Merge branch 'arm_soft_fp_abi' into develop

Makefile.arm

@@ -1,27 +1,32 @@
-# ifeq logical or
+#ifeq logical or
 ifeq ($(CORE), $(filter $(CORE),CORTEXA9 CORTEXA15))
 ifeq ($(OSNAME), Android)
-CCOMMON_OPT += -marm -mfpu=neon  -mfloat-abi=hard -march=armv7-a
-FCOMMON_OPT += -marm -mfpu=neon  -mfloat-abi=hard -march=armv7-a
+CCOMMON_OPT += -mfpu=neon  -march=armv7-a
+FCOMMON_OPT += -mfpu=neon  -march=armv7-a
 else
-CCOMMON_OPT += -marm -mfpu=vfpv3  -mfloat-abi=hard -march=armv7-a
-FCOMMON_OPT += -marm -mfpu=vfpv3  -mfloat-abi=hard -march=armv7-a
+CCOMMON_OPT += -mfpu=vfpv3 -march=armv7-a
+FCOMMON_OPT += -mfpu=vfpv3 -march=armv7-a
 endif
 endif
 
 ifeq ($(CORE), ARMV7)
 ifeq ($(OSNAME), Android)
-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
+ifeq ($(ARM_SOFTFP_ABI), 1)
+CCOMMON_OPT += -mfpu=neon -march=armv7-a
+FCOMMON_OPT += -mfpu=neon -march=armv7-a
 else
-CCOMMON_OPT += -marm -mfpu=vfpv3  -mfloat-abi=hard -march=armv7-a
-FCOMMON_OPT += -marm -mfpu=vfpv3  -mfloat-abi=hard -march=armv7-a
+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
+FCOMMON_OPT += -mfpu=vfpv3 -march=armv7-a
 endif
 endif
 
 ifeq ($(CORE), ARMV6)
-CCOMMON_OPT += -marm -mfpu=vfp -mfloat-abi=hard  -march=armv6
-FCOMMON_OPT += -marm -mfpu=vfp -mfloat-abi=hard  -march=armv6
+CCOMMON_OPT += -mfpu=vfp -march=armv6
+FCOMMON_OPT += -mfpu=vfp -march=armv6
 endif
 
 

Makefile.system

@@ -482,6 +482,17 @@ endif
 ifeq ($(ARCH), arm)
 NO_BINARY_MODE  = 1
 BINARY_DEFINED  = 1
+
+CCOMMON_OPT += -marm
+FCOMMON_OPT += -marm
+
+ifeq ($(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
+endif
 endif
 
 ifeq ($(ARCH), arm64)

common_arm.h

@@ -111,6 +111,11 @@ static inline int blas_quickdivide(blasint x, blasint y){
 
 #define PROFCODE
 
+#ifdef __ARM_PCS
+//-mfloat-abi=softfp
+#define SOFT_FLOAT_ABI
+#endif
+
 #endif
 
 

kernel/arm/axpy_vfp.S

@@ -38,11 +38,18 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define STACKSIZE 256
 
+#ifndef ARM_SOFTFP_ABI
+//hard	
 #define	OLD_INC_X	[fp, #0 ]
 #define	OLD_Y		[fp, #4 ]
 #define	OLD_INC_Y	[fp, #8 ]
-
-
+#else
+#define	OLD_X		[fp, #0 ]
+#define	OLD_INC_X	[fp, #4 ]
+#define	OLD_Y		[fp, #8 ]
+#define	OLD_INC_Y	[fp, #12 ]
+#endif
+
 #define	N	r0
 #define Y	r1
 #define	INC_X	r2
@@ -363,6 +370,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
         add     fp, sp, #8
 	sub     sp, sp, #STACKSIZE                              // reserve stack
 
+#ifdef ARM_SOFTFP_ABI
+#ifndef DOUBLE
+	vmov	s0, r3     //move alpha to s0
+	ldr	X, OLD_X
+#endif
+#endif
+
 	ldr    INC_X , OLD_INC_X
 	ldr         Y, OLD_Y
 	ldr    INC_Y , OLD_INC_Y

kernel/arm/gemv_n_vfpv3.S

@@ -38,12 +38,25 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define STACKSIZE 256
 
