loongarch64: Add optimizations for scal.

Author: yancheng

kernel/loongarch64/KERNEL.LOONGSON2K1000

@@ -4,4 +4,7 @@ SDOTKERNEL  = dot_lsx.S
 DSDOTKERNEL = dot_lsx.S
 DDOTKERNEL  = dot_lsx.S
 
+SSCALKERNEL  = sscal_lsx.S
+DSCALKERNEL  = dscal_lsx.S
+
 endif

kernel/loongarch64/KERNEL.LOONGSON3R5

@@ -4,6 +4,9 @@ SDOTKERNEL  = dot_lasx.S
 DSDOTKERNEL = dot_lasx.S
 DDOTKERNEL  = dot_lasx.S
 
+SSCALKERNEL  = sscal_lasx.S
+DSCALKERNEL  = dscal_lasx.S
+
 DGEMMKERNEL    = dgemm_kernel_16x4.S
 DGEMMINCOPY    = dgemm_ncopy_16.S
 DGEMMITCOPY    = dgemm_tcopy_16.S

kernel/loongarch64/dscal_lasx.S

@@ -0,0 +1,194 @@
+#define ASSEMBLER
+#include "common.h"
+
+#define N $r4
+#define ALPHA $f0
+#define X  $r7
+#define INCX $r8
+#define I $r12
+#define TEMP $r13
+#define t1 $r14
+#define t2 $r18
+#define t3 $r15
+#define t4 $r17
+#define XX $r16
+#define VX0 $xr12
+#define VX1 $xr13
+#define VT0 $xr14
+#define VT1 $xr15
+#define VALPHA $xr19
+#define a1 $f8
+#define a2 $f23
+
+    PROLOGUE
+
+    bge $r0, N, .L999
+    bge $r0, INCX, .L999
+    li.d TEMP, 1
+    movgr2fr.d a1, $r0
+    ffint.d.l a1, a1
+    movgr2fr.d a2, TEMP
+    ffint.d.l a2, a2
+    slli.d TEMP, TEMP, BASE_SHIFT
+    slli.d INCX, INCX, BASE_SHIFT
+    fcmp.ceq.d $fcc0, ALPHA, a1
+    bcnez $fcc0, .L20   //ALPHA==0
+    fcmp.ceq.d $fcc0, ALPHA, a2
+    bcnez $fcc0, .L999  //ALPHA==1 return
+    srai.d I, N, 3
+    beq INCX, TEMP, .L30    //ALPHA！=0|1 and INCX==1
+    movfr2gr.d TEMP, ALPHA
+    xvreplgr2vr.d VALPHA, TEMP
+    move XX, X
+    .align 3
+
+.L10:
+    bge $r0, I, .L32
+    .align 3
+.L11:
+    ld.d t1, X, 0 * SIZE
+    add.d X, X, INCX
+    ld.d t2, X, 0 * SIZE
+    add.d X, X, INCX
+    ld.d t3, X, 0 * SIZE
+    add.d X, X, INCX
+    ld.d t4, X, 0 * SIZE
+    add.d X, X, INCX
+    xvinsgr2vr.d VX0, t1, 0
+    xvinsgr2vr.d VX0, t2, 1
+    xvinsgr2vr.d VX0, t3, 2
+    xvinsgr2vr.d VX0, t4, 3
+    ld.d t1, X, 0 * SIZE
+    add.d X, X, INCX
+    ld.d t2, X, 0 * SIZE
+    add.d X, X, INCX
+    xvfmul.d VT0, VX0, VALPHA
+    ld.d t3, X, 0 * SIZE
+    add.d X, X, INCX
+    ld.d t4, X, 0 * SIZE
+    add.d X, X, INCX
+    xvinsgr2vr.d VX1, t1, 0
+    xvinsgr2vr.d VX1, t2, 1
+    xvinsgr2vr.d VX1, t3, 2
+    xvinsgr2vr.d VX1, t4, 3
+    xvstelm.d VT0, XX, 0, 0
+    add.d XX, XX, INCX
