/*

AUTOGENERATED KERNEL
Settings:
 LMUL=1
 M=16
 M_tail_scalar_from=2
 N=8
 __riscv_='__riscv_'
 complex=False
 conjugate=False
 cpu='zvl256b'
 force_acc_double=False
 index_type='BLASLONG'
 op='gemm'
 param_precision='float'
 reg_width_bits=256
 tail_policy=''
 trace=False

Derived:
 ELEN_ACC=32
 ELEN_PARAM=32
 LMUL_ACC=1
 VFMACC='__riscv_vfmacc_vf_f32m1'
 VFMUL='__riscv_vfmul_vf_f32m1'
 VLEV='__riscv_vle32_v_f32m1'
 VLSEV='__riscv_vlse32_v_f32m1'
 VMACC_TO_ACC='__riscv_vfmacc_vf_f32m1'
 VMUL_TO_ACC='__riscv_vfmul_vf_f32m1'
 VSETVL='__riscv_vsetvl_e32m1'
 VSEV='__riscv_vse32_v_f32m1'
 VSSEV='__riscv_vsse32_v_f32m1'
 acc_vector_t='vfloat32m1_t'
 output='sgemm_kernel_16x8_zvl256b.c'
 param_scalar_t='float'
 param_vector_t='vfloat32m1_t'

*/

#include "common.h"


int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, FLOAT* A, FLOAT* B, FLOAT* C, BLASLONG ldc)

{
    BLASLONG gvl = 0;
    BLASLONG m_top = 0;
    BLASLONG n_top = 0;


    // -- MAIN PASS

    for (BLASLONG j=0; j<N/8; j+=1) {
        m_top = 0;
        BLASLONG gvl = __riscv_vsetvl_e32m1(8);


        for (BLASLONG i=0; i<M/16; i+=1) {
            BLASLONG ai=m_top*K;
            BLASLONG bi=n_top*K;
            float B0 = B[bi+0];
            float B1 = B[bi+1];
            float B2 = B[bi+2];
            float B3 = B[bi+3];
            float B4 = B[bi+4];
            float B5 = B[bi+5];
            float B6 = B[bi+6];
            float B7 = B[bi+7];
            bi += 8;

            vfloat32m1_t A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
            vfloat32m1_t A1 = __riscv_vle32_v_f32m1( &A[ai+1*gvl], gvl );
            ai += 16;

            vfloat32m1_t result0 = __riscv_vfmul_vf_f32m1( A0, B0, gvl);
            vfloat32m1_t result1 = __riscv_vfmul_vf_f32m1( A1, B0, gvl);
            vfloat32m1_t result2 = __riscv_vfmul_vf_f32m1( A0, B1, gvl);
            vfloat32m1_t result3 = __riscv_vfmul_vf_f32m1( A1, B1, gvl);
            vfloat32m1_t result4 = __riscv_vfmul_vf_f32m1( A0, B2, gvl);
            vfloat32m1_t result5 = __riscv_vfmul_vf_f32m1( A1, B2, gvl);
            vfloat32m1_t result6 = __riscv_vfmul_vf_f32m1( A0, B3, gvl);
            vfloat32m1_t result7 = __riscv_vfmul_vf_f32m1( A1, B3, gvl);
            vfloat32m1_t result8 = __riscv_vfmul_vf_f32m1( A0, B4, gvl);
            vfloat32m1_t result9 = __riscv_vfmul_vf_f32m1( A1, B4, gvl);
            vfloat32m1_t result10 = __riscv_vfmul_vf_f32m1( A0, B5, gvl);
            vfloat32m1_t result11 = __riscv_vfmul_vf_f32m1( A1, B5, gvl);
            vfloat32m1_t result12 = __riscv_vfmul_vf_f32m1( A0, B6, gvl);
            vfloat32m1_t result13 = __riscv_vfmul_vf_f32m1( A1, B6, gvl);
            vfloat32m1_t result14 = __riscv_vfmul_vf_f32m1( A0, B7, gvl);
            vfloat32m1_t result15 = __riscv_vfmul_vf_f32m1( A1, B7, gvl);

            for(BLASLONG k=1; k<K; k++) {
                B0 = B[bi+0];
                B1 = B[bi+1];
                B2 = B[bi+2];
                B3 = B[bi+3];
                B4 = B[bi+4];
                B5 = B[bi+5];
                B6 = B[bi+6];
                B7 = B[bi+7];
                bi += 8;

                A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
                A1 = __riscv_vle32_v_f32m1( &A[ai+1*gvl], gvl );
                ai += 16;

                result0 = __riscv_vfmacc_vf_f32m1( result0, B0, A0, gvl);
                result1 = __riscv_vfmacc_vf_f32m1( result1, B0, A1, gvl);
                result2 = __riscv_vfmacc_vf_f32m1( result2, B1, A0, gvl);
                result3 = __riscv_vfmacc_vf_f32m1( result3, B1, A1, gvl);
                result4 = __riscv_vfmacc_vf_f32m1( result4, B2, A0, gvl);
                result5 = __riscv_vfmacc_vf_f32m1( result5, B2, A1, gvl);
                result6 = __riscv_vfmacc_vf_f32m1( result6, B3, A0, gvl);
                result7 = __riscv_vfmacc_vf_f32m1( result7, B3, A1, gvl);
                result8 = __riscv_vfmacc_vf_f32m1( result8, B4, A0, gvl);
                result9 = __riscv_vfmacc_vf_f32m1( result9, B4, A1, gvl);
                result10 = __riscv_vfmacc_vf_f32m1( result10, B5, A0, gvl);
                result11 = __riscv_vfmacc_vf_f32m1( result11, B5, A1, gvl);
                result12 = __riscv_vfmacc_vf_f32m1( result12, B6, A0, gvl);
                result13 = __riscv_vfmacc_vf_f32m1( result13, B6, A1, gvl);
                result14 = __riscv_vfmacc_vf_f32m1( result14, B7, A0, gvl);
                result15 = __riscv_vfmacc_vf_f32m1( result15, B7, A1, gvl);
            }


