Optimized scal kernels for CORTEXA57

Co-Authored-By: Ralph Campbell <ralph.campbell@broadcom.com>
This commit is contained in:
Ashwin Sekhar T K 2015-10-06 14:36:31 +05:30
parent fb4be3b3eb
commit ea1465cdf8
3 changed files with 532 additions and 0 deletions

View File

@ -40,4 +40,9 @@ DROTKERNEL = rot.S
CROTKERNEL = zrot.S
ZROTKERNEL = zrot.S
SCALKERNEL = scal.S
DSCALKERNEL = scal.S
CSCALKERNEL = zscal.S
ZSCALKERNEL = zscal.S

253
kernel/arm64/scal.S Normal file
View File

@ -0,0 +1,253 @@
/*******************************************************************************
Copyright (c) 2015, 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 x3 /* X vector address */
#define X_COPY x5 /* X vector address */
#define INC_X x4 /* X stride */
#define I x1 /* loop variable */
/*******************************************************************************
* Macro definitions
*******************************************************************************/
#if !defined(DOUBLE)
#define DA s0 /* scale input value */
#define DAV {v0.s}[0]
#define TMPF s1
#define TMPVF {v1.s}[0]
#define SZ 4
#else
#define DA d0 /* scale input value */
#define DAV {v0.d}[0]
#define TMPF d1
#define TMPVF {v1.d}[0]
#define SZ 8
#endif
/******************************************************************************/
.macro KERNEL_F1
ldr TMPF, [X]
fmul TMPF, TMPF, DA
str TMPF, [X], #SZ
.endm
.macro KERNEL_INIT_F8
#if !defined(DOUBLE)
ins v0.s[1], v0.s[0]
ins v0.s[2], v0.s[0]
ins v0.s[3], v0.s[0]
#else
ins v0.d[1], v0.d[0]
#endif
.endm
.macro KERNEL_F8
#if !defined(DOUBLE)
ld1 {v1.4s, v2.4s}, [X]
fmul v1.4s, v1.4s, v0.4s
fmul v2.4s, v2.4s, v0.4s
st1 {v1.4s, v2.4s}, [X], #32
#else // DOUBLE
ld1 {v1.2d, v2.2d, v3.2d, v4.2d}, [X]
fmul v1.2d, v1.2d, v0.2d
fmul v2.2d, v2.2d, v0.2d
fmul v3.2d, v3.2d, v0.2d
fmul v4.2d, v4.2d, v0.2d
st1 {v1.2d, v2.2d, v3.2d, v4.2d}, [X], #64
#endif
PRFM PLDL1KEEP, [X, #1024]
.endm
.macro INIT_S
#if !defined(DOUBLE)
lsl INC_X, INC_X, #2
#else
lsl INC_X, INC_X, #3
#endif
.endm
.macro KERNEL_S1
ldr TMPF, [X]
fmul TMPF, TMPF, DA
st1 TMPVF, [X], INC_X
.endm
.macro KERNEL_S4
#if !defined(DOUBLE)
ldr s1, [X]
add X, X, INC_X
fmul s1, s1, s0
str s1, [X_COPY]
add X_COPY, X_COPY, INC_X
ldr s2, [X]
add X, X, INC_X
fmul s2, s2, s0
str s2, [X_COPY]
add X_COPY, X_COPY, INC_X
ldr s3, [X]
add X, X, INC_X
fmul s3, s3, s0
str s3, [X_COPY]
add X_COPY, X_COPY, INC_X
ldr s4, [X]
add X, X, INC_X
fmul s4, s4, s0
str s4, [X_COPY]
add X_COPY, X_COPY, INC_X
#else
ldr d1, [X]
add X, X, INC_X
fmul d1, d1, d0
str d1, [X_COPY]
add X_COPY, X_COPY, INC_X
ldr d2, [X]
add X, X, INC_X
fmul d2, d2, d0
str d2, [X_COPY]
add X_COPY, X_COPY, INC_X
ldr d3, [X]
add X, X, INC_X
fmul d3, d3, d0
str d3, [X_COPY]
add X_COPY, X_COPY, INC_X
ldr d4, [X]
add X, X, INC_X
fmul d4, d4, d0
str d4, [X_COPY]
add X_COPY, X_COPY, INC_X
#endif
.endm
/*******************************************************************************
* End of macro definitions
*******************************************************************************/
PROLOGUE
cmp N, xzr
ble scal_kernel_L999
fcmp DA, #0.0
beq scal_kernel_zero
cmp INC_X, #1
bne scal_kernel_S_BEGIN
scal_kernel_F_BEGIN:
asr I, N, #3
cmp I, xzr
beq scal_kernel_F1
KERNEL_INIT_F8
scal_kernel_F8:
KERNEL_F8
subs I, I, #1
bne scal_kernel_F8
scal_kernel_F1:
ands I, N, #7
ble scal_kernel_L999
scal_kernel_F10:
KERNEL_F1
subs I, I, #1
bne scal_kernel_F10
mov w0, wzr
ret
scal_kernel_S_BEGIN:
INIT_S
mov X_COPY, X
asr I, N, #2
cmp I, xzr
ble scal_kernel_S1
scal_kernel_S4:
KERNEL_S4
subs I, I, #1
bne scal_kernel_S4
scal_kernel_S1:
ands I, N, #3
ble scal_kernel_L999
scal_kernel_S10:
KERNEL_S1
subs I, I, #1
bne scal_kernel_S10
scal_kernel_L999:
mov w0, wzr
ret
scal_kernel_zero:
INIT_S
scal_kernel_Z1:
st1 DAV, [X], INC_X
subs N, N, #1
bne scal_kernel_Z1
mov w0, wzr
ret
EPILOGUE

