loongarch64: Add optimizations for nrm2.

Author: yancheng

kernel/loongarch64/KERNEL.LOONGSON2K1000

@@ -52,4 +52,7 @@ DASUMKERNEL =  dasum_lsx.S
 SROTKERNEL  =  srot_lsx.S
 DROTKERNEL  =  drot_lsx.S
 
+SNRM2KERNEL =  snrm2_lsx.S
+DNRM2KERNEL =  dnrm2_lsx.S
+
 endif

kernel/loongarch64/KERNEL.LOONGSON3R5

@@ -52,6 +52,9 @@ DASUMKERNEL =  dasum_lasx.S
 SROTKERNEL  =  srot_lasx.S
 DROTKERNEL  =  drot_lasx.S
 
+SNRM2KERNEL =  snrm2_lasx.S
+DNRM2KERNEL =  dnrm2_lasx.S
+
 DGEMMKERNEL    = dgemm_kernel_16x4.S
 DGEMMINCOPY    = dgemm_ncopy_16.S
 DGEMMITCOPY    = dgemm_tcopy_16.S

kernel/loongarch64/dnrm2_lasx.S

@@ -0,0 +1,233 @@
+#define ASSEMBLER
+
+#include "common.h"
+
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define XX     $r19
+#define I      $r17
+#define TEMP   $r18
+#define t1     $r12
+#define t2     $r13
+#define t3     $r14
+#define t4     $r15
+#define VX0    $xr15
+#define VX1    $xr16
+#define VM0    $xr17
+#define VM1    $xr18
+#define VM2    $xr13
+#define VM3    $xr14
+#define res1   $xr19
+#define res2   $xr20
+#define VALPHA $xr21
+#define INF    $f23
+#define a1     $f22
+#define max    $f17
+#define ALPHA  $f12
+
+    PROLOGUE
+
+#ifdef F_INTERFACE
+    LDINT   N,     0(N)
+    LDINT   INCX,  0(INCX)
+#endif
+
+    xvxor.v   res1, res1, res1
+    xvxor.v   res2, res2, res2
+    bge $r0, N, .L999
+    beq $r0, INCX, .L999
+    move  XX, X
+    // Init INF
+    addi.d  TEMP, $r0,  0x7FF
+    slli.d  TEMP, TEMP, 52
+    MTC  INF, TEMP
+    li.d  TEMP, SIZE
+    slli.d INCX, INCX, BASE_SHIFT
+    srai.d I, N, 3
+    bne INCX, TEMP, .L20
+    xvld VM0, X, 0
+    bge $r0,  I, .L97
+    .align 3
+
+.L10:
+    xvld VX0, X, 0 * SIZE
+    xvld VX1, X, 4 * SIZE
+    xvfmaxa.d VM1, VX1, VX0
+    xvfmaxa.d VM0, VM0, VM1
+    addi.d I, I, -1
+    addi.d X, X, 8 * SIZE
+    blt $r0, I, .L10
+    b .L96
+    .align 3
+
+.L20: // INCX!=1
+    move TEMP, X // initialize the maxa value
+    ld.d t1, TEMP, 0 * SIZE
+    add.d TEMP, TEMP, INCX
+    xvinsgr2vr.d VM0, t1, 0
+    srai.d I, N, 3
+    bge $r0, I, .L97
+    ld.d t2, TEMP, 0 * SIZE
+    add.d TEMP, TEMP, INCX
+    xvinsgr2vr.d VM0, t2, 1
+    .align 3
+
+.L21:
+    ld.d t1, X, 0 * SIZE
+    add.d X, X, INCX
+    xvinsgr2vr.d VX0, t1, 0
+    ld.d t2, X, 0 * SIZE
+    add.d X, X, INCX
+    xvinsgr2vr.d VX0, t2, 1
+    ld.d t3, X, 0 * SIZE
+    add.d X, X, INCX
+    xvinsgr2vr.d VX0, t3, 2
+    ld.d t4, X, 0 * SIZE
+    add.d X, X, INCX
+    xvinsgr2vr.d VX0, t4, 3
+    ld.d t1, X, 0 * SIZE
+    add.d X, X, INCX
+    xvinsgr2vr.d VX1, t1, 0
+    ld.d t2, X, 0 * SIZE
+    add.d X, X, INCX
+    xvinsgr2vr.d VX1, t2, 1
+    ld.d t3, X, 0 * SIZE
+    add.d X, X, INCX
+    xvinsgr2vr.d VX1, t3, 2
+    ld.d t4, X, 0 * SIZE
+    add.d X, X, INCX
+    xvinsgr2vr.d VX1, t4, 3
+    xvfmaxa.d VM1, VX0, VX1