            BLASLONG ci=n_top*ldc+m_top;

            vfloat32m1_t c0 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += gvl;
            vfloat32m1_t c1 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*1;
            vfloat32m1_t c2 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += gvl;
            vfloat32m1_t c3 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*1;
            vfloat32m1_t c4 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += gvl;
            vfloat32m1_t c5 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*1;
            vfloat32m1_t c6 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += gvl;
            vfloat32m1_t c7 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*1;
            vfloat32m1_t c8 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += gvl;
            vfloat32m1_t c9 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*1;
            vfloat32m1_t c10 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += gvl;
            vfloat32m1_t c11 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*1;
            vfloat32m1_t c12 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += gvl;
            vfloat32m1_t c13 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*1;
            vfloat32m1_t c14 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += gvl;
            vfloat32m1_t c15 = __riscv_vle32_v_f32m1( &C[ci], gvl);
            c0 = __riscv_vfmacc_vf_f32m1( c0, alpha, result0, gvl );
            c1 = __riscv_vfmacc_vf_f32m1( c1, alpha, result1, gvl );
            c2 = __riscv_vfmacc_vf_f32m1( c2, alpha, result2, gvl );
            c3 = __riscv_vfmacc_vf_f32m1( c3, alpha, result3, gvl );
            c4 = __riscv_vfmacc_vf_f32m1( c4, alpha, result4, gvl );
            c5 = __riscv_vfmacc_vf_f32m1( c5, alpha, result5, gvl );
            c6 = __riscv_vfmacc_vf_f32m1( c6, alpha, result6, gvl );
            c7 = __riscv_vfmacc_vf_f32m1( c7, alpha, result7, gvl );
            c8 = __riscv_vfmacc_vf_f32m1( c8, alpha, result8, gvl );
            c9 = __riscv_vfmacc_vf_f32m1( c9, alpha, result9, gvl );
            c10 = __riscv_vfmacc_vf_f32m1( c10, alpha, result10, gvl );
            c11 = __riscv_vfmacc_vf_f32m1( c11, alpha, result11, gvl );
            c12 = __riscv_vfmacc_vf_f32m1( c12, alpha, result12, gvl );
            c13 = __riscv_vfmacc_vf_f32m1( c13, alpha, result13, gvl );
            c14 = __riscv_vfmacc_vf_f32m1( c14, alpha, result14, gvl );
            c15 = __riscv_vfmacc_vf_f32m1( c15, alpha, result15, gvl );

            ci=n_top*ldc+m_top;

            __riscv_vse32_v_f32m1( &C[ci], c0, gvl); ci += gvl;
            __riscv_vse32_v_f32m1( &C[ci], c1, gvl); ci += ldc-gvl*1;
            __riscv_vse32_v_f32m1( &C[ci], c2, gvl); ci += gvl;
            __riscv_vse32_v_f32m1( &C[ci], c3, gvl); ci += ldc-gvl*1;
            __riscv_vse32_v_f32m1( &C[ci], c4, gvl); ci += gvl;
            __riscv_vse32_v_f32m1( &C[ci], c5, gvl); ci += ldc-gvl*1;
            __riscv_vse32_v_f32m1( &C[ci], c6, gvl); ci += gvl;
            __riscv_vse32_v_f32m1( &C[ci], c7, gvl); ci += ldc-gvl*1;
            __riscv_vse32_v_f32m1( &C[ci], c8, gvl); ci += gvl;
            __riscv_vse32_v_f32m1( &C[ci], c9, gvl); ci += ldc-gvl*1;
            __riscv_vse32_v_f32m1( &C[ci], c10, gvl); ci += gvl;
            __riscv_vse32_v_f32m1( &C[ci], c11, gvl); ci += ldc-gvl*1;
            __riscv_vse32_v_f32m1( &C[ci], c12, gvl); ci += gvl;
            __riscv_vse32_v_f32m1( &C[ci], c13, gvl); ci += ldc-gvl*1;
            __riscv_vse32_v_f32m1( &C[ci], c14, gvl); ci += gvl;
            __riscv_vse32_v_f32m1( &C[ci], c15, gvl);
            m_top += 16;
        }



        // -- tails for main pass

        if( M & 8 ) {
            gvl = __riscv_vsetvl_e32m1(8);

            BLASLONG ai=m_top*K;
            BLASLONG bi=n_top*K;
            float B0 = B[bi+0];
            float B1 = B[bi+1];
            float B2 = B[bi+2];
            float B3 = B[bi+3];
            float B4 = B[bi+4];
            float B5 = B[bi+5];
            float B6 = B[bi+6];
            float B7 = B[bi+7];
            bi += 8;

            vfloat32m1_t A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
            ai += 8;

            vfloat32m1_t result0 = __riscv_vfmul_vf_f32m1( A0, B0, gvl);
            vfloat32m1_t result1 = __riscv_vfmul_vf_f32m1( A0, B1, gvl);
            vfloat32m1_t result2 = __riscv_vfmul_vf_f32m1( A0, B2, gvl);
            vfloat32m1_t result3 = __riscv_vfmul_vf_f32m1( A0, B3, gvl);
            vfloat32m1_t result4 = __riscv_vfmul_vf_f32m1( A0, B4, gvl);
            vfloat32m1_t result5 = __riscv_vfmul_vf_f32m1( A0, B5, gvl);
            vfloat32m1_t result6 = __riscv_vfmul_vf_f32m1( A0, B6, gvl);
            vfloat32m1_t result7 = __riscv_vfmul_vf_f32m1( A0, B7, gvl);

            for(BLASLONG k=1; k<K; k++) {
                B0 = B[bi+0];
                B1 = B[bi+1];
                B2 = B[bi+2];
                B3 = B[bi+3];
                B4 = B[bi+4];
                B5 = B[bi+5];
                B6 = B[bi+6];
                B7 = B[bi+7];
                bi += 8;

