RISC-V Vector Extension

#include <riscv_vector.h>
#include <stdio.h>

void vec_add_rvv(int *a, int *b, int *c, size_t n) {
    size_t vl;
    vint32m2_t va, vb, vc;
    for (; vl = vsetvl_e32m2(n); n -= vl) {
        vb = vle32_v_i32m2(b, vl);
        vc = vle32_v_i32m2(c, vl);
        va = vadd_vv_i32m2(vb, vc, vl);
        vse32_v_i32m2(a, va, vl);
        a += vl;
        b += vl;
        c += vl;
    }
}

int x[10] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 0};
int y[10] = {0, 9, 8, 7, 6, 5, 4, 3, 2, 1};
int z[10];

int main() {
    int i;
    vec_add_rvv(z, x, y, 10);
    for (i = 0; i < 10; i++) printf("%d ", z[i]);
    printf("\n");
    return 0;
}

li t0, 10
vsetvli <rd>, <rs1>, e8, m1, ta, ma

int len = 1021;
int data[len] = {0};

/*
 * 32 类似于 VLMAX
 * max 相当于 vsetvli
 */

do {
    int seg = min(len, 32);
    /* process data[0..seg] */
    data += seg;
    len -= seg;
} while (len > 0)

## t1 是之前保存的 vtype
vsetivl t0, 10, t1
vsetivli t0, 10, e8, m2
## t2 保存着要求的 vl
vsetvl t0, t2, t1
vsetvli t0, t2, e8 ,2, ta, mu

## 不使用 mask
vop.v* v1, ...
## 使用 mask, 其中 mask 保存在 v0 中, 每 bit 代表一个 element 是否被 mask
vop.v* v1, ..., v0.t

    .global matmul
    .global main

    .text
    #define row_index s3
    #define col_index s4
    #define tmp t0
matmul:
    #define vlen s0
    #define a_offset s1
    #define b_offset s2
    #define remain s6
    li row_index, 0
.Louter:
    li col_index, 0
.Linner:
    mul tmp, row_index, a4
    slli a_offset, tmp, 2
    add a_offset, a_offset, a0

    slli b_offset, col_index, 2
    add b_offset, b_offset, a1

    mv vlen, a4
    mv remain, a4
.Lloop:
    vsetvli vlen, remain, e32
    ## load a
    vle32.v v0, (a_offset)
    slli tmp, vlen, 2
    add a_offset, a_offset, tmp
    ## load b
    slli tmp, a5, 2
    vlse32.v v1, (b_offset), tmp
    mul tmp, tmp, vlen
    add b_offset, b_offset, tmp
    ## calc
    sub remain, remain, vlen
    vmul.vv v0, v0, v1
    vredsum.vs v2, v0, v2
    bnez remain, .Lloop

    ## save result
    vmv.x.s tmp, v2
    vmv.s.x v2, zero
    sw tmp, (a2)
    addi a2, a2, 4
    addi col_index, col_index, 1
    blt col_index, a5, .Linner

    addi row_index, row_index, 1
    blt row_index, a3, .Louter
    ## finish
    ret


dump:
    addi sp, sp, -4
    sw ra, 4(sp)

    #define buf s0
    #define total_row s1
    #define total_col s2

    mv buf, a0
    mv total_row, a1
    mv total_col, a2
    mv row_index, zero
2:
    mv col_index, zero
1:
    la a0, fmt
    lw a1, (buf)
    addi buf, buf, 4
    call printf
    addi col_index, col_index, 1
    blt col_index, total_col, 1b

    la a0, newline
    call printf
    addi row_index, row_index, 1
    blt row_index, total_row, 2b

    lw ra, 4(sp)
    addi sp, sp, 4
    ret


main:
    addi sp, sp, -4
    sw ra, 4(sp)

    la a0, a
    la a1, b
    la a2, c
    li a3, 2
    li a4, 3
    li a5, 2

    call matmul

    la a0, c
    li a1, 2
    li a2, 2
    call dump

    lw ra, 4(sp)
    addi sp, sp, 4
    ret

    .data
a:
    .word 1
    .word 2
    .word 3
    .word 4
    .word 5
    .word 6

b:
    .word 1
    .word 2
    .word 3
    .word 4
    .word 5
    .word 6

c:
    .word 0
    .word 0
    .word 0
    .word 0

fmt:
    .asciz "%d "

newline:
    .asciz "\n"

// 2022-05-06 09:49
#include <riscv_vector.h>
#include <stdio.h>

/* c=a@b, a.shape=[x,y] b.shape=[y,z], c.shape=[x,z] */
void matmul(int *a, int *b, int *c, int x, int y, int z) {
    int tmp[] = {0, 0, 0};
    for (int i = 0; i < x; i++) {
        for (int j = 0; j < z; j++) {
            int *a_offset = a + i * y;
            int *b_offset = b + j;
            vint32m1_t accu = vmv_v_x_i32m1(0, 1);
            int remain = y;
            while (remain != 0) {
                int vlen = vsetvl_e32m1(remain);
                vint32m1_t aa = vle32_v_i32m1(a_offset, vlen);
                vint32m1_t bb = vlse32_v_i32m1(b_offset, z * 4, vlen);
                vint32m1_t cc = vmul_vv_i32m1(aa, bb, vlen);
                accu = vredsum_vs_i32m1_i32m1(accu, cc, accu, vlen);
                a_offset += vlen;
                b_offset += vlen * z;
                remain -= vlen;
            }
            int result = vmv_x_s_i32m1_i32(accu);
            *(c++) = result;
        }
    }
}

int main(int argc, char *argv[]) {
    int a[] = {1, 2, 3, 4, 5, 6};
    int b[] = {1, 2, 3, 4, 5, 6};
    int c[] = {0, 0, 0, 0};

    matmul(a, b, c, 2, 3, 2);

    for (int i = 0; i < 2; i++) {
        for (int j = 0; j < 2; j++) {
            printf("%d ", c[i * 2 + j]);
        }
        printf("\n");
    }
}