add single and double precision gemv_n kernel for ARMV6

kernel/arm/KERNEL.ARMV6

@@ -70,8 +70,8 @@ DSWAPKERNEL  = swap_vfp.S
 CSWAPKERNEL  = swap_vfp.S
 ZSWAPKERNEL  = swap_vfp.S
 
-SGEMVNKERNEL = gemv_n.c
-DGEMVNKERNEL = gemv_n.c
+SGEMVNKERNEL = gemv_n_vfp.S
+DGEMVNKERNEL = gemv_n_vfp.S
 CGEMVNKERNEL = zgemv_n.c
 ZGEMVNKERNEL = zgemv_n.c
 

kernel/arm/gemv_n_vfp.S

@@ -0,0 +1,675 @@
+/***************************************************************************
+Copyright (c) 2013, 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.
+*****************************************************************************/
+
+/**************************************************************************************
+* 2013/11/24 Saar
+* 	 BLASTEST 		: OK
+* 	 CTEST			: OK
+* 	 TEST			: OK
+*
+**************************************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+
+#define STACKSIZE 256
+
+#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
+#define	OLD_M		r0
+
+#define AO1	r0
+#define N	r1
+#define J	r2
+
+#define AO2	r4
+#define XO	r5
+#define YO	r6
+#define LDA	r7
+#define INC_X	r8
+#define INC_Y	r9
+
+#define I	r12
+
+#define M	[fp, #-252 ]
+#define A	[fp, #-256 ]
+
+
+#define X_PRE	64
+#define Y_PRE	0
+#define A_PRE	0
+
+/**************************************************************************************
+* Macro definitions
+**************************************************************************************/
+
+
+#if	defined(DOUBLE)
+
+.macro INIT_F4
+
+	pld     [ YO , #Y_PRE ]
+
+	vsub.f64	d12 , d12 , d12
+	vmov.f64	d13 , d12
+	vmov.f64	d14 , d12
+	vmov.f64	d15 , d12
+
+.endm
+
+.macro KERNEL_F4X4
+
+	pld     [ XO , #X_PRE ]
+	KERNEL_F4X1
+	KERNEL_F4X1
+	KERNEL_F4X1
+	KERNEL_F4X1
+
+.endm
+
+
+.macro KERNEL_F4X1
+
+	fldmiad	XO! ,  { d2 }
+	fldmiad	AO1 ,  { d8 - d11 }
+
+	vmla.f64	d12 , d2 , d8
+	pld	[ AO2 , #A_PRE ]
+	vmla.f64	d13 , d2 , d9
+	add		AO1, AO1, LDA
+	vmla.f64	d14 , d2 , d10
+	vmla.f64	d15 , d2 , d11
+	add		AO2, AO2, LDA
+
+.endm
+
+.macro	SAVE_F4
+
+	fldmiad	YO,  { d4 - d7 }
+
+	vmla.f64	d4 , d0, d12
+	vmla.f64	d5 , d0, d13
+	vmla.f64	d6 , d0, d14
+	vmla.f64	d7 , d0, d15
+
+	fstmiad	YO!, { d4 - d7 }
+
+.endm
+
+
+.macro INIT_F1
+
+	vsub.f64	d12 , d12 , d12
+
+.endm
+
+
+
+.macro KERNEL_F1X1
+
+	fldmiad	XO! ,  { d2 }
+	fldmiad	AO1 ,  { d8 }
+	vmla.f64	d12 , d2 , d8
+	add		AO1, AO1, LDA
+
+.endm
+
+.macro	SAVE_F1
+
+	fldmiad	YO,  { d4 }
+	vmla.f64	d4, d0, d12
+	fstmiad	YO!, { d4 }
+
+.endm
+
+/*********************************************************************************************/