                A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
                ai += 8;

                result0 = __riscv_vfmacc_vf_f32m1( result0, B0, A0, gvl);
                result1 = __riscv_vfmacc_vf_f32m1( result1, B1, A0, gvl);
                result2 = __riscv_vfmacc_vf_f32m1( result2, B2, A0, gvl);
                result3 = __riscv_vfmacc_vf_f32m1( result3, B3, A0, gvl);
                result4 = __riscv_vfmacc_vf_f32m1( result4, B4, A0, gvl);
                result5 = __riscv_vfmacc_vf_f32m1( result5, B5, A0, gvl);
                result6 = __riscv_vfmacc_vf_f32m1( result6, B6, A0, gvl);
                result7 = __riscv_vfmacc_vf_f32m1( result7, B7, A0, gvl);
            }


            BLASLONG ci=n_top*ldc+m_top;

            vfloat32m1_t c0 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c1 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c2 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c3 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c4 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c5 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c6 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c7 = __riscv_vle32_v_f32m1( &C[ci], gvl);
            c0 = __riscv_vfmacc_vf_f32m1( c0, alpha, result0, gvl );
            c1 = __riscv_vfmacc_vf_f32m1( c1, alpha, result1, gvl );
            c2 = __riscv_vfmacc_vf_f32m1( c2, alpha, result2, gvl );
            c3 = __riscv_vfmacc_vf_f32m1( c3, alpha, result3, gvl );
            c4 = __riscv_vfmacc_vf_f32m1( c4, alpha, result4, gvl );
            c5 = __riscv_vfmacc_vf_f32m1( c5, alpha, result5, gvl );
            c6 = __riscv_vfmacc_vf_f32m1( c6, alpha, result6, gvl );
            c7 = __riscv_vfmacc_vf_f32m1( c7, alpha, result7, gvl );

            ci=n_top*ldc+m_top;

            __riscv_vse32_v_f32m1( &C[ci], c0, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c1, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c2, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c3, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c4, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c5, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c6, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c7, gvl);
            m_top += 8;
        }


        if( M & 4 ) {
            gvl = __riscv_vsetvl_e32m1(4);

            BLASLONG ai=m_top*K;
            BLASLONG bi=n_top*K;
            float B0 = B[bi+0];
            float B1 = B[bi+1];
            float B2 = B[bi+2];
            float B3 = B[bi+3];
            float B4 = B[bi+4];
            float B5 = B[bi+5];
            float B6 = B[bi+6];
            float B7 = B[bi+7];
            bi += 8;

            vfloat32m1_t A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
            ai += 4;

            vfloat32m1_t result0 = __riscv_vfmul_vf_f32m1( A0, B0, gvl);
            vfloat32m1_t result1 = __riscv_vfmul_vf_f32m1( A0, B1, gvl);
            vfloat32m1_t result2 = __riscv_vfmul_vf_f32m1( A0, B2, gvl);
            vfloat32m1_t result3 = __riscv_vfmul_vf_f32m1( A0, B3, gvl);
            vfloat32m1_t result4 = __riscv_vfmul_vf_f32m1( A0, B4, gvl);
            vfloat32m1_t result5 = __riscv_vfmul_vf_f32m1( A0, B5, gvl);
            vfloat32m1_t result6 = __riscv_vfmul_vf_f32m1( A0, B6, gvl);
            vfloat32m1_t result7 = __riscv_vfmul_vf_f32m1( A0, B7, gvl);

            for(BLASLONG k=1; k<K; k++) {
                B0 = B[bi+0];
                B1 = B[bi+1];
                B2 = B[bi+2];
                B3 = B[bi+3];
                B4 = B[bi+4];
                B5 = B[bi+5];
                B6 = B[bi+6];
                B7 = B[bi+7];
                bi += 8;

                A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
                ai += 4;

                result0 = __riscv_vfmacc_vf_f32m1( result0, B0, A0, gvl);
                result1 = __riscv_vfmacc_vf_f32m1( result1, B1, A0, gvl);
                result2 = __riscv_vfmacc_vf_f32m1( result2, B2, A0, gvl);
                result3 = __riscv_vfmacc_vf_f32m1( result3, B3, A0, gvl);
                result4 = __riscv_vfmacc_vf_f32m1( result4, B4, A0, gvl);
                result5 = __riscv_vfmacc_vf_f32m1( result5, B5, A0, gvl);
                result6 = __riscv_vfmacc_vf_f32m1( result6, B6, A0, gvl);
                result7 = __riscv_vfmacc_vf_f32m1( result7, B7, A0, gvl);
            }


