
+	      vec_t rowA_l = MERGE_LOW (rowA[0], vzero);
+	      MMA (&acc0, rowB_h, rowA_h);
+	      MMA (&acc1, rowB_l, rowA_h);
+	      MMA (&acc2, rowB_h, rowA_l);
+	      MMA (&acc3, rowB_l, rowA_l);
+	    }
+	  SAVE_ACC (&acc0, 0);
+	  SAVE_ACC (&acc2, 4);
+	  SAVE_ACC1 (&acc1, 0);
+	  SAVE_ACC1 (&acc3, 4);
+	  CO += 8;
+	  AO += (k << 3);
+	  BO += (k << 3);
+	}
+      i = (m & 7) >> 2;
+      /* Loop for m >= 4. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  __builtin_mma_xxsetaccz (&acc1);
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & (AO[l << 3]);
+	      vec_t *rowB = (vec_t *) & (BO[l << 4]);
+	      vec_t rowA_mrg = MERGE_ROW (rowA[0]);
+	      MMA (&acc0, MERGE_HIGH (rowB[0], rowB[1]), rowA_mrg);
+	      MMA (&acc1, MERGE_LOW (rowB[0], rowB[1]), rowA_mrg);
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 3;
+	      vector short rowA =
+		{ AO[l + 0], 0, AO[l + 1], 0, AO[l + 2], 0, AO[l + 3], 0 };
+	      vec_t *rowB = (vec_t *) & (BO[l << 1]);
+	      MMA (&acc0, MERGE_HIGH (rowB[0], rowB[1]), (vec_t) rowA);
+	      MMA (&acc1, MERGE_LOW (rowB[0], rowB[1]), (vec_t) rowA);
+	    }
+	  SAVE_ACC (&acc0, 0);
+	  SAVE_ACC1 (&acc1, 0);
+	  CO += 4;
+	  AO += (k << 2);
+	  BO += (k << 3);
+	}
+      i = (m & 3) >> 1;
+      /* Loop for m >= 2. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v2sf_t *rowC;
+	  v2sf_t result[8];
+	  __vector_quad acc0, acc1;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  __builtin_mma_xxsetaccz (&acc1);
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vector short rowA =
+		{ AO[(l << 2) + 0], AO[(l << 2) + 2], AO[(l << 2) + 1],
+		AO[(l << 2) + 3],
+		0, 0, 0, 0
+	      };
+	      vec_t *rowB = (vec_t *) & (BO[l << 4]);
+	      MMA (&acc0, MERGE_HIGH (rowB[0], rowB[1]), (vec_t) rowA);
+	      MMA (&acc1, MERGE_LOW (rowB[0], rowB[1]), (vec_t) rowA);
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 2;
+	      vector short rowA = { AO[l + 0], 0, AO[l + 1], 0, 0, 0, 0, 0 };
+	      vec_t *rowB = (vec_t *) & (BO[(l << 2)]);
+	      MMA (&acc0, MERGE_HIGH (rowB[0], rowB[1]), (vec_t) rowA);
+	      MMA (&acc1, MERGE_LOW (rowB[0], rowB[1]), (vec_t) rowA);
+	    }
+	  SAVE4x2_ACC (&acc0, 0);
+	  SAVE4x2_ACC1 (&acc1, 0);
+	  CO += 2;
+	  AO += (k << 1);
+	  BO += (k << 3);
+	}
+      i = (m & 1) >> 0;
+      /* Loop for m = 1. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0, 0, 0 }
+	  , t1 =
+	  {
+	  0, 0, 0, 0};
+	  for (l = 0; l < k; l++)
+	    {
+	      v4sf_t rowA =
+		{ BF16TOF32 (AO[l]), BF16TOF32 (AO[l]), BF16TOF32 (AO[l]),
+		BF16TOF32 (AO[l])
+	      };
+	      v4sf_t rowB =
+		{ BF16TOF32 (BO[l << 3]), BF16TOF32 (BO[(l << 3) + 1]),
+		BF16TOF32 (BO[(l << 3) + 2]),
+		BF16TOF32 (BO[(l << 3) + 3])
+	      };
+	      v4sf_t rowB1 =
+		{ BF16TOF32 (BO[(l << 3) + 4]), BF16TOF32 (BO[(l << 3) + 5]),
+		BF16TOF32 (BO[(l << 3) + 6]),