+    xvfmaxa.d VM0, VM0, VM1
+    addi.d I, I, -1
+    blt $r0, I, .L21
+    b .L96
+    .align 3
+
+.L96:
+    xvpickve.d VX0, VM0, 1
+    xvpickve.d VX1, VM0, 2
+    xvpickve.d VM3, VM0, 3
+    xvfmaxa.d VM1, VX0, VX1
+    xvfmaxa.d VM2, VM3, VM0
+    xvfmaxa.d VM0, VM1, VM2
+    .align 3
+
+.L97:
+    andi I, N, 7
+    bge $r0, I, .L99
+    .align 3
+
+.L98:
+    xvld VX1, X, 0
+    xvfmaxa.d VM0, VM0, VX1
+    addi.d I, I, -1
+    add.d X, X, INCX
+    blt $r0, I, .L98
+    .align 3
+
+.L99:
+    fabs.d  max, max
+    lu12i.w  TEMP, 0x3f800 // 1
+    movgr2fr.d  a1, $r0
+    movgr2fr.w  ALPHA, TEMP
+    CMPEQ  $fcc0, max, a1
+    fcvt.d.s  ALPHA, ALPHA
+    bcnez  $fcc0, .L999
+    fdiv.d  ALPHA, ALPHA, max
+    CMPEQ  $fcc0, INF, ALPHA
+    bcnez  $fcc0, .L999
+    movfr2gr.d TEMP, ALPHA
+    xvreplgr2vr.d VALPHA, TEMP
+
+.L100:
+    li.d TEMP, SIZE
+    bne INCX, TEMP, .L120
+    srai.d  I, N, 3
+    bge $r0, I, .L997
+    .align 3
+
+.L110:
+    xvld VX0, XX, 0 * SIZE
+    xvld VX1, XX, 4 * SIZE
+    xvfmul.d VM0, VX0, VALPHA
+    xvfmul.d VM1, VX1, VALPHA
+    xvfmadd.d res1, VM0, VM0, res1
+    xvfmadd.d res2, VM1, VM1, res2
+    addi.d XX, XX, 8 * SIZE
+    addi.d I, I, -1
+    blt $r0, I, .L110
+    b .L996
+    .align 3
+
+.L120:
+    srai.d I, N, 3
+    bge $r0, I, .L997
+
+.L121:
+    ld.d t1, XX,  0 * SIZE
+    add.d XX, XX, INCX
+    ld.d t2, XX, 0 * SIZE
+    add.d XX, XX, INCX
+    ld.d t3, XX, 0 * SIZE
+    add.d XX, XX, INCX
+    ld.d t4, XX, 0 * SIZE
+    add.d XX, XX, INCX
+    xvinsgr2vr.d VX0, t1, 0
+    xvinsgr2vr.d VX0, t2, 1
+    xvinsgr2vr.d VX0, t3, 2
+    xvinsgr2vr.d VX0, t4, 3
+    ld.d t1, XX, 0 * SIZE
+    add.d XX, XX, INCX
+    ld.d t2, XX, 0 * SIZE
+    add.d XX, XX, INCX
+    ld.d t3, XX, 0 * SIZE
+    add.d XX, XX, INCX
+    ld.d t4, XX, 0 * SIZE
+    add.d XX, XX, INCX
+    xvinsgr2vr.d VX0, t1, 0
+    xvinsgr2vr.d VX0, t2, 1
+    xvinsgr2vr.d VX1, t3, 2
+    xvinsgr2vr.d VX1, t4, 3
+    xvfmul.d VM0, VX0, VALPHA
+    xvfmul.d VM1, VX1, VALPHA
+    xvfmadd.d res1, VM0, VM0, res1
+    xvfmadd.d res2, VM1, VM1, res2
+    addi.d  I, I, -1
+    blt $r0, I, .L121
+    b .L996
+    .align 3
+
+.L996:
+    xvfadd.d res1, res1, res2
+    xvpickve.d VX0, res1, 1
+    xvpickve.d VX1, res1, 2
+    xvpickve.d VM0, res1, 3
+    xvfadd.d res1, VX0, res1
+    xvfadd.d VX1, VX1, VM0
+    xvfadd.d res1, VX1, res1
+    .align 3
+
+.L997:
+    andi I, N, 7
+    bge $r0, I, .L999
+    .align 3
+
+.L998:
+    fld.d $f15, XX, 0 * SIZE
+    addi.d I, I, -1
+    fmul.d $f15, $f15, ALPHA
+    fmadd.d $f19, $f15, $f15, $f19
+    add.d XX, XX , INCX
+    blt $r0, I, .L998
+    fsqrt.d $f19, $f19
+    fmul.d $f0, max, $f19
+    jirl    $r0, $r1, 0x0
+    .align 3
+
+.L999:
+    fmov.d $f0, $f19
+    jirl $r0, $r1, 0x0
+
+    EPILOGUE