            BLASLONG ci=n_top*ldc+m_top;

            vfloat32m1_t c0 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c1 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c2 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c3 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c4 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c5 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c6 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c7 = __riscv_vle32_v_f32m1( &C[ci], gvl);
            c0 = __riscv_vfmacc_vf_f32m1( c0, alpha, result0, gvl );
            c1 = __riscv_vfmacc_vf_f32m1( c1, alpha, result1, gvl );
            c2 = __riscv_vfmacc_vf_f32m1( c2, alpha, result2, gvl );
            c3 = __riscv_vfmacc_vf_f32m1( c3, alpha, result3, gvl );
            c4 = __riscv_vfmacc_vf_f32m1( c4, alpha, result4, gvl );
            c5 = __riscv_vfmacc_vf_f32m1( c5, alpha, result5, gvl );
            c6 = __riscv_vfmacc_vf_f32m1( c6, alpha, result6, gvl );
            c7 = __riscv_vfmacc_vf_f32m1( c7, alpha, result7, gvl );

            ci=n_top*ldc+m_top;

            __riscv_vse32_v_f32m1( &C[ci], c0, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c1, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c2, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c3, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c4, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c5, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c6, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c7, gvl);
            m_top += 4;
        }


        if( M & 2 ) {
            float result0 = 0;
            float result1 = 0;
            float result2 = 0;
            float result3 = 0;
            float result4 = 0;
            float result5 = 0;
            float result6 = 0;
            float result7 = 0;
            float result8 = 0;
            float result9 = 0;
            float result10 = 0;
            float result11 = 0;
            float result12 = 0;
            float result13 = 0;
            float result14 = 0;
            float result15 = 0;
            BLASLONG ai=m_top*K;
            BLASLONG bi=n_top*K;

            for(BLASLONG k=0; k<K; k++) {
                result0+=A[ai+0]*B[bi+0];
                result1+=A[ai+1]*B[bi+0];
                result2+=A[ai+0]*B[bi+1];
                result3+=A[ai+1]*B[bi+1];
                result4+=A[ai+0]*B[bi+2];
                result5+=A[ai+1]*B[bi+2];
                result6+=A[ai+0]*B[bi+3];
                result7+=A[ai+1]*B[bi+3];
                result8+=A[ai+0]*B[bi+4];
                result9+=A[ai+1]*B[bi+4];
                result10+=A[ai+0]*B[bi+5];
                result11+=A[ai+1]*B[bi+5];
                result12+=A[ai+0]*B[bi+6];
                result13+=A[ai+1]*B[bi+6];
                result14+=A[ai+0]*B[bi+7];
                result15+=A[ai+1]*B[bi+7];
                ai+=2;
                bi+=8;
            }

            BLASLONG ci=n_top*ldc+m_top;
            C[ci+0*ldc+0] += alpha * result0;
            C[ci+0*ldc+1] += alpha * result1;
            C[ci+1*ldc+0] += alpha * result2;
            C[ci+1*ldc+1] += alpha * result3;
            C[ci+2*ldc+0] += alpha * result4;
            C[ci+2*ldc+1] += alpha * result5;
            C[ci+3*ldc+0] += alpha * result6;
            C[ci+3*ldc+1] += alpha * result7;
            C[ci+4*ldc+0] += alpha * result8;
            C[ci+4*ldc+1] += alpha * result9;
            C[ci+5*ldc+0] += alpha * result10;
            C[ci+5*ldc+1] += alpha * result11;
            C[ci+6*ldc+0] += alpha * result12;
            C[ci+6*ldc+1] += alpha * result13;
            C[ci+7*ldc+0] += alpha * result14;
            C[ci+7*ldc+1] += alpha * result15;
            m_top+=2;
        }


        if( M & 1 ) {
            float result0 = 0;
            float result1 = 0;
            float result2 = 0;
            float result3 = 0;
            float result4 = 0;
            float result5 = 0;
            float result6 = 0;
            float result7 = 0;
            BLASLONG ai=m_top*K;
            BLASLONG bi=n_top*K;

            for(BLASLONG k=0; k<K; k++) {
                result0+=A[ai+0]*B[bi+0];
                result1+=A[ai+0]*B[bi+1];
                result2+=A[ai+0]*B[bi+2];
                result3+=A[ai+0]*B[bi+3];
                result4+=A[ai+0]*B[bi+4];
                result5+=A[ai+0]*B[bi+5];
                result6+=A[ai+0]*B[bi+6];
                result7+=A[ai+0]*B[bi+7];
                ai+=1;
                bi+=8;
            }

            BLASLONG ci=n_top*ldc+m_top;
            C[ci+0*ldc+0] += alpha * result0;
            C[ci+1*ldc+0] += alpha * result1;
            C[ci+2*ldc+0] += alpha * result2;
            C[ci+3*ldc+0] += alpha * result3;
            C[ci+4*ldc+0] += alpha * result4;
            C[ci+5*ldc+0] += alpha * result5;
            C[ci+6*ldc+0] += alpha * result6;
            C[ci+7*ldc+0] += alpha * result7;
            m_top+=1;
        }

        n_top += 8;
    }



    // -- tails for N=4

    if( N & 4 ) {
        gvl = __riscv_vsetvl_e32m1(8);
        m_top = 0;

        for (BLASLONG i=0; i<M/16; i+=1) {
            BLASLONG ai=m_top*K;
            BLASLONG bi=n_top*K;
            float B0 = B[bi+0];
            float B1 = B[bi+1];
            float B2 = B[bi+2];
            float B3 = B[bi+3];
            bi += 4;

            vfloat32m1_t A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
            vfloat32m1_t A1 = __riscv_vle32_v_f32m1( &A[ai+1*gvl], gvl );
            ai += 16;

            vfloat32m1_t result0 = __riscv_vfmul_vf_f32m1( A0, B0, gvl);
            vfloat32m1_t result1 = __riscv_vfmul_vf_f32m1( A1, B0, gvl);
            vfloat32m1_t result2 = __riscv_vfmul_vf_f32m1( A0, B1, gvl);
            vfloat32m1_t result3 = __riscv_vfmul_vf_f32m1( A1, B1, gvl);
            vfloat32m1_t result4 = __riscv_vfmul_vf_f32m1( A0, B2, gvl);
            vfloat32m1_t result5 = __riscv_vfmul_vf_f32m1( A1, B2, gvl);
            vfloat32m1_t result6 = __riscv_vfmul_vf_f32m1( A0, B3, gvl);
            vfloat32m1_t result7 = __riscv_vfmul_vf_f32m1( A1, B3, gvl);

            for(BLASLONG k=1; k<K; k++) {
                B0 = B[bi+0];
                B1 = B[bi+1];
                B2 = B[bi+2];
                B3 = B[bi+3];
                bi += 4;

