Support ARM softfp ABI for sgemm on ARMV7.

make ARM_SOFTFP_ABI=1

diff --git a/Makefile.arm b/Makefile.arm
index 11f8092..c189b0c 100644 (file)
--- a/Makefile.arm
+++ b/Makefile.arm
@@ -1,4 +1,4 @@
-ifeq logical or
+#ifeq logical or
 ifeq ($(CORE), $(filter $(CORE),CORTEXA9 CORTEXA15))
 ifeq ($(OSNAME), Android)
 CCOMMON_OPT += -mfpu=neon  -march=armv7-a
@@ -11,12 +11,12 @@ endif
 
 ifeq ($(CORE), ARMV7)
 ifeq ($(OSNAME), Android)
-ifeq ($(ARM_SOFTFP), 1)
+ifeq ($(ARM_SOFTFP_ABI), 1)
 CCOMMON_OPT += -mfpu=neon -march=armv7-a
 FCOMMON_OPT += -mfpu=neon -march=armv7-a
 else
-CCOMMON_OPT += -marm -mfpu=neon  -mfloat-abi=hard -march=armv7-a -Wl,--no-warn-mismatch
-FCOMMON_OPT += -marm -mfpu=neon  -mfloat-abi=hard -march=armv7-a -Wl,--no-warn-mismatch
+CCOMMON_OPT += -mfpu=neon -march=armv7-a -Wl,--no-warn-mismatch
+FCOMMON_OPT += -mfpu=neon -march=armv7-a -Wl,--no-warn-mismatch
 endif
 else
 CCOMMON_OPT += -mfpu=vfpv3 -march=armv7-a

diff --git a/Makefile.system b/Makefile.system
index c4fcc01..cc78629 100644 (file)
--- a/Makefile.system
+++ b/Makefile.system
@@ -480,9 +480,9 @@ BINARY_DEFINED  = 1
 CCOMMON_OPT += -marm
 FCOMMON_OPT += -marm
 
-ifeq ($(ARM_SOFT_FLOAT_ABI), 1)
-CCOMMON_OPT += -mfloat-abi=softfp
-FCOMMON_OPT += -mfloat-abi=softfp
+ifeq ($(ARM_SOFTFP_ABI), 1)
+CCOMMON_OPT += -mfloat-abi=softfp -DARM_SOFTFP_ABI
+FCOMMON_OPT += -mfloat-abi=softfp -DARM_SOFTFP_ABI
 else
 CCOMMON_OPT += -mfloat-abi=hard
 FCOMMON_OPT += -mfloat-abi=hard

diff --git a/kernel/arm/sgemm_kernel_4x4_vfpv3.S b/kernel/arm/sgemm_kernel_4x4_vfpv3.S
index 1852726..86198ac 100644 (file)
--- a/kernel/arm/sgemm_kernel_4x4_vfpv3.S
+++ b/kernel/arm/sgemm_kernel_4x4_vfpv3.S
@@ -58,8 +58,14 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define        OLD_M   r0
 #define        OLD_N   r1
 #define        OLD_K   r2
+
+#ifdef ARM_SOFTFP_ABI
+#define OLD_ALPHA r3
+//#define OLD_A        
+#else //hard
 #define        OLD_A   r3
 #define OLD_ALPHA s0
+#endif
 
 /******************************************************
 * [fp, #-128] - [fp, #-64] is reserved
@@ -71,7 +77,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define M      [fp, #-256 ]
 #define N      [fp, #-260 ]
 #define K      [fp, #-264 ]
+
+#ifndef ARM_SOFTFP_ABI
 #define A      [fp, #-268 ]
+#endif
 
 #define FP_ZERO [fp, #-240]
 #define FP_ZERO_0 [fp, #-240]
@@ -79,10 +88,17 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define ALPHA   [fp, #-280]
 
+#ifdef ARM_SOFTFP_ABI
+#define A      [fp, #4 ]
+#define B      [fp, #8 ]
+#define C      [fp, #12 ]
+#define OLD_LDC        [fp, #16 ]
+#else //hard
 #define B      [fp, #4 ]
 #define C      [fp, #8 ]
 #define OLD_LDC        [fp, #12 ]
-
+#endif
+       
 #define I      r0
 #define J      r1
 #define L      r2
@@ -854,9 +870,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
        str     OLD_M, M
        str     OLD_N, N
        str     OLD_K, K
+       
+#ifdef ARM_SOFTFP_ABI
+       str     OLD_ALPHA, ALPHA
+#else //hard
        str     OLD_A, A
        vstr    OLD_ALPHA, ALPHA
-
+#endif
        sub     r3, fp, #128
        vstm    r3, { s8 - s31}                                 // store floating point registers