+		BF16TOF32 (BO[(l << 3) + 7])
+	      };
+	      t += rowA * rowB;
+	      t1 += rowA * rowB1;
+	    }
+	  t = t * valpha;
+	  t1 = t1 * valpha;
+	  CO[0 * ldc] += t[0];
+	  CO[1 * ldc] += t[1];
+	  CO[2 * ldc] += t[2];
+	  CO[3 * ldc] += t[3];
+	  CO[4 * ldc] += t1[0];
+	  CO[5 * ldc] += t1[1];
+	  CO[6 * ldc] += t1[2];
+	  CO[7 * ldc] += t1[3];
+	  CO += 1;
+	  AO += k;
+	  BO += (k << 3);
+	}
+      B += k << 3;
+    }
+  N = (n & 7) >> 2;
+  /* Loop for n >= 4. */
+  for (i1 = 0; i1 < N; i1++)
+    {
+      BLASLONG i, j;
+      FLOAT *CO;
+      IFLOAT *AO;
+      CO = C;
+      C += ldc << 2;
+      AO = A;
+      i = m >> 5;
+      /* Loop for m >= 32. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  IFLOAT *A1 = AO + (16 * k);
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1, acc2, acc3, acc4, acc5, acc6, acc7;
+	  SET_ACC_ZERO8 ();
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & (AO[l << 5]);
+	      vec_t *rowA1 = (vec_t *) & (A1[l << 5]);
+	      vec_t *rowB = (vec_t *) & (BO[l << 3]);
+	      vec_t rowB_mrg = MERGE_ROW (rowB[0]);
+	      MMA (&acc0, rowB_mrg, MERGE_HIGH (rowA[0], rowA[2]));
+	      MMA (&acc1, rowB_mrg, MERGE_LOW (rowA[0], rowA[2]));
+	      MMA (&acc2, rowB_mrg, MERGE_HIGH (rowA[1], rowA[3]));
+	      MMA (&acc3, rowB_mrg, MERGE_LOW (rowA[1], rowA[3]));
+	      MMA (&acc4, rowB_mrg, MERGE_HIGH (rowA1[0], rowA1[2]));
+	      MMA (&acc5, rowB_mrg, MERGE_LOW (rowA1[0], rowA1[2]));
+	      MMA (&acc6, rowB_mrg, MERGE_HIGH (rowA1[1], rowA1[3]));
+	      MMA (&acc7, rowB_mrg, MERGE_LOW (rowA1[1], rowA1[3]));
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 3;
+	      vec_t *rowA = (vec_t *) & (AO[(l << 2)]);
+	      vec_t *rowA1 = (vec_t *) & (A1[(l << 2)]);
+	      vec_t *rowB = (vec_t *) & (BO[l]);
+	      vec_t rowB_mrg = MERGE_ROW (rowB[0]);
+	      MMA (&acc0, rowB_mrg, MERGE_HIGH (rowA[0], vzero));
+	      MMA (&acc1, rowB_mrg, MERGE_LOW (rowA[0], vzero));
+	      MMA (&acc2, rowB_mrg, MERGE_HIGH (rowA[1], vzero));
+	      MMA (&acc3, rowB_mrg, MERGE_LOW (rowA[1], vzero));
+	      MMA (&acc4, rowB_mrg, MERGE_HIGH (rowA1[0], vzero));
+	      MMA (&acc5, rowB_mrg, MERGE_LOW (rowA1[0], vzero));
+	      MMA (&acc6, rowB_mrg, MERGE_HIGH (rowA1[1], vzero));
+	      MMA (&acc7, rowB_mrg, MERGE_LOW (rowA1[1], vzero));
+	    }
+
+	  SAVE_ACC (&acc0, 0);
+	  SAVE_ACC (&acc1, 4);
+	  CO += 8;
+	  SAVE_ACC (&acc2, 0);
+	  SAVE_ACC (&acc3, 4);
+	  CO += 8;
+	  SAVE_ACC (&acc4, 0);
+	  SAVE_ACC (&acc5, 4);
+	  CO += 8;
+	  SAVE_ACC (&acc6, 0);
+	  SAVE_ACC (&acc7, 4);
+	  CO += 8;
+	  AO += k << 5;
+	  BO += k << 2;
+	}
+      i = (m & 31) >> 4;
+      /* Loop for m >= 16. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1, acc2, acc3;
+	  SET_ACC_ZERO4 ();
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & (AO[l << 5]);
+	      vec_t *rowB = (vec_t *) & (BO[l << 3]);