+    xvstelm.d VT0, XX, 0, 1
+    add.d XX, XX, INCX
+    xvstelm.d VT0, XX, 0, 2
+    add.d XX, XX, INCX
+    xvstelm.d VT0, XX, 0, 3
+    add.d XX, XX, INCX
+    xvfmul.d VT1, VX1, VALPHA
+    xvstelm.d VT1, XX, 0, 0
+    add.d XX, XX, INCX
+    xvstelm.d VT1, XX, 0, 1
+    add.d XX, XX, INCX
+    xvstelm.d VT1, XX, 0, 2
+    add.d XX, XX, INCX
+    xvstelm.d VT1, XX, 0, 3
+    add.d XX, XX, INCX
+    addi.d  I, I, -1
+    blt $r0, I, .L11
+    b .L32
+    .align 3
+
+.L20:
+    srai.d I, N, 3
+    beq INCX, TEMP, .L24
+    bge $r0, I, .L22
+    .align 3
+
+.L21:
+    fst.d a1, X, 0
+    add.d X, X, INCX
+    fst.d a1, X, 0
+    add.d X, X, INCX
+    fst.d a1, X, 0
+    add.d X, X, INCX
+    fst.d a1, X, 0
+    add.d X, X, INCX
+    fst.d a1, X, 0
+    add.d X, X, INCX
+    fst.d a1, X, 0
+    add.d X, X, INCX
+    fst.d a1, X, 0
+    add.d X, X, INCX
+    fst.d a1, X, 0
+    add.d X, X, INCX
+    addi.d  I, I, -1
+    blt $r0, I, .L21
+    .align 3
+
+.L22:
+    andi I, N, 7
+    bge $r0, I, .L999
+    .align 3
+.L23:
+    fst.d a1, X, 0 * SIZE
+    addi.d I, I, -1
+    add.d X, X, INCX
+    blt $r0, I, .L23
+    jirl $r0, $r1, 0
+    .align 3
+
+.L24:
+    bge $r0, I, .L26 /*N<8 INCX==1*/
+    .align 3
+.L25:
+    xvxor.v VX0, VX0, VX0
+    xvst VX0, X, 0 * SIZE
+    xvst VX0, X, 4 * SIZE
+    addi.d I, I, -1
+    addi.d X, X, 8 * SIZE
+    blt $r0, I, .L25
+    .align 3
+
+.L26:
+    andi I, N, 7
+    bge $r0, I, .L999
+    .align 3
+.L27:
+    fst.d a1, X, 0 * SIZE
+    addi.d I, I, -1
+    addi.d X, X, SIZE
+    blt $r0, I, .L27
+    jirl $r0, $r1, 0
+    .align 3
+
+.L30:
+    bge $r0, I, .L32/*N<8 INCX==1*/
+    movfr2gr.d TEMP, ALPHA
+    xvreplgr2vr.d VALPHA , TEMP
+    .align 3
+
+.L31:
+    xvld VX0, X, 0 * SIZE
+    xvld VX1, X, 4 * SIZE
+    xvfmul.d VT0, VX0, VALPHA
+    xvfmul.d VT1, VX1, VALPHA
+    addi.d I, I, -1
+    xvst VT0, X, 0 * SIZE
+    xvst VT1, X, 4 * SIZE
+    addi.d X, X, 8 * SIZE
+    blt $r0, I, .L31
+    .align 3
+
+.L32:
+    andi I, N, 7
+    bge $r0, I, .L999
+    .align 3
+.L33:
+    fld.d a1, X, 0 * SIZE
+    addi.d I, I, -1
+    fmul.d a1, ALPHA, a1
+    fst.d a1, X, 0 * SIZE
+    add.d X, X, INCX
+    blt $r0, I, .L33
+    jirl $r0, $r1, 0
+    .align 3
+
+.L999:
+    jirl    $r0, $r1, 0x0
+
+    EPILOGUE