+
+.macro INIT_S4
+
+	vsub.f64	d12 , d12 , d12
+	vmov.f64	d13 , d12
+	vmov.f64	d14 , d12
+	vmov.f64	d15 , d12
+
+.endm
+
+.macro KERNEL_S4X4
+
+	KERNEL_S4X1
+	KERNEL_S4X1
+	KERNEL_S4X1
+	KERNEL_S4X1
+
+.endm
+
+
+.macro KERNEL_S4X1
+
+	pld	[ AO2 , #A_PRE ]
+	fldmiad	XO  ,  { d2 }
+	fldmiad	AO1 ,  { d8 - d11 }
+
+	vmla.f64	d12 , d2 , d8
+	add		AO1, AO1, LDA
+	vmla.f64	d13 , d2 , d9
+	add		AO2, AO2, LDA
+	vmla.f64	d14 , d2 , d10
+	vmla.f64	d15 , d2 , d11
+	add		XO, XO , INC_X
+
+.endm
+
+.macro	SAVE_S4
+
+	fldmiad	YO,  { d4 }
+	vmla.f64	d4 , d0, d12
+	fstmiad	YO,  { d4 }
+	add	YO, YO, INC_Y
+
+	fldmiad	YO,  { d5 }
+	vmla.f64	d5 , d0, d13
+	fstmiad	YO,  { d5 }
+	add	YO, YO, INC_Y
+
+	fldmiad	YO,  { d4 }
+	vmla.f64	d4 , d0, d14
+	fstmiad	YO,  { d4 }
+	add	YO, YO, INC_Y
+
+	fldmiad	YO,  { d5 }
+	vmla.f64	d5 , d0, d15
+	fstmiad	YO,  { d5 }
+	add	YO, YO, INC_Y
+
+.endm
+
+
+.macro INIT_S1
+
+	vsub.f64	d12 , d12 , d12
+
+.endm
+
+
+
+.macro KERNEL_S1X1
+
+	fldmiad	XO  ,  { d2 }
+	fldmiad	AO1 ,  { d8 }
+	vmla.f64	d12 , d2 , d8
+	add		AO1, AO1, LDA
+	add		XO, XO , INC_X
+
+.endm
+
+.macro	SAVE_S1
+
+	fldmiad	YO,  { d4 }
+	vmla.f64	d4, d0, d12
+	fstmiad	YO , { d4 }
+	add	YO, YO, INC_Y
+
+.endm
+
+
+
+
+#else	/************************* SINGLE PRECISION *****************************************/
+
+.macro INIT_F4
+
+	pld     [ YO , #Y_PRE ]
+
+	vsub.f32	s12 , s12 , s12
+	vmov.f32	s13 , s12
+	vmov.f32	s14 , s12
+	vmov.f32	s15 , s12
+
+.endm
+
+.macro KERNEL_F4X4
+
+	pld     [ XO , #X_PRE ]
+	KERNEL_F4X1
+	KERNEL_F4X1
+	KERNEL_F4X1
+	KERNEL_F4X1
+
+.endm
+
+
+.macro KERNEL_F4X1
+
+	fldmias	XO! ,  { s2 }
+	fldmias	AO1 ,  { s8 - s11 }
+
+	vmla.f32	s12 , s2 , s8
+	vmla.f32	s13 , s2 , s9
+	vmla.f32	s14 , s2 , s10
+	vmla.f32	s15 , s2 , s11
+	add		AO1, AO1, LDA
+	add		AO2, AO2, LDA
+
+.endm
+
+.macro	SAVE_F4
+
+	fldmias	YO,  { s4 - s7 }
+
+	vmla.f32	s4 , s0, s12
+	vmla.f32	s5 , s0, s13
+	vmla.f32	s6 , s0, s14
+	vmla.f32	s7 , s0, s15
+
+	fstmias	YO!, { s4 - s7 }
+
+.endm
+
+
+.macro INIT_F1
+
+	vsub.f32	s12 , s12 , s12
+
+.endm
+
+
+
+.macro KERNEL_F1X1
+
+	fldmias	XO! ,  { s2 }
+	fldmias	AO1 ,  { s8 }
+	vmla.f32	s12 , s2 , s8
+	add		AO1, AO1, LDA
+
+.endm
+
+.macro	SAVE_F1
+
+	fldmias	YO,  { s4 }
+	vmla.f32	s4, s0, s12
+	fstmias	YO!, { s4 }
+
+.endm
+
+/*********************************************************************************************/
+
+.macro INIT_S4
+
+	vsub.f32	s12 , s12 , s12
+	vmov.f32	s13 , s12
+	vmov.f32	s14 , s12
+	vmov.f32	s15 , s12
+
+.endm