+	      vec_t rowB_mrg = MERGE_ROW (rowB[0]);
+	      MMA (&acc0, rowB_mrg, MERGE_HIGH (rowA[0], rowA[2]));
+	      MMA (&acc1, rowB_mrg, MERGE_LOW (rowA[0], rowA[2]));
+	      MMA (&acc2, rowB_mrg, MERGE_HIGH (rowA[1], rowA[3]));
+	      MMA (&acc3, rowB_mrg, MERGE_LOW (rowA[1], rowA[3]));
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 3;
+	      vec_t *rowA = (vec_t *) & (AO[(l << 2)]);
+	      vec_t *rowB = (vec_t *) & (BO[l]);
+	      vec_t rowB_mrg = MERGE_ROW (rowB[0]);
+	      MMA (&acc0, rowB_mrg, MERGE_HIGH (rowA[0], vzero));
+	      MMA (&acc1, rowB_mrg, MERGE_LOW (rowA[0], vzero));
+	      MMA (&acc2, rowB_mrg, MERGE_HIGH (rowA[1], vzero));
+	      MMA (&acc3, rowB_mrg, MERGE_LOW (rowA[1], vzero));
+	    }
+
+	  SAVE_ACC (&acc0, 0);
+	  SAVE_ACC (&acc1, 4);
+	  CO += 8;
+	  SAVE_ACC (&acc2, 0);
+	  SAVE_ACC (&acc3, 4);
+	  CO += 8;
+	  AO += k << 4;
+	  BO += k << 2;
+	}
+      i = (m & 15) >> 3;
+      /* Loop for m >= 8. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  __builtin_mma_xxsetaccz (&acc1);
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & (AO[l << 4]);
+	      vec_t *rowB = (vec_t *) & (BO[l << 3]);
+	      vec_t rowB_mrg = MERGE_ROW (rowB[0]);
+	      MMA (&acc0, rowB_mrg, MERGE_HIGH (rowA[0], rowA[1]));
+	      MMA (&acc1, rowB_mrg, MERGE_LOW (rowA[0], rowA[1]));
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 3;
+	      vec_t *rowA = (vec_t *) & (AO[l << 1]);
+	      vec_t *rowB = (vec_t *) & (BO[l]);
+	      vec_t rowB_mrg = MERGE_ROW (rowB[0]);
+	      MMA (&acc0, rowB_mrg, MERGE_HIGH (rowA[0], vzero));
+	      MMA (&acc1, rowB_mrg, MERGE_LOW (rowA[0], vzero));
+	    }
+	  SAVE_ACC (&acc0, 0);
+	  SAVE_ACC (&acc1, 4);
+	  CO += 8;
+	  AO += k << 3;
+	  BO += k << 2;
+	}
+      i = (m & 7) >> 2;
+      /* Loop for m >= 4. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v4sf_t *rowC;
+	  __vector_quad acc0;
+	  v4sf_t result[4];
+	  BLASLONG l = 0;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vec_t *rowA = (vec_t *) & (AO[l << 3]);
+	      vec_t *rowB = (vec_t *) & (BO[l << 3]);
+	      MMA (&acc0, MERGE_ROW (rowB[0]), MERGE_ROW (rowA[0]));
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 3;
+	      vector short rowA =
+		{ AO[l], 0, AO[l + 1], 0, AO[l + 2], 0, AO[l + 3], 0 };
+	      vec_t *rowB = (vec_t *) & (BO[l]);
+	      MMA (&acc0, MERGE_ROW (rowB[0]), (vec_t) rowA);
+	    }
+	  SAVE_ACC (&acc0, 0);
+	  CO += 4;
+	  AO += k << 2;
+	  BO += k << 2;
+	}
+      i = (m & 3) >> 1;
+      /* Loop for m >= 2. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v2sf_t *rowC;
+	  v2sf_t result[8];
+	  __vector_quad acc0;
+	  BLASLONG l = 0;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vector short rowA =
+		{ AO[(l << 2) + 0], AO[(l << 2) + 2], AO[(l << 2) + 1],
+		AO[(l << 2) + 3],
+		0, 0, 0, 0
+	      };
