loongarch64: Add and refine iamax optimization functions.

Author: CNClareChen

common_loongarch64.h

@@ -139,6 +139,7 @@ static inline int WhereAmI(void){
 #define XVFMAX  xvfmax.d
 #define XVFMAXA xvfmaxa.d
 #define XVCMPEQ xvfcmp.ceq.d
+#define XVCMPLE xvfcmp.cle.d
 #define XVCMPLT xvfcmp.clt.d
 #define XVMUL   xvfmul.d
 #define XVMSUB  xvfmsub.d
@@ -151,6 +152,7 @@ static inline int WhereAmI(void){
 #define VFMAX  vfmax.d
 #define VFMAXA vfmaxa.d
 #define VCMPEQ vfcmp.ceq.d
+#define VCMPLE vfcmp.cle.d
 #define VCMPLT vfcmp.clt.d
 #define VMUL   vfmul.d
 #define VMSUB  vfmsub.d
@@ -189,6 +191,7 @@ static inline int WhereAmI(void){
 #define XVFMAX  xvfmax.s
 #define XVFMAXA xvfmaxa.s
 #define XVCMPEQ xvfcmp.ceq.s
+#define XVCMPLE xvfcmp.cle.s
 #define XVCMPLT xvfcmp.clt.s
 #define XVMUL   xvfmul.s
 #define XVMSUB  xvfmsub.s
@@ -201,6 +204,7 @@ static inline int WhereAmI(void){
 #define VFMAX  vfmax.s
 #define VFMAXA vfmaxa.s
 #define VCMPEQ vfcmp.ceq.s
+#define VCMPLE vfcmp.cle.s
 #define VCMPLT vfcmp.clt.s
 #define VMUL   vfmul.s
 #define VMSUB  vfmsub.s

kernel/loongarch64/KERNEL.LOONGSON2K1000

@@ -9,6 +9,7 @@ DSCALKERNEL  = dscal_lsx.S
 
 SAMAXKERNEL =  amax_lsx.S
 DAMAXKERNEL =  amax_lsx.S
+CAMAXKERNEL =  camax_lsx.S
 
 SAMINKERNEL =  amin_lsx.S
 DAMINKERNEL =  amin_lsx.S
@@ -25,8 +26,10 @@ IDMAXKERNEL =  imax_lsx.S
 ISMINKERNEL =  imin_lsx.S
 IDMINKERNEL =  imin_lsx.S
 
-ISAMAXKERNEL = isamax_lsx.S
-IDAMAXKERNEL = idamax_lsx.S
+ISAMAXKERNEL = iamax_lsx.S
+IDAMAXKERNEL = iamax_lsx.S
+ICAMAXKERNEL = icamax_lsx.S
+IZAMAXKERNEL = icamax_lsx.S
 
 ISAMINKERNEL = iamin_lsx.S
 IDAMINKERNEL = iamin_lsx.S

kernel/loongarch64/KERNEL.LOONGSON3R5

@@ -9,6 +9,7 @@ DSCALKERNEL  = dscal_lasx.S
 
 SAMAXKERNEL =  amax_lasx.S
 DAMAXKERNEL =  amax_lasx.S
+CAMAXKERNEL =  camax_lasx.S
 
 SAMINKERNEL =  amin_lasx.S
 DAMINKERNEL =  amin_lasx.S
@@ -25,8 +26,10 @@ IDMAXKERNEL =  imax_lasx.S
 ISMINKERNEL =  imin_lasx.S
 IDMINKERNEL =  imin_lasx.S
 
-ISAMAXKERNEL = isamax_lasx.S
-IDAMAXKERNEL = idamax_lasx.S
+ISAMAXKERNEL = iamax_lasx.S
+IDAMAXKERNEL = iamax_lasx.S
+ICAMAXKERNEL = icamax_lasx.S
+IZAMAXKERNEL = icamax_lasx.S
 