+
+.macro KERNEL_S4X4
+
+	KERNEL_S4X1
+	KERNEL_S4X1
+	KERNEL_S4X1
+	KERNEL_S4X1
+
+.endm
+
+
+.macro KERNEL_S4X1
+
+	pld	[ AO2 , #A_PRE ]
+	fldmias	XO  ,  { s2 }
+	fldmias	AO1 ,  { s8 - s11 }
+
+	vmla.f32	s12 , s2 , s8
+	vmla.f32	s13 , s2 , s9
+	vmla.f32	s14 , s2 , s10
+	vmla.f32	s15 , s2 , s11
+	add		AO1, AO1, LDA
+	add		AO2, AO2, LDA
+	add		XO, XO , INC_X
+
+.endm
+
+.macro	SAVE_S4
+
+	fldmias	YO,  { s4 }
+	vmla.f32	s4 , s0, s12
+	fstmias	YO,  { s4 }
+	add	YO, YO, INC_Y
+
+	fldmias	YO,  { s5 }
+	vmla.f32	s5 , s0, s13
+	fstmias	YO,  { s5 }
+	add	YO, YO, INC_Y
+
+	fldmias	YO,  { s4 }
+	vmla.f32	s4 , s0, s14
+	fstmias	YO,  { s4 }
+	add	YO, YO, INC_Y
+
+	fldmias	YO,  { s5 }
+	vmla.f32	s5 , s0, s15
+	fstmias	YO,  { s5 }
+	add	YO, YO, INC_Y
+
+.endm
+
+
+.macro INIT_S1
+
+	vsub.f32	s12 , s12 , s12
+
+.endm
+
+
+
+.macro KERNEL_S1X1
+
+	fldmias	XO  ,  { s2 }
+	fldmias	AO1 ,  { s8 }
+	vmla.f32	s12 , s2 , s8
+	add		AO1, AO1, LDA
+	add		XO, XO , INC_X
+
+.endm
+
+.macro	SAVE_S1
+
+	fldmias	YO,  { s4 }
+	vmla.f32	s4, s0, s12
+	fstmias	YO , { s4 }
+	add	YO, YO, INC_Y
+
+.endm
+
+
+
+
+#endif
+
+/**************************************************************************************
+* End of macro definitions
+**************************************************************************************/
+
+	PROLOGUE
+
+	.align 5
+	push    {r4 - r9 , fp}
+        add     fp, sp, #28
+	sub     sp, sp, #STACKSIZE                              // reserve stack
+
+        sub     r12, fp, #192
+
+#if	defined(DOUBLE)
+        vstm    r12, { d8 - d15 }                                 // store floating point registers
+#else
+        vstm    r12, { s8 - s15 }                                 // store floating point registers
+#endif
+
+	cmp	OLD_M, #0
+	ble	gemvn_kernel_L999
+
+	cmp	N, #0
+	ble	gemvn_kernel_L999
+
+	str	OLD_A, A
+	str	OLD_M, M
+
+	ldr    INC_X , OLD_INC_X
+	ldr    INC_Y , OLD_INC_Y
+
+	cmp	INC_X, #0
+	beq	gemvn_kernel_L999
+
+	cmp	INC_Y, #0
+	beq	gemvn_kernel_L999
+
+	ldr	LDA, OLD_LDA
+
+
+#if defined(DOUBLE)
+	lsl	LDA, LDA, #3				// LDA * SIZE
+#else
+	lsl	LDA, LDA, #2				// LDA * SIZE
+#endif
+
+	cmp	INC_X, #1
+	bne	gemvn_kernel_S4_BEGIN
+
+	cmp	INC_Y, #1
+	bne	gemvn_kernel_S4_BEGIN
+
+
+gemvn_kernel_F4_BEGIN:
+
+	ldr	YO , Y
+
+	ldr	I, M
+	asrs	I, I, #2					// I = M / 4
+	ble	gemvn_kernel_F1_BEGIN
+
+gemvn_kernel_F4X4:
+
+	ldr	AO1, A
+	add	AO2, AO1, LDA
+	add	r3 , AO1, #4*SIZE
+	str	r3 , A
+
+	ldr	XO , X
+
+	INIT_F4
+
+	asrs	J, N, #2					// J = N / 4
+	ble	gemvn_kernel_F4X1
+
+