+	      vec_t *rowB = (vec_t *) & (BO[l << 3]);
+	      MMA (&acc0, MERGE_ROW (rowB[0]), (vec_t) rowA);
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 2;
+	      vector short rowA = { AO[l], 0, AO[l + 1], 0, 0, 0, 0, 0 };
+	      vec_t *rowB = (vec_t *) & (BO[l << 1]);
+	      MMA (&acc0, MERGE_ROW (rowB[0]), (vec_t) rowA);
+	    }
+	  SAVE4x2_ACC (&acc0, 0);
+	  CO += 2;
+	  AO += k << 1;
+	  BO += k << 2;
+	}
+      i = (m & 1) >> 0;
+      /* Loop for m = 1. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0, 0, 0 };
+	  for (l = 0; l < k; l++)
+	    {
+	      v4sf_t rowA =
+		{ BF16TOF32 (AO[l]), BF16TOF32 (AO[l]), BF16TOF32 (AO[l]),
+		BF16TOF32 (AO[l])
+	      };
+	      v4sf_t rowB =
+		{ BF16TOF32 (BO[l << 2]), BF16TOF32 (BO[(l << 2) + 1]),
+		BF16TOF32 (BO[(l << 2) + 2]),
+		BF16TOF32 (BO[(l << 2) + 3])
+	      };
+	      t += rowA * rowB;
+	    }
+	  t = t * valpha;
+	  CO[0 * ldc] += t[0];
+	  CO[1 * ldc] += t[1];
+	  CO[2 * ldc] += t[2];
+	  CO[3 * ldc] += t[3];
+	  AO += k;
+	  BO += (k << 2);
+	  CO += 1;
+	}
+
+      B += k << 2;
+    }
+  N = (n & 3) >> 1;
+  /* Loop for n >= 2. */
+  for (i1 = 0; i1 < N; i1++)
+    {
+      BLASLONG i, j;
+      FLOAT *CO;
+      IFLOAT *AO;
+      CO = C;
+      C += ldc << 1;
+      AO = A;
+      i = m >> 5;
+      /* Loop for m >= 32. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  IFLOAT *A1 = AO + (16 * k);
+	  __vector_quad acc0, acc1, acc2, acc3, acc4, acc5, acc6, acc7;
+	  SET_ACC_ZERO8 ();
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vector short rowB =
+		{ BO[(l << 2) + 0], BO[(l << 2) + 2], BO[(l << 2) + 1],
+		BO[(l << 2) + 3],
+		0, 0, 0, 0
+	      };
+	      vec_t *rowA = (vec_t *) & (AO[l << 5]);
+	      vec_t *rowA1 = (vec_t *) & (A1[l << 5]);
+	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], rowA[2]));
+	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], rowA[2]));
+	      MMA (&acc2, (vec_t) rowB, MERGE_HIGH (rowA[1], rowA[3]));
+	      MMA (&acc3, (vec_t) rowB, MERGE_LOW (rowA[1], rowA[3]));
+	      MMA (&acc4, (vec_t) rowB, MERGE_HIGH (rowA1[0], rowA1[2]));
+	      MMA (&acc5, (vec_t) rowB, MERGE_LOW (rowA1[0], rowA1[2]));
+	      MMA (&acc6, (vec_t) rowB, MERGE_HIGH (rowA1[1], rowA1[3]));
+	      MMA (&acc7, (vec_t) rowB, MERGE_LOW (rowA1[1], rowA1[3]));
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 2;
+	      vector short rowB = { BO[l + 0], 0, BO[l + 1], 0, 0, 0, 0, 0 };
+	      vec_t *rowA = (vec_t *) & (AO[l << 3]);
+	      vec_t *rowA1 = (vec_t *) & (A1[l << 3]);
+	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], rowA[2]));
+	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], rowA[2]));
+	      MMA (&acc2, (vec_t) rowB, MERGE_HIGH (rowA[1], rowA[3]));
+	      MMA (&acc3, (vec_t) rowB, MERGE_LOW (rowA[1], rowA[3]));
+	      MMA (&acc4, (vec_t) rowB, MERGE_HIGH (rowA1[0], rowA1[2]));