+#ifndef ARM_SOFTFP_ABI
+//hard
 #define	OLD_LDA		[fp, #0 ]
 #define	X		[fp, #4 ]
 #define	OLD_INC_X	[fp, #8 ]
 #define	Y		[fp, #12 ]
 #define	OLD_INC_Y	[fp, #16 ]
 #define OLD_A		r3
+#else
+#define OLD_A_SOFTFP	[fp, #0 ]
+#define	OLD_LDA		[fp, #4 ]
+#define	X		[fp, #8 ]
+#define	OLD_INC_X	[fp, #12 ]
+#define	Y		[fp, #16 ]
+#define	OLD_INC_Y	[fp, #20 ]
+#define OLD_ALPHA	r3
+#define OLD_A		r3
+#endif
+
 #define	OLD_M		r0
 
 #define AO1	r0
@@ -552,9 +565,18 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	cmp	N, #0
 	ble	gemvn_kernel_L999
 
+#ifndef DOUBLE
+#ifdef ARM_SOFTFP_ABI
+
+	vmov	s0, 	OLD_ALPHA
+	ldr	OLD_A,	OLD_A_SOFTFP
+#endif
+#endif
+
 	str	OLD_A, A
 	str	OLD_M, M
-
+
+
 	ldr    INC_X , OLD_INC_X
 	ldr    INC_Y , OLD_INC_Y
 

kernel/arm/gemv_t_vfp.S

@@ -38,12 +38,25 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define STACKSIZE 256
 
+#ifndef ARM_SOFTFP_ABI
+//hard abi	
 #define	OLD_LDA		[fp, #0 ]
 #define	X		[fp, #4 ]
 #define	OLD_INC_X	[fp, #8 ]
 #define	Y		[fp, #12 ]
 #define	OLD_INC_Y	[fp, #16 ]
 #define OLD_A		r3
+#else
+#define OLD_A_SOFTFP	[fp, #0 ]
+#define	OLD_LDA		[fp, #4 ]
+#define	X		[fp, #8 ]
+#define	OLD_INC_X	[fp, #12 ]
+#define	Y		[fp, #16 ]
+#define	OLD_INC_Y	[fp, #20 ]
+#define OLD_ALPHA	r3
+#define OLD_A		r3
+#endif
+
 #define	OLD_N		r1
 
 #define M	r0
@@ -505,6 +518,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	cmp	OLD_N, #0
 	ble	gemvt_kernel_L999
 
+#ifndef DOUBLE
+#ifdef ARM_SOFTFP_ABI
+	vmov	s0,	OLD_ALPHA
+	ldr	OLD_A,	OLD_A_SOFTFP
+#endif
+#endif
+
 	str	OLD_A, A
 	str	OLD_N, N
 

kernel/arm/nrm2_vfpv3.S

@@ -503,6 +503,9 @@ nrm2_kernel_L999:
 #else
 	vsqrt.f32	s1, s1
 	vmul.f32	s0, s0, s1
+#ifdef ARM_SOFTFP_ABI
+	vmov		r0, s0
+#endif
 #endif
 
 	bx	lr

kernel/arm/sdot_vfp.S

@@ -332,10 +332,16 @@ sdot_kernel_L999:
 
 	vadd.f64	d0 , d0, d1				// set return value
 
+#ifdef ARM_SOFTFP_ABI
+	vmov	r0, r1, d0
+#endif
+
 #else
 
 	vadd.f32	s0 , s0, s1				// set return value
-
+#ifdef ARM_SOFTFP_ABI
+	vmov	r0, s0
+#endif
 #endif
 	sub	sp, fp, #24
 	pop	{r4 - r9, fp}

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
@@ -71,18 +77,28 @@ 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]
 #define FP_ZERO_1 [fp, #-236]
 
 #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