 ISAMINKERNEL = iamin_lasx.S
 IDAMINKERNEL = iamin_lasx.S

kernel/loongarch64/camax_lasx.S

@@ -0,0 +1,194 @@
+/***************************************************************************
+Copyright (c) 2023, 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    $r4
+#define X    $r5
+#define INCX $r6
+#define I    $r12
+#define t1   $f14
+#define t2   $f18
+#define t3   $f15
+#define t4   $f17
+#define s1   $f22
+#define s2   $f9
+#define s3   $f10
+#define s4   $f11
+#define TEMP $r16
+#define a0   $f20
+#define a1   $f21
+#define x1   $xr9
+#define x2   $xr10
+#define x3   $xr11
+#define x4   $xr12
+#define VT0  $xr13
+#define VT1  $xr14
+#define res0 $xr18
+#define neg1 $xr19
+#define VX0  $xr20
+#define VX1  $xr21
+#define VM0  $xr22
+#define VM1  $xr23
+
+    PROLOGUE
+    xvxor.v VM0, VM0, VM0
+    xvxor.v res0, res0, res0
+    bge $r0, N, .L999
+    bge $r0, INCX, .L999
+    li.d TEMP, 1
+    li.w I, -1
+    slli.d TEMP, TEMP, ZBASE_SHIFT
+    slli.d INCX, INCX, ZBASE_SHIFT
+    xvreplgr2vr.w neg1, I
+    xvffint.s.w neg1, neg1
+    srai.d I, N, 3
+    bne INCX, TEMP, .L20
+    bge $r0, I, .L23
+    .align 3
+
+.L10:
+    xvld VX0, X, 0 * SIZE
+    xvld VX1, X, 8 * SIZE
+    addi.d I, I, -1
+    xvpickev.w x1, VX1, VX0
+    xvpickod.w x2, VX1, VX0
+    xvfmul.s x3, neg1, x1
+    xvfmul.s x4, neg1, x2
+    xvfcmp.clt.s VT0, x1, res0
+    xvfcmp.clt.s VT1, x2, res0
+    xvbitsel.v x1, x1, x3, VT0
+    xvbitsel.v x2, x2, x4, VT1
+    addi.d X, X, 16 * SIZE
+    xvfadd.s VM1, x1, x2
+    xvfmax.s VM0, VM0, VM1
+    blt $r0, I, .L10
+    .align 3
+
+.L11:
+    xvpickve.w x1, VM0, 0
+    xvpickve.w x2, VM0, 1
+    xvpickve.w x3, VM0, 2
+    xvpickve.w x4, VM0, 3
+    xvfmax.s VM1, x1, x2
+    xvfmax.s VM0, x3, x4
+    xvfmax.s VM0, VM0, VM1
+    b .L23
+    .align 3
+
+.L20: // INCX!=1
+    bge $r0, I, .L23
+    .align 3
+
+.L21:
+    fld.s t1, X, 0 * SIZE
+    fld.s t2, X, 1 * SIZE
+    add.d X, X, INCX
+    fld.s t3, X, 0 * SIZE
+    fld.s t4, X, 1 * SIZE
+    add.d X, X, INCX
+    fabs.s t1, t1
+    fabs.s t2, t2
+    fabs.s t3, t3
+    fabs.s t4, t4
+    fadd.s t1, t1, t2
+    fadd.s t3, t3, t4
+    fmax.s s1, t1, t3
+    fld.s t1, X, 0 * SIZE
+    fld.s t2, X, 1 * SIZE
+    add.d X, X, INCX
+    fld.s t3, X, 0 * SIZE
+    fld.s t4, X, 1 * SIZE
+    add.d X, X, INCX
+    fabs.s t1, t1
+    fabs.s t2, t2
+    fabs.s t3, t3
+    fabs.s t4, t4
+    fadd.s t1, t1, t2
+    fadd.s t3, t3, t4
+    fmax.s s1, t1, t3
+    fld.s t1, X, 0 * SIZE
+    fld.s t2, X, 1 * SIZE
+    add.d X, X, INCX
+    fld.s t3, X, 0 * SIZE
+    fld.s t4, X, 1 * SIZE
+    add.d X, X, INCX
+    fabs.s t1, t1
+    fabs.s t2, t2
+    fabs.s t3, t3
+    fabs.s t4, t4
+    addi.d I, I, -1
+    fadd.s t1, t1, t2
+    fadd.s t3, t3, t4
+    fmax.s s3, t1, t3
+    fld.s t1, X, 0 * SIZE
+    fld.s t2, X, 1 * SIZE
+    add.d X, X, INCX
+    fld.s t3, X, 0 * SIZE
+    fld.s t4, X, 1 * SIZE
+    add.d X, X, INCX
+    fabs.s t1, t1
+    fabs.s t2, t2
+    fabs.s t3, t3
+    fabs.s t4, t4
+    fadd.s t1, t1, t2
+    fadd.s t3, t3, t4
+    fmax.s s4, t1, t3
+    blt $r0, I, .L21
+    .align 3
+
+.L22:
+    fmax.s s1, s1, s2
+    fmax.s s3, s3, s4
+    fmax.s s1, s1, s3
+    .align 3
+
+.L23: //N<8
+    andi I, N, 7
+    bge $r0, I, .L999
+    .align 3
+
+.L24:
+    LD a0, X, 0 * SIZE
+    LD a1, X, 1 * SIZE
+    addi.d I, I, -1
+    FABS a0, a0
+    FABS a1, a1
+    ADD a0, a0, a1
+    add.d  X, X, INCX
+    fmax.s s1, a0, s1
+    blt $r0, I, .L24
+    .align 3
+
+.L999:
+    fmov.s $f0, $f22
+    jirl $r0, $r1, 0x0
+    .align 3
+
+    EPILOGUE