+	      MMA (&acc5, (vec_t) rowB, MERGE_LOW (rowA1[0], rowA1[2]));
+	      MMA (&acc6, (vec_t) rowB, MERGE_HIGH (rowA1[1], rowA1[3]));
+	      MMA (&acc7, (vec_t) rowB, MERGE_LOW (rowA1[1], rowA1[3]));
+	    }
+	  SAVE2x4_ACC (&acc0, 0);
+	  SAVE2x4_ACC (&acc1, 4);
+	  SAVE2x4_ACC (&acc2, 8);
+	  SAVE2x4_ACC (&acc3, 12);
+	  CO += 16;
+	  SAVE2x4_ACC (&acc4, 0);
+	  SAVE2x4_ACC (&acc5, 4);
+	  SAVE2x4_ACC (&acc6, 8);
+	  SAVE2x4_ACC (&acc7, 12);
+	  CO += 16;
+	  AO += k << 5;
+	  BO += k << 1;
+	}
+      i = (m & 31) >> 4;
+      /* Loop for m >= 16. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1, acc2, acc3;
+	  SET_ACC_ZERO4 ();
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vector short rowB =
+		{ BO[(l << 2) + 0], BO[(l << 2) + 2], BO[(l << 2) + 1],
+		BO[(l << 2) + 3],
+		0, 0, 0, 0
+	      };
+	      vec_t *rowA = (vec_t *) & (AO[l << 5]);
+	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], rowA[2]));
+	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], rowA[2]));
+	      MMA (&acc2, (vec_t) rowB, MERGE_HIGH (rowA[1], rowA[3]));
+	      MMA (&acc3, (vec_t) rowB, MERGE_LOW (rowA[1], rowA[3]));
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 2;
+	      vector short rowB = { BO[l + 0], 0, BO[l + 1], 0, 0, 0, 0, 0 };
+	      vec_t *rowA = (vec_t *) & (AO[l << 3]);
+	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], rowA[2]));
+	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], rowA[2]));
+	      MMA (&acc2, (vec_t) rowB, MERGE_HIGH (rowA[1], rowA[3]));
+	      MMA (&acc3, (vec_t) rowB, MERGE_LOW (rowA[1], rowA[3]));
+	    }
+	  SAVE2x4_ACC (&acc0, 0);
+	  SAVE2x4_ACC (&acc1, 4);
+	  SAVE2x4_ACC (&acc2, 8);
+	  SAVE2x4_ACC (&acc3, 12);
+	  CO += 16;
+	  AO += k << 4;
+	  BO += k << 1;
+	}
+      i = (m & 15) >> 3;
+      /* Loop for m >= 8. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0, acc1;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  __builtin_mma_xxsetaccz (&acc1);
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vector short rowB =
+		{ BO[(l << 2) + 0], BO[(l << 2) + 2], BO[(l << 2) + 1],
+		BO[(l << 2) + 3],
+		0, 0, 0, 0
+	      };
+	      vec_t *rowA = (vec_t *) & (AO[l << 4]);
+	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], rowA[1]));
+	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], rowA[1]));
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 2;
+	      vector short rowB = { BO[l + 0], 0, BO[l + 1], 0, 0, 0, 0, 0 };
+	      vec_t *rowA = (vec_t *) & (AO[(l << 2)]);
+	      MMA (&acc0, (vec_t) rowB, MERGE_HIGH (rowA[0], rowA[1]));
+	      MMA (&acc1, (vec_t) rowB, MERGE_LOW (rowA[0], rowA[1]));
+	    }
+	  SAVE2x4_ACC (&acc0, 0);
+	  SAVE2x4_ACC (&acc1, 4);
+	  CO += 8;
+	  AO += k << 3;
+	  BO += k << 1;
+	}
+      i = (m & 7) >> 2;
+      /* Loop for m >= 4. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  v4sf_t *rowC;