274
kernel/arm64/zscal.S Normal file
View File

@ -0,0 +1,274 @@
/*******************************************************************************
Copyright (c) 2015, 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 x3 /* X vector address */
#define INC_X x4 /* X stride */
#define I x5 /* loop variable */
/*******************************************************************************
* Macro definitions
*******************************************************************************/
#if !defined(DOUBLE)
#define DA_R s0 /* real scale input value */
#define DA_I s1 /* imaginary scale input value */
#else
#define DA_R d0 /* real scale input value */
#define DA_I d1 /* imaginary scale input value */
#endif
/******************************************************************************/
.macro INIT
#if !defined(DOUBLE)
ins v0.s[1], v0.s[0] // v0 = DA_R, DA_R
fneg s2, DA_I
ins v1.s[1], v2.s[0] // v1 = -DA_I, DA_I
ext v1.8b, v1.8b, v1.8b, #4 // v1 = DA_I, -DA_I
#else
ins v0.d[1], v0.d[0] // v0 = DA_R, DA_R
fneg d2, DA_I
ins v1.d[1], v2.d[0] // v1 = DA_I, DA_I
ext v1.16b, v1.16b, v1.16b, #8 // v1 = DA_I, -DA_I
#endif
.endm
.macro KERNEL_F1
#if !defined(DOUBLE)
ld1 {v2.2s}, [X] // X1, X0
ext v3.8b, v2.8b, v2.8b, #4 // X0, X1
fmul v2.2s, v2.2s, v0.2s // DA_R*X1, DA_R*X0
fmla v2.2s, v3.2s, v1.2s // DA_R*X1+DA_I*X0, DA_R*X0-DA_I*X1
st1 {v2.2s}, [X], #8
#else
ld1 {v2.2d}, [X] // X1, X0
ext v3.16b, v2.16b, v2.16b, #8 // X0, X1
fmul v2.2d, v2.2d, v0.2d // DA_R*X1, DA_R*X0
fmla v2.2d, v3.2d, v1.2d // DA_R*X1+DA_I*X0, DA_R*X0-DA_I*X1
st1 {v2.2d}, [X], #16
#endif
.endm
.macro KERNEL_INIT_F4
#if !defined(DOUBLE)
// Replicate the lower 2 floats into the upper 2 slots
ins v0.d[1], v0.d[0] // v0 = DA_R, DA_R, DA_R, DA_R
ins v1.d[1], v1.d[0] // v1 = DA_I, DA_I, DA_I, DA_I
#endif
.endm
.macro KERNEL_F4
#if !defined(DOUBLE)
ld1 {v2.4s,v3.4s}, [X] // V2 = X[3], X[2], X[1], X[0]
// V3 = X[7], X[6], X[5], X[4]
ext v6.8b, v2.8b, v2.8b, #4 // V6 = - , - , X[0], X[1]
ins v6.s[2], v2.s[3] // V6 = - , X[3], X[0], X[1]
ins v6.s[3], v2.s[2] // V6 = X[2], X[3], X[0], X[1]
fmul v2.4s, v0.4s, v2.4s // X'[ix] += DA_R * X[ix]
// X'[ix+1] += DA_R * X[ix+1]
fmla v2.4s, v1.4s, v6.4s // X'[ix] += -DA_I * X[ix+1]
// X'[ix+1] += DA_I * X[ix]
ext v7.8b, v3.8b, v3.8b, #4 // V7 = - , - , X[4], X[5]
ins v7.s[2], v3.s[3] // V7 = - , X[7], X[4], X[5]
ins v7.s[3], v3.s[2] // V7 = X[6], X[7], X[4], X[5]
fmul v3.4s, v0.4s, v3.4s // X'[ix] += DA_R * X[ix]
