THUNDERX2T99: Add Optimized CNRM2 Implementation

kernel/arm64/KERNEL.THUNDERX2T99

@@ -1,6 +1,8 @@
 include $(KERNELDIR)/KERNEL.CORTEXA57
 
 SNRM2KERNEL    = snrm2_thunderx2t99.S
+CNRM2KERNEL    = cnrm2_thunderx2t99.S
+
 DAXPYKERNEL    = daxpy_thunderx2t99.S
 
 ifndef SMP

kernel/arm64/cnrm2_thunderx2t99.S

@@ -0,0 +1,228 @@
+/*******************************************************************************
+Copyright (c) 2017, 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.
+*******************************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+
+#define	N	x0	/* vector length */
+#define	X	x1	/* X vector address */
+#define	INC_X	x2	/* X stride */
+#define I	x5	/* loop variable */
+
+/*******************************************************************************
+* Macro definitions
+*******************************************************************************/
+
+#define TMPF	d16
+#define SSQ	s0
+#define SSQD	d0
+
+/******************************************************************************/
+
+.macro INIT
+	fmov	SSQD, xzr
+	fmov	d1, xzr
+	fmov	d2, xzr
+	fmov	d3, xzr
+	fmov	d4, xzr
+	fmov	d5, xzr
+	fmov	d6, xzr
+	fmov	d7, xzr
+.endm
+
+.macro KERNEL_F1
+	ldr	TMPF, [X]
+	add	X, X, #8
+	fcvtl	v16.2d, v16.2s
+	fmla	v0.2d, v16.2d, v16.2d
+.endm
+
+.macro KERNEL_F16
+	ldur	q16, [X]
+	ldur	q18, [X, #16]
+	ldur	q20, [X, #32]
+	ldur	q22, [X, #48]
+	ldur	q24, [X, #64]
+	ldur	q26, [X, #80]
+	ldur	q28, [X, #96]
+	ldur	q30, [X, #112]
+
+	add	X, X, #128
+
+	fcvtl2	v17.2d, v16.4s
+	fcvtl	v16.2d, v16.2s
+	fcvtl2	v19.2d, v18.4s
+	fcvtl	v18.2d, v18.2s
+	fcvtl2	v21.2d, v20.4s
+	fcvtl	v20.2d, v20.2s
+	fcvtl2	v23.2d, v22.4s
+	fcvtl	v22.2d, v22.2s
+	fcvtl2	v25.2d, v24.4s
+	fcvtl	v24.2d, v24.2s
+	fcvtl2	v27.2d, v26.4s
+	fcvtl	v26.2d, v26.2s
+	fcvtl2	v29.2d, v28.4s
+	fcvtl	v28.2d, v28.2s
+	fcvtl2	v31.2d, v30.4s
+	fcvtl	v30.2d, v30.2s
+
+	fmla	v0.2d, v16.2d, v16.2d
+	fmla	v1.2d, v17.2d, v17.2d
+	fmla	v2.2d, v18.2d, v18.2d
+	fmla	v3.2d, v19.2d, v19.2d
+	fmla	v4.2d, v20.2d, v20.2d
+	fmla	v5.2d, v21.2d, v21.2d
+	fmla	v6.2d, v22.2d, v22.2d
+	fmla	v7.2d, v23.2d, v23.2d
+
+	fmla	v0.2d, v24.2d, v24.2d
+	fmla	v1.2d, v25.2d, v25.2d
+	fmla	v2.2d, v26.2d, v26.2d
+	fmla	v3.2d, v27.2d, v27.2d
+	fmla	v4.2d, v28.2d, v28.2d
+	fmla	v5.2d, v29.2d, v29.2d
+	fmla	v6.2d, v30.2d, v30.2d
+	fmla	v7.2d, v31.2d, v31.2d
+
+	prfm	PLDL1KEEP, [X, #1024]
+	prfm	PLDL1KEEP, [X, #1024+64]
+.endm
+
+.macro KERNEL_F16_FINALIZE
+	fadd	v0.2d, v0.2d, v1.2d
+	fadd	v2.2d, v2.2d, v3.2d
+	fadd	v4.2d, v4.2d, v5.2d
+	fadd	v6.2d, v6.2d, v7.2d
+
+	fadd	v0.2d, v0.2d, v2.2d
+	fadd	v4.2d, v4.2d, v6.2d
+
+	fadd	v0.2d, v0.2d, v4.2d
+.endm
+
+.macro KERNEL_FINALIZE
+	faddp	SSQD, v0.2d
+.endm
+
+.macro INIT_S
+	lsl	INC_X, INC_X, #3
+.endm
+
+.macro KERNEL_S1
+	ldr	TMPF, [X]
+	add	X, X, INC_X
+	fcvtl	v16.2d, v16.2s
+	fmla	v0.2d, v16.2d, v16.2d
+.endm
+
+/*******************************************************************************
+* End of macro definitions
+*******************************************************************************/
+
+	PROLOGUE
+
+	INIT
+
+	cmp	N, xzr
+	ble	nrm2_kernel_zero
+	cmp	INC_X, xzr
+	ble	nrm2_kernel_zero
+	cmp	INC_X, #1
+	bne	nrm2_kernel_S_BEGIN
+
+nrm2_kernel_F_BEGIN:
+
+	asr	I, N, #4
+	cmp	I, xzr
+	beq	nrm2_kernel_S_BEGIN
+
+	.align 5
+nrm2_kernel_F16:
+
+	KERNEL_F16
+
+	subs	I, I, #1
+	bne	nrm2_kernel_F16
+
+	KERNEL_F16_FINALIZE
+
+nrm2_kernel_F1:
+
+	ands	I, N, #15
+	ble	nrm2_kernel_L999
+
+nrm2_kernel_F10:
+
+	KERNEL_F1
+
+	subs    I, I, #1
+	bne     nrm2_kernel_F10
+
+	b	nrm2_kernel_L999
+
+nrm2_kernel_S_BEGIN:
+
+	INIT_S
+
+	asr	I, N, #2
+	cmp	I, xzr
+	ble	nrm2_kernel_S1
+
+nrm2_kernel_S4:
+
+	KERNEL_S1
+	KERNEL_S1
+	KERNEL_S1
+	KERNEL_S1
+
+	subs	I, I, #1
+	bne	nrm2_kernel_S4
+
+nrm2_kernel_S1:
+
+	ands	I, N, #3
+	ble	nrm2_kernel_L999
+
+nrm2_kernel_S10:
+
+	KERNEL_S1
+
+	subs    I, I, #1
+	bne     nrm2_kernel_S10
+
+nrm2_kernel_L999:
+	KERNEL_FINALIZE
+	fsqrt	SSQD, SSQD
+	fcvt	SSQ, SSQD
+	ret
+
+nrm2_kernel_zero:
+	fmov	SSQ, wzr
+
+	ret
+
+	EPILOGUE