+	  v4sf_t result[4];
+	  __vector_quad acc0;
+	  __builtin_mma_xxsetaccz (&acc0);
+	  BLASLONG l = 0;
+	  for (l = 0; l < k / 2; l++)
+	    {
+	      vector short rowB =
+		{ BO[(l << 2) + 0], BO[(l << 2) + 2], BO[(l << 2) + 1],
+		BO[(l << 2) + 3],
+		0, 0, 0, 0
+	      };
+	      vec_t *rowA = (vec_t *) & (AO[l << 3]);
+	      MMA (&acc0, (vec_t) rowB, MERGE_ROW (rowA[0]));
+	    }
+	  if (k % 2 == 1)
+	    {
+	      if (k > 1)
+		l = (k / 2) << 2;
+	      vector short rowB = { BO[l + 0], 0, BO[l + 1], 0, 0, 0, 0, 0 };
+	      vec_t *rowA = (vec_t *) & (AO[l << 1]);
+	      MMA (&acc0, (vec_t) rowB, MERGE_ROW (rowA[0]));
+	    }
+	  SAVE2x4_ACC (&acc0, 0);
+	  CO += 4;
+	  AO += k << 2;
+	  BO += k << 1;
+	}
+      i = (m & 3) >> 1;
+      /* Loop for m >= 2. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0, 0, 0 };
+	  for (l = 0; l < (k << 1); l += 2)
+	    {
+	      v4sf_t rowA =
+		{ BF16TOF32 (AO[l]), BF16TOF32 (AO[l]), BF16TOF32 (AO[l + 1]),
+		BF16TOF32 (AO[l + 1])
+	      };
+	      v4sf_t rowB =
+		{ BF16TOF32 (BO[l]), BF16TOF32 (BO[l + 1]), BF16TOF32 (BO[l]),
+		BF16TOF32 (BO[l + 1])
+	      };
+	      t += rowA * rowB;
+	    }
+	  t = t * valpha;
+	  CO[0 * ldc] += t[0];
+	  CO[1 * ldc] += t[1];
+	  CO[0 * ldc + 1] += t[2];
+	  CO[1 * ldc + 1] += t[3];
+	  CO += 2;
+	  AO += k << 1;
+	  BO += k << 1;
+	}
+      i = (m & 1) >> 0;
+      /* Loop for m = 1. */
+      for (j = 0; j < i; j++)
+	{
+	  IFLOAT *BO = B;
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0, 0, 0 };
+	  for (l = 0; l < k; l++)
+	    {
+	      v4sf_t rowA = { BF16TOF32 (AO[l]), BF16TOF32 (AO[l]), 0, 0 };
+	      v4sf_t rowB =
+		{ BF16TOF32 (BO[l << 1]), BF16TOF32 (BO[(l << 1) + 1]), 0,
+		0
+	      };
+	      t += rowA * rowB;
+	    }
+	  CO[0 * ldc] += t[0] * alpha;
+	  CO[1 * ldc] += t[1] * alpha;
+	  CO += 1;
+	  AO += k;
+	  BO += k << 1;
+	}
+      B += k << 1;
+    }
+  N = (n & 1) >> 0;
+  /* Loop for n = 1. */
+  for (i1 = 0; i1 < N; i1++)
+    {
+      BLASLONG i;
+      FLOAT *CO;
+      IFLOAT *AO;
+      CO = C;
+      C += ldc;
+      AO = A;
+      i = m;
+      /* Loop for m >= 16. */
+      while (i >= 16)
+	{
+	  IFLOAT *BO = B;
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0, 0, 0 };
+	  v4sf_t t1 = { 0, 0, 0, 0 };
+	  v4sf_t t2 = { 0, 0, 0, 0 };
+	  v4sf_t t3 = { 0, 0, 0, 0 };
+	  for (l = 0; l < k; l++)
+	    {
+	      v4sf_t rowB =
+		{ BF16TOF32 (BO[l]), BF16TOF32 (BO[l]), BF16TOF32 (BO[l]),
+		BF16TOF32 (BO[l])
+	      };
+	      v4sf_t rowA =
+		{ BF16TOF32 (AO[l << 4]), BF16TOF32 (AO[(l << 4) + 1]),
+		BF16TOF32 (AO[(l << 4) + 2]),
+		BF16TOF32 (AO[(l << 4) + 3])
+	      };
+	      v4sf_t rowA1 =
+		{ BF16TOF32 (AO[(l << 4) + 4]), BF16TOF32 (AO[(l << 4) + 5]),
+		BF16TOF32 (AO[(l << 4) + 6]),
+		BF16TOF32 (AO[(l << 4) + 7])
+	      };
+	      v4sf_t rowA2 =