                A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
                A1 = __riscv_vle32_v_f32m1( &A[ai+1*gvl], gvl );
                ai += 16;

                result0 = __riscv_vfmacc_vf_f32m1( result0, B0, A0, gvl);
                result1 = __riscv_vfmacc_vf_f32m1( result1, B0, A1, gvl);
                result2 = __riscv_vfmacc_vf_f32m1( result2, B1, A0, gvl);
                result3 = __riscv_vfmacc_vf_f32m1( result3, B1, A1, gvl);
                result4 = __riscv_vfmacc_vf_f32m1( result4, B2, A0, gvl);
                result5 = __riscv_vfmacc_vf_f32m1( result5, B2, A1, gvl);
                result6 = __riscv_vfmacc_vf_f32m1( result6, B3, A0, gvl);
                result7 = __riscv_vfmacc_vf_f32m1( result7, B3, A1, gvl);
            }


            BLASLONG ci=n_top*ldc+m_top;

            vfloat32m1_t c0 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += gvl;
            vfloat32m1_t c1 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*1;
            vfloat32m1_t c2 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += gvl;
            vfloat32m1_t c3 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*1;
            vfloat32m1_t c4 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += gvl;
            vfloat32m1_t c5 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*1;
            vfloat32m1_t c6 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += gvl;
            vfloat32m1_t c7 = __riscv_vle32_v_f32m1( &C[ci], gvl);
            c0 = __riscv_vfmacc_vf_f32m1( c0, alpha, result0, gvl );
            c1 = __riscv_vfmacc_vf_f32m1( c1, alpha, result1, gvl );
            c2 = __riscv_vfmacc_vf_f32m1( c2, alpha, result2, gvl );
            c3 = __riscv_vfmacc_vf_f32m1( c3, alpha, result3, gvl );
            c4 = __riscv_vfmacc_vf_f32m1( c4, alpha, result4, gvl );
            c5 = __riscv_vfmacc_vf_f32m1( c5, alpha, result5, gvl );
            c6 = __riscv_vfmacc_vf_f32m1( c6, alpha, result6, gvl );
            c7 = __riscv_vfmacc_vf_f32m1( c7, alpha, result7, gvl );

            ci=n_top*ldc+m_top;

            __riscv_vse32_v_f32m1( &C[ci], c0, gvl); ci += gvl;
            __riscv_vse32_v_f32m1( &C[ci], c1, gvl); ci += ldc-gvl*1;
            __riscv_vse32_v_f32m1( &C[ci], c2, gvl); ci += gvl;
            __riscv_vse32_v_f32m1( &C[ci], c3, gvl); ci += ldc-gvl*1;
            __riscv_vse32_v_f32m1( &C[ci], c4, gvl); ci += gvl;
            __riscv_vse32_v_f32m1( &C[ci], c5, gvl); ci += ldc-gvl*1;
            __riscv_vse32_v_f32m1( &C[ci], c6, gvl); ci += gvl;
            __riscv_vse32_v_f32m1( &C[ci], c7, gvl);
            m_top += 16;
        }


        if( M & 8 ) {
            gvl = __riscv_vsetvl_e32m1(8);

            BLASLONG ai=m_top*K;
            BLASLONG bi=n_top*K;
            float B0 = B[bi+0];
            float B1 = B[bi+1];
            float B2 = B[bi+2];
            float B3 = B[bi+3];
            bi += 4;

            vfloat32m1_t A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
            ai += 8;

            vfloat32m1_t result0 = __riscv_vfmul_vf_f32m1( A0, B0, gvl);
            vfloat32m1_t result1 = __riscv_vfmul_vf_f32m1( A0, B1, gvl);
            vfloat32m1_t result2 = __riscv_vfmul_vf_f32m1( A0, B2, gvl);
            vfloat32m1_t result3 = __riscv_vfmul_vf_f32m1( A0, B3, gvl);

            for(BLASLONG k=1; k<K; k++) {
                B0 = B[bi+0];
                B1 = B[bi+1];
                B2 = B[bi+2];
                B3 = B[bi+3];
                bi += 4;

                A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
                ai += 8;

                result0 = __riscv_vfmacc_vf_f32m1( result0, B0, A0, gvl);
                result1 = __riscv_vfmacc_vf_f32m1( result1, B1, A0, gvl);
                result2 = __riscv_vfmacc_vf_f32m1( result2, B2, A0, gvl);
                result3 = __riscv_vfmacc_vf_f32m1( result3, B3, A0, gvl);
            }


            BLASLONG ci=n_top*ldc+m_top;

            vfloat32m1_t c0 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c1 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c2 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c3 = __riscv_vle32_v_f32m1( &C[ci], gvl);
            c0 = __riscv_vfmacc_vf_f32m1( c0, alpha, result0, gvl );
            c1 = __riscv_vfmacc_vf_f32m1( c1, alpha, result1, gvl );
            c2 = __riscv_vfmacc_vf_f32m1( c2, alpha, result2, gvl );
            c3 = __riscv_vfmacc_vf_f32m1( c3, alpha, result3, gvl );

            ci=n_top*ldc+m_top;