+gemvn_kernel_F4X4_10:
+
+	KERNEL_F4X4
+
+	subs	J, J, #1
+	bne	gemvn_kernel_F4X4_10
+
+
+gemvn_kernel_F4X1:
+
+	ands	J, N , #3
+	ble	gemvn_kernel_F4_END
+
+gemvn_kernel_F4X1_10:
+
+	KERNEL_F4X1
+
+	subs	J, J, #1
+	bne	gemvn_kernel_F4X1_10
+
+
+gemvn_kernel_F4_END:
+
+	SAVE_F4
+
+	subs	I , I , #1
+	bne	gemvn_kernel_F4X4
+
+
+gemvn_kernel_F1_BEGIN:
+
+	ldr	I, M
+	ands	I,  I , #3
+	ble	gemvn_kernel_L999
+
+gemvn_kernel_F1X1:
+
+	ldr	AO1, A
+	add	r3, AO1, #SIZE
+	str	r3, A
+	
+	ldr	XO , X
+
+	INIT_F1
+
+	mov	J, N
+
+
+gemvn_kernel_F1X1_10:
+
+	KERNEL_F1X1
+
+	subs	J, J, #1
+	bne	gemvn_kernel_F1X1_10
+
+
+gemvn_kernel_F1_END:
+
+	SAVE_F1
+
+	subs	I , I , #1
+	bne	gemvn_kernel_F1X1
+
+	b	gemvn_kernel_L999
+
+
+
+/*************************************************************************************************************/
+
+gemvn_kernel_S4_BEGIN:
+
+#if defined(DOUBLE)
+	lsl	INC_X, INC_X, #3				// INC_X * SIZE
+	lsl	INC_Y, INC_Y, #3				// INC_Y * SIZE
+#else
+	lsl	INC_X, INC_X, #2				// INC_X * SIZE
+	lsl	INC_Y, INC_Y, #2				// INC_Y * SIZE
+#endif
+
+	ldr	YO , Y
+
+	ldr	I, M
+	asrs	I, I, #2					// I = M / 4
+	ble	gemvn_kernel_S1_BEGIN
+
+gemvn_kernel_S4X4:
+
+	ldr	AO1, A
+	add	AO2, AO1, LDA
+	add	r3 , AO1, #4*SIZE
+	str	r3 , A
+
+	ldr	XO , X
+
+	INIT_S4
+
+	asrs	J, N, #2					// J = N / 4
+	ble	gemvn_kernel_S4X1
+
+
+gemvn_kernel_S4X4_10:
+
+	KERNEL_S4X4
+
+	subs	J, J, #1
+	bne	gemvn_kernel_S4X4_10
+
+
+gemvn_kernel_S4X1:
+
+	ands	J, N , #3
+	ble	gemvn_kernel_S4_END
+
+gemvn_kernel_S4X1_10:
+
+	KERNEL_S4X1
+
+	subs	J, J, #1
+	bne	gemvn_kernel_S4X1_10
+
+
+gemvn_kernel_S4_END:
+
+	SAVE_S4
+
+	subs	I , I , #1
+	bne	gemvn_kernel_S4X4
+
+
+gemvn_kernel_S1_BEGIN:
+
+	ldr	I, M
+	ands	I,  I , #3
+	ble	gemvn_kernel_L999
+
+gemvn_kernel_S1X1:
+
+	ldr	AO1, A
+	add	r3, AO1, #SIZE
+	str	r3, A
+	
+	ldr	XO , X
+
+	INIT_S1
+
+	mov	J, N
+
+
+gemvn_kernel_S1X1_10:
+
+	KERNEL_S1X1
+
+	subs	J, J, #1
+	bne	gemvn_kernel_S1X1_10
+
+
+gemvn_kernel_S1_END:
+
+	SAVE_S1
+
+	subs	I , I , #1
+	bne	gemvn_kernel_S1X1
+
+
+/*************************************************************************************************************/
+
+gemvn_kernel_L999:
+
+        sub     r3, fp, #192
+
+#if	defined(DOUBLE)
+        vldm    r3, { d8 - d15 }                                 // restore floating point registers
+#else
+        vldm    r3, { s8 - s15 }                                 // restore floating point registers
+#endif
+
+	mov	r0, #0		// set return value
+
+	sub     sp, fp, #28
+	pop     {r4 -r9 ,fp}
+	bx	lr
+
+	EPILOGUE
+