+		{ BF16TOF32 (AO[(l << 4) + 8]), BF16TOF32 (AO[(l << 4) + 9]),
+		BF16TOF32 (AO[(l << 4) + 10]),
+		BF16TOF32 (AO[(l << 4) + 11])
+	      };
+	      v4sf_t rowA3 = { BF16TOF32 (AO[(l << 4) + 12]),
+		BF16TOF32 (AO[(l << 4) + 13]), BF16TOF32 (AO[(l << 4) + 14]),
+		BF16TOF32 (AO[(l << 4) + 15])
+	      };
+	      t += rowA * rowB;
+	      t1 += rowA1 * rowB;
+	      t2 += rowA2 * rowB;
+	      t3 += rowA3 * rowB;
+	    }
+	  t = t * valpha;
+	  t1 = t1 * valpha;
+	  t2 = t2 * valpha;
+	  t3 = t3 * valpha;
+	  CO[0] += t[0];
+	  CO[1] += t[1];
+	  CO[2] += t[2];
+	  CO[3] += t[3];
+	  CO[4] += t1[0];
+	  CO[5] += t1[1];
+	  CO[6] += t1[2];
+	  CO[7] += t1[3];
+	  CO[8] += t2[0];
+	  CO[9] += t2[1];
+	  CO[10] += t2[2];
+	  CO[11] += t2[3];
+	  CO[12] += t3[0];
+	  CO[13] += t3[1];
+	  CO[14] += t3[2];
+	  CO[15] += t3[3];
+	  AO += k << 4;
+	  BO += k;
+	  CO += 16;
+	  i -= 16;
+	}
+      /* Loop for m >= 8. */
+      while (i >= 8)
+	{
+	  IFLOAT *BO = B;
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0, 0, 0 };
+	  v4sf_t t1 = { 0, 0, 0, 0 };
+	  for (l = 0; l < k; l++)
+	    {
+	      v4sf_t rowB =
+		{ BF16TOF32 (BO[l]), BF16TOF32 (BO[l]), BF16TOF32 (BO[l]),
+		BF16TOF32 (BO[l])
+	      };
+	      v4sf_t rowA =
+		{ BF16TOF32 (AO[l << 3]), BF16TOF32 (AO[(l << 3) + 1]),
+		BF16TOF32 (AO[(l << 3) + 2]),
+		BF16TOF32 (AO[(l << 3) + 3])
+	      };
+	      v4sf_t rowA1 =
+		{ BF16TOF32 (AO[(l << 3) + 4]), BF16TOF32 (AO[(l << 3) + 5]),
+		BF16TOF32 (AO[(l << 3) + 6]),
+		BF16TOF32 (AO[(l << 3) + 7])
+	      };
+	      t += rowA * rowB;
+	      t1 += rowA1 * rowB;
+	    }
+	  t = t * valpha;
+	  t1 = t1 * valpha;
+	  CO[0] += t[0];
+	  CO[1] += t[1];
+	  CO[2] += t[2];
+	  CO[3] += t[3];
+	  CO[4] += t1[0];
+	  CO[5] += t1[1];
+	  CO[6] += t1[2];
+	  CO[7] += t1[3];
+	  AO += k << 3;
+	  BO += k;
+	  CO += 8;
+	  i -= 8;
+	}
+      /* Loop for m >= 4. */
+      while (i >= 4)
+	{
+	  IFLOAT *BO = B;
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0, 0, 0 };
+	  for (l = 0; l < k; l++)
+	    {
+	      v4sf_t rowB =
+		{ BF16TOF32 (BO[l]), BF16TOF32 (BO[l]), BF16TOF32 (BO[l]),
+		BF16TOF32 (BO[l])
+	      };
+	      v4sf_t rowA =
+		{ BF16TOF32 (AO[l << 2]), BF16TOF32 (AO[(l << 2) + 1]),
+		BF16TOF32 (AO[(l << 2) + 2]),
+		BF16TOF32 (AO[(l << 2) + 3])
+	      };
+	      t += rowA * rowB;
+	    }
+	  t = t * valpha;
+	  CO[0] += t[0];
+	  CO[1] += t[1];
+	  CO[2] += t[2];
+	  CO[3] += t[3];
+	  AO += k << 2;
+	  BO += k;
+	  CO += 4;
+	  i -= 4;
+	}
+      /* Loop for m >= 2. */
+      while (i >= 2)
+	{
+	  IFLOAT *BO = B;
+	  BLASLONG l = 0;
+	  v4sf_t t = { 0, 0, 0, 0 };
+	  for (l = 0; l < k; l++)
+	    {
+	      v4sf_t rowB = { BF16TOF32 (BO[l]), BF16TOF32 (BO[l]), 0, 0 };
+	      v4sf_t rowA =
+		{ BF16TOF32 (AO[l << 1]), BF16TOF32 (AO[(l << 1) + 1]), 0,
+		0
+	      };