            __riscv_vse32_v_f32m1( &C[ci], c0, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c1, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c2, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c3, gvl);
            m_top += 8;
        }


        if( M & 4 ) {
            gvl = __riscv_vsetvl_e32m1(4);

            BLASLONG ai=m_top*K;
            BLASLONG bi=n_top*K;
            float B0 = B[bi+0];
            float B1 = B[bi+1];
            float B2 = B[bi+2];
            float B3 = B[bi+3];
            bi += 4;

            vfloat32m1_t A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
            ai += 4;

            vfloat32m1_t result0 = __riscv_vfmul_vf_f32m1( A0, B0, gvl);
            vfloat32m1_t result1 = __riscv_vfmul_vf_f32m1( A0, B1, gvl);
            vfloat32m1_t result2 = __riscv_vfmul_vf_f32m1( A0, B2, gvl);
            vfloat32m1_t result3 = __riscv_vfmul_vf_f32m1( A0, B3, gvl);

            for(BLASLONG k=1; k<K; k++) {
                B0 = B[bi+0];
                B1 = B[bi+1];
                B2 = B[bi+2];
                B3 = B[bi+3];
                bi += 4;

                A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
                ai += 4;

                result0 = __riscv_vfmacc_vf_f32m1( result0, B0, A0, gvl);
                result1 = __riscv_vfmacc_vf_f32m1( result1, B1, A0, gvl);
                result2 = __riscv_vfmacc_vf_f32m1( result2, B2, A0, gvl);
                result3 = __riscv_vfmacc_vf_f32m1( result3, B3, A0, gvl);
            }


            BLASLONG ci=n_top*ldc+m_top;

            vfloat32m1_t c0 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c1 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c2 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c3 = __riscv_vle32_v_f32m1( &C[ci], gvl);
            c0 = __riscv_vfmacc_vf_f32m1( c0, alpha, result0, gvl );
            c1 = __riscv_vfmacc_vf_f32m1( c1, alpha, result1, gvl );
            c2 = __riscv_vfmacc_vf_f32m1( c2, alpha, result2, gvl );
            c3 = __riscv_vfmacc_vf_f32m1( c3, alpha, result3, gvl );

            ci=n_top*ldc+m_top;

            __riscv_vse32_v_f32m1( &C[ci], c0, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c1, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c2, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c3, gvl);
            m_top += 4;
        }


        if( M & 2 ) {
            float result0 = 0;
            float result1 = 0;
            float result2 = 0;
            float result3 = 0;
            float result4 = 0;
            float result5 = 0;
            float result6 = 0;
            float result7 = 0;
            BLASLONG ai=m_top*K;
            BLASLONG bi=n_top*K;

            for(BLASLONG k=0; k<K; k++) {
                result0+=A[ai+0]*B[bi+0];
                result1+=A[ai+1]*B[bi+0];
                result2+=A[ai+0]*B[bi+1];
                result3+=A[ai+1]*B[bi+1];
                result4+=A[ai+0]*B[bi+2];
                result5+=A[ai+1]*B[bi+2];
                result6+=A[ai+0]*B[bi+3];
                result7+=A[ai+1]*B[bi+3];
                ai+=2;
                bi+=4;
            }

            BLASLONG ci=n_top*ldc+m_top;
            C[ci+0*ldc+0] += alpha * result0;
            C[ci+0*ldc+1] += alpha * result1;
            C[ci+1*ldc+0] += alpha * result2;
            C[ci+1*ldc+1] += alpha * result3;
            C[ci+2*ldc+0] += alpha * result4;
            C[ci+2*ldc+1] += alpha * result5;
            C[ci+3*ldc+0] += alpha * result6;
            C[ci+3*ldc+1] += alpha * result7;
            m_top+=2;
        }


        if( M & 1 ) {
            float result0 = 0;
            float result1 = 0;
            float result2 = 0;
            float result3 = 0;
            BLASLONG ai=m_top*K;
            BLASLONG bi=n_top*K;

            for(BLASLONG k=0; k<K; k++) {
                result0+=A[ai+0]*B[bi+0];
                result1+=A[ai+0]*B[bi+1];
                result2+=A[ai+0]*B[bi+2];
                result3+=A[ai+0]*B[bi+3];
                ai+=1;
                bi+=4;
            }

            BLASLONG ci=n_top*ldc+m_top;
            C[ci+0*ldc+0] += alpha * result0;
            C[ci+1*ldc+0] += alpha * result1;
            C[ci+2*ldc+0] += alpha * result2;
            C[ci+3*ldc+0] += alpha * result3;
            m_top+=1;
        }

        n_top += 4;
    }



    // -- tails for N=2

    if( N & 2 ) {
        gvl = __riscv_vsetvl_e32m1(8);
        m_top = 0;

        for (BLASLONG i=0; i<M/16; i+=1) {
            BLASLONG ai=m_top*K;
            BLASLONG bi=n_top*K;
            float B0 = B[bi+0];
            float B1 = B[bi+1];
            bi += 2;

            vfloat32m1_t A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
            vfloat32m1_t A1 = __riscv_vle32_v_f32m1( &A[ai+1*gvl], gvl );
            ai += 16;

            vfloat32m1_t result0 = __riscv_vfmul_vf_f32m1( A0, B0, gvl);
            vfloat32m1_t result1 = __riscv_vfmul_vf_f32m1( A1, B0, gvl);
            vfloat32m1_t result2 = __riscv_vfmul_vf_f32m1( A0, B1, gvl);
            vfloat32m1_t result3 = __riscv_vfmul_vf_f32m1( A1, B1, gvl);

            for(BLASLONG k=1; k<K; k++) {
                B0 = B[bi+0];
                B1 = B[bi+1];
                bi += 2;