// X'[ix+1] += DA_R * X[ix+1]
fmla v3.4s, v1.4s, v7.4s // X'[ix] += -DA_I * X[ix+1]
// X'[ix+1] += DA_I * X[ix]
st1 {v2.4s,v3.4s}, [X], #32
#else // DOUBLE
ld1 {v2.2d,v3.2d,v4.2d,v5.2d}, [X] // CX0, CX1, CX2, CX3
ext v20.16b, v2.16b, v2.16b, #8 // X[ix], X[ix+1]
ext v21.16b, v3.16b, v3.16b, #8 // X[ix], X[ix+1]
ext v22.16b, v4.16b, v4.16b, #8 // X[ix], X[ix+1]
ext v23.16b, v5.16b, v5.16b, #8 // X[ix], X[ix+1]
fmul v2.2d, v0.2d, v2.2d
fmla v2.2d, v1.2d, v20.2d
fmul v3.2d, v0.2d, v3.2d
fmla v3.2d, v1.2d, v21.2d
st1 {v2.2d,v3.2d}, [X], #32
fmul v4.2d, v0.2d, v4.2d
fmla v4.2d, v1.2d, v22.2d
fmul v5.2d, v0.2d, v5.2d
fmla v5.2d, v1.2d, v23.2d
st1 {v4.2d,v5.2d}, [X], #32
#endif
PRFM PLDL1KEEP, [X, #1024]
.endm
.macro INIT_S
#if !defined(DOUBLE)
lsl INC_X, INC_X, #3
#else
lsl INC_X, INC_X, #4
#endif
.endm
.macro KERNEL_S1
#if !defined(DOUBLE)
ld1 {v2.2s}, [X] // X1, X0
ext v3.8b, v2.8b, v2.8b, #4 // X0, X1
fmul v2.2s, v2.2s, v0.2s // DA_R*X1, DA_R*X0
fmla v2.2s, v3.2s, v1.2s // DA_R*X1+DA_I*X0, DA_R*X0-DA_I*X1
st1 {v2.2s}, [X], INC_X
#else
ld1 {v2.2d}, [X] // X1, X0
ext v3.16b, v2.16b, v2.16b, #8 // X0, X1
fmul v2.2d, v2.2d, v0.2d // DA_R*X1, DA_R*X0
fmla v2.2d, v3.2d, v1.2d // DA_R*X1+DA_I*X0, DA_R*X0-DA_I*X1
st1 {v2.2d}, [X], INC_X
#endif
.endm
/*******************************************************************************
* End of macro definitions
*******************************************************************************/
PROLOGUE
cmp N, xzr
ble zscal_kernel_L999
fcmp DA_R, #0.0
bne zscal_kernel_1
fcmp DA_I, #0.0
beq zscal_kernel_zero
// TODO: special case DA_R == 0 && DA_I != 0
zscal_kernel_1:
// TODO: special case DA_R != 0 && DA_I == 0
INIT
cmp INC_X, #1
bne zscal_kernel_S_BEGIN
zscal_kernel_F_BEGIN:
asr I, N, #2
cmp I, xzr
beq zscal_kernel_F1
KERNEL_INIT_F4
zscal_kernel_F4:
KERNEL_F4
subs I, I, #1
bne zscal_kernel_F4
zscal_kernel_F1:
ands I, N, #3
ble zscal_kernel_L999
zscal_kernel_F10:
KERNEL_F1
subs I, I, #1
bne zscal_kernel_F10
mov w0, wzr
ret
zscal_kernel_S_BEGIN:
INIT_S
asr I, N, #2
cmp I, xzr
ble zscal_kernel_S1
zscal_kernel_S4:
KERNEL_S1
KERNEL_S1
KERNEL_S1
KERNEL_S1
subs I, I, #1
bne zscal_kernel_S4
zscal_kernel_S1:
ands I, N, #3
ble zscal_kernel_L999
zscal_kernel_S10:
KERNEL_S1
subs I, I, #1
bne zscal_kernel_S10
zscal_kernel_L999:
mov w0, wzr
ret
zscal_kernel_zero:
INIT_S
zscal_kernel_Z1:
stp DA_R, DA_I, [X]
add X, X, INC_X
subs N, N, #1
bne zscal_kernel_Z1
mov w0, wzr
ret
EPILOGUE