+	      t += rowA * rowB;
+	    }
+	  t = t * valpha;
+	  CO[0] += t[0];
+	  CO[1] += t[1];
+	  AO += k << 1;
+	  BO += k;
+	  CO += 2;
+	  i -= 2;
+	}
+      /* Loop for m = 1. */
+      while (i >= 1)
+	{
+	  IFLOAT *BO = B;
+	  BLASLONG l = 0;
+	  FLOAT t = 0;
+	  for (l = 0; l < k; l++)
+	    {
+	      t += BF16TOF32 (AO[l]) * BF16TOF32 (BO[l]);
+	    }
+	  AO += k;
+	  BO += k;
+	  CO[0] += t * alpha;
+	  CO += 1;
+	  i -= 1;
+	}
+
+      B += k;
+    }
+
+  return 0;
+}
diff --git a/kernel/x86_64/sgemm_kernel_8x4_haswell_2.c b/kernel/x86_64/sgemm_kernel_8x4_haswell_2.c
index 5ab3e6d1f..a2e78c58d 100644
--- a/kernel/x86_64/sgemm_kernel_8x4_haswell_2.c
+++ b/kernel/x86_64/sgemm_kernel_8x4_haswell_2.c
@@ -1,4 +1,4 @@
-/* %0 = "+r"(a_pointer), %1 = "+r"(b_pointer), %2 = "+r"(c_pointer), %3 = "+r"(ldc_in_bytes), %4 for k_count, %5 for c_store, %6 = b_pref */
+/* %0 = "+r"(a_pointer), %1 = "+r"(b_pointer), %2 = "+r"(c_pointer), %3 = "+r"(ldc_in_bytes), %4 for k_count, %5 for c_store, %6 = b_pref */
 /* r10 = tmp, r11 = m_counter, r12 = k << 2(const), r13 = tmp, r14 = b_head_pos(const), r15 = tmp */
 
 /* m = 8 *//* ymm0 for alpha, ymm1-ymm3 for temporary use, ymm4-ymm15 for accumulators */
diff --git a/kernel/x86_64/strsm_kernel_8x4_haswell_LN.c b/kernel/x86_64/strsm_kernel_8x4_haswell_LN.c
index 4131debb1..5410bd4ae 100644
--- a/kernel/x86_64/strsm_kernel_8x4_haswell_LN.c
+++ b/kernel/x86_64/strsm_kernel_8x4_haswell_LN.c
@@ -1,4 +1,4 @@
-#include "common.h"
+#include "common.h"
 #include <stdint.h>
 #include "strsm_kernel_8x4_haswell_L_common.h"
 
diff --git a/kernel/x86_64/strsm_kernel_8x4_haswell_L_common.h b/kernel/x86_64/strsm_kernel_8x4_haswell_L_common.h
index cfa56da97..2862a5b8d 100644
--- a/kernel/x86_64/strsm_kernel_8x4_haswell_L_common.h
+++ b/kernel/x86_64/strsm_kernel_8x4_haswell_L_common.h
@@ -1,4 +1,4 @@
-/* r11 = m_counter, r12 = size_of_k_elements, r13 = kk, r14 = b_head, r15 = a_head */
+/* r11 = m_counter, r12 = size_of_k_elements, r13 = kk, r14 = b_head, r15 = a_head */
 /* register i/o: %0 = a_ptr, %1 = b_ptr, %2 = c_ptr, %3 = c_tmp, %4 = ldc, %5 = k_counter */
 /* memory input: %6 = K, %7 = offset, %8 = {1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0}, %9 = {0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0}, %10 = M */
 
diff --git a/param.h b/param.h
index fd0ea7599..e8cf53f0a 100644
--- a/param.h
+++ b/param.h
@@ -2297,6 +2297,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #endif
 
+#if defined(POWER10)
+#undef SHGEMM_DEFAULT_UNROLL_N
+#undef SHGEMM_DEFAULT_UNROLL_M
+#undef SHGEMM_DEFAULT_P
+#undef SHGEMM_DEFAULT_R
+#undef SHGEMM_DEFAULT_Q
+#define SHGEMM_DEFAULT_UNROLL_M 16
+#define SHGEMM_DEFAULT_UNROLL_N 8
+#define SHGEMM_DEFAULT_P 832
+#define SHGEMM_DEFAULT_Q 1026
+#define SHGEMM_DEFAULT_R 4096
+#endif
+
 #if defined(SPARC) && defined(V7)
 
 #define SNUMOPT		4