                A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
                A1 = __riscv_vle32_v_f32m1( &A[ai+1*gvl], gvl );
                ai += 16;

                result0 = __riscv_vfmacc_vf_f32m1( result0, B0, A0, gvl);
                result1 = __riscv_vfmacc_vf_f32m1( result1, B0, A1, gvl);
                result2 = __riscv_vfmacc_vf_f32m1( result2, B1, A0, gvl);
                result3 = __riscv_vfmacc_vf_f32m1( result3, B1, A1, gvl);
            }


            BLASLONG ci=n_top*ldc+m_top;

            vfloat32m1_t c0 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += gvl;
            vfloat32m1_t c1 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*1;
            vfloat32m1_t c2 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += gvl;
            vfloat32m1_t c3 = __riscv_vle32_v_f32m1( &C[ci], gvl);
            c0 = __riscv_vfmacc_vf_f32m1( c0, alpha, result0, gvl );
            c1 = __riscv_vfmacc_vf_f32m1( c1, alpha, result1, gvl );
            c2 = __riscv_vfmacc_vf_f32m1( c2, alpha, result2, gvl );
            c3 = __riscv_vfmacc_vf_f32m1( c3, alpha, result3, gvl );

            ci=n_top*ldc+m_top;

            __riscv_vse32_v_f32m1( &C[ci], c0, gvl); ci += gvl;
            __riscv_vse32_v_f32m1( &C[ci], c1, gvl); ci += ldc-gvl*1;
            __riscv_vse32_v_f32m1( &C[ci], c2, gvl); ci += gvl;
            __riscv_vse32_v_f32m1( &C[ci], c3, gvl);
            m_top += 16;
        }


        if( M & 8 ) {
            gvl = __riscv_vsetvl_e32m1(8);

            BLASLONG ai=m_top*K;
            BLASLONG bi=n_top*K;
            float B0 = B[bi+0];
            float B1 = B[bi+1];
            bi += 2;

            vfloat32m1_t A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
            ai += 8;

            vfloat32m1_t result0 = __riscv_vfmul_vf_f32m1( A0, B0, gvl);
            vfloat32m1_t result1 = __riscv_vfmul_vf_f32m1( A0, B1, gvl);

            for(BLASLONG k=1; k<K; k++) {
                B0 = B[bi+0];
                B1 = B[bi+1];
                bi += 2;

                A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
                ai += 8;

                result0 = __riscv_vfmacc_vf_f32m1( result0, B0, A0, gvl);
                result1 = __riscv_vfmacc_vf_f32m1( result1, B1, A0, gvl);
            }


            BLASLONG ci=n_top*ldc+m_top;

            vfloat32m1_t c0 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c1 = __riscv_vle32_v_f32m1( &C[ci], gvl);
            c0 = __riscv_vfmacc_vf_f32m1( c0, alpha, result0, gvl );
            c1 = __riscv_vfmacc_vf_f32m1( c1, alpha, result1, gvl );

            ci=n_top*ldc+m_top;

            __riscv_vse32_v_f32m1( &C[ci], c0, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c1, gvl);
            m_top += 8;
        }


        if( M & 4 ) {
            gvl = __riscv_vsetvl_e32m1(4);

            BLASLONG ai=m_top*K;
            BLASLONG bi=n_top*K;
            float B0 = B[bi+0];
            float B1 = B[bi+1];
            bi += 2;

            vfloat32m1_t A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
            ai += 4;

            vfloat32m1_t result0 = __riscv_vfmul_vf_f32m1( A0, B0, gvl);
            vfloat32m1_t result1 = __riscv_vfmul_vf_f32m1( A0, B1, gvl);

            for(BLASLONG k=1; k<K; k++) {
                B0 = B[bi+0];
                B1 = B[bi+1];
                bi += 2;

                A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
                ai += 4;

                result0 = __riscv_vfmacc_vf_f32m1( result0, B0, A0, gvl);
                result1 = __riscv_vfmacc_vf_f32m1( result1, B1, A0, gvl);
            }


            BLASLONG ci=n_top*ldc+m_top;

            vfloat32m1_t c0 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += ldc-gvl*0;
            vfloat32m1_t c1 = __riscv_vle32_v_f32m1( &C[ci], gvl);
            c0 = __riscv_vfmacc_vf_f32m1( c0, alpha, result0, gvl );
            c1 = __riscv_vfmacc_vf_f32m1( c1, alpha, result1, gvl );

            ci=n_top*ldc+m_top;

            __riscv_vse32_v_f32m1( &C[ci], c0, gvl); ci += ldc-gvl*0;
            __riscv_vse32_v_f32m1( &C[ci], c1, gvl);
            m_top += 4;
        }


        if( M & 2 ) {
            float result0 = 0;
            float result1 = 0;
            float result2 = 0;
            float result3 = 0;
            BLASLONG ai=m_top*K;
            BLASLONG bi=n_top*K;

            for(BLASLONG k=0; k<K; k++) {
                result0+=A[ai+0]*B[bi+0];
                result1+=A[ai+1]*B[bi+0];
                result2+=A[ai+0]*B[bi+1];
                result3+=A[ai+1]*B[bi+1];
                ai+=2;
                bi+=2;
            }

            BLASLONG ci=n_top*ldc+m_top;
            C[ci+0*ldc+0] += alpha * result0;
            C[ci+0*ldc+1] += alpha * result1;
            C[ci+1*ldc+0] += alpha * result2;
            C[ci+1*ldc+1] += alpha * result3;
            m_top+=2;
        }


        if( M & 1 ) {
            float result0 = 0;
            float result1 = 0;
            BLASLONG ai=m_top*K;
            BLASLONG bi=n_top*K;

            for(BLASLONG k=0; k<K; k++) {
                result0+=A[ai+0]*B[bi+0];
                result1+=A[ai+0]*B[bi+1];
                ai+=1;
                bi+=2;
            }

            BLASLONG ci=n_top*ldc+m_top;
            C[ci+0*ldc+0] += alpha * result0;
            C[ci+1*ldc+0] += alpha * result1;
            m_top+=1;
        }

        n_top += 2;
    }



    // -- tails for N=1

    if( N & 1 ) {
        gvl = __riscv_vsetvl_e32m1(8);
        m_top = 0;

        for (BLASLONG i=0; i<M/16; i+=1) {
            BLASLONG ai=m_top*K;
            BLASLONG bi=n_top*K;
            float B0 = B[bi+0];
            bi += 1;

            vfloat32m1_t A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
            vfloat32m1_t A1 = __riscv_vle32_v_f32m1( &A[ai+1*gvl], gvl );
            ai += 16;

            vfloat32m1_t result0 = __riscv_vfmul_vf_f32m1( A0, B0, gvl);
            vfloat32m1_t result1 = __riscv_vfmul_vf_f32m1( A1, B0, gvl);

            for(BLASLONG k=1; k<K; k++) {
                B0 = B[bi+0];
                bi += 1;

                A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
                A1 = __riscv_vle32_v_f32m1( &A[ai+1*gvl], gvl );
                ai += 16;

                result0 = __riscv_vfmacc_vf_f32m1( result0, B0, A0, gvl);
                result1 = __riscv_vfmacc_vf_f32m1( result1, B0, A1, gvl);
            }


            BLASLONG ci=n_top*ldc+m_top;

            vfloat32m1_t c0 = __riscv_vle32_v_f32m1( &C[ci], gvl); ci += gvl;
            vfloat32m1_t c1 = __riscv_vle32_v_f32m1( &C[ci], gvl);
            c0 = __riscv_vfmacc_vf_f32m1( c0, alpha, result0, gvl );
            c1 = __riscv_vfmacc_vf_f32m1( c1, alpha, result1, gvl );

            ci=n_top*ldc+m_top;

            __riscv_vse32_v_f32m1( &C[ci], c0, gvl); ci += gvl;
            __riscv_vse32_v_f32m1( &C[ci], c1, gvl);
            m_top += 16;
        }


        if( M & 8 ) {
            gvl = __riscv_vsetvl_e32m1(8);

            BLASLONG ai=m_top*K;
            BLASLONG bi=n_top*K;
            float B0 = B[bi+0];
            bi += 1;

            vfloat32m1_t A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
            ai += 8;

            vfloat32m1_t result0 = __riscv_vfmul_vf_f32m1( A0, B0, gvl);

            for(BLASLONG k=1; k<K; k++) {
                B0 = B[bi+0];
                bi += 1;

                A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
                ai += 8;

                result0 = __riscv_vfmacc_vf_f32m1( result0, B0, A0, gvl);
            }


            BLASLONG ci=n_top*ldc+m_top;

            vfloat32m1_t c0 = __riscv_vle32_v_f32m1( &C[ci], gvl);
            c0 = __riscv_vfmacc_vf_f32m1( c0, alpha, result0, gvl );

            ci=n_top*ldc+m_top;

            __riscv_vse32_v_f32m1( &C[ci], c0, gvl);
            m_top += 8;
        }


        if( M & 4 ) {
            gvl = __riscv_vsetvl_e32m1(4);

            BLASLONG ai=m_top*K;
            BLASLONG bi=n_top*K;
            float B0 = B[bi+0];
            bi += 1;

            vfloat32m1_t A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
            ai += 4;

            vfloat32m1_t result0 = __riscv_vfmul_vf_f32m1( A0, B0, gvl);

            for(BLASLONG k=1; k<K; k++) {
                B0 = B[bi+0];
                bi += 1;

                A0 = __riscv_vle32_v_f32m1( &A[ai+0*gvl], gvl );
                ai += 4;

                result0 = __riscv_vfmacc_vf_f32m1( result0, B0, A0, gvl);
            }


            BLASLONG ci=n_top*ldc+m_top;

            vfloat32m1_t c0 = __riscv_vle32_v_f32m1( &C[ci], gvl);
            c0 = __riscv_vfmacc_vf_f32m1( c0, alpha, result0, gvl );

            ci=n_top*ldc+m_top;

            __riscv_vse32_v_f32m1( &C[ci], c0, gvl);
            m_top += 4;
        }


        if( M & 2 ) {
            float result0 = 0;
            float result1 = 0;
            BLASLONG ai=m_top*K;
            BLASLONG bi=n_top*K;

            for(BLASLONG k=0; k<K; k++) {
                result0+=A[ai+0]*B[bi+0];
                result1+=A[ai+1]*B[bi+0];
                ai+=2;
                bi+=1;
            }

            BLASLONG ci=n_top*ldc+m_top;
            C[ci+0*ldc+0] += alpha * result0;
            C[ci+0*ldc+1] += alpha * result1;
            m_top+=2;
        }


        if( M & 1 ) {
            float result0 = 0;
            BLASLONG ai=m_top*K;
            BLASLONG bi=n_top*K;

            for(BLASLONG k=0; k<K; k++) {
                result0+=A[ai+0]*B[bi+0];
                ai+=1;
                bi+=1;
            }

            BLASLONG ci=n_top*ldc+m_top;
            C[ci+0*ldc+0] += alpha * result0;
            m_top+=1;
        }

        n_top += 1;
    }

    return 0;
}