_asm {
    mov             esi, K8;
    sub             esi, 8;
    shl             esi, 2;
    xor             edi, edi;
    mov             edx, a;
    mov             ebx, bb;
    vxorps          ymm3, ymm3, ymm3;
Lrep:
    cmp             edi, esi;
    jg              Lexit;
    vmovups         ymm0, ymmword ptr[edx + edi];
    vfmadd231ps     ymm3, ymm0, ymmword ptr[ebx + edi];
    add             edi, 32;
    jmp             Lrep;
Lexit:
    vmovups         ymmword ptr[c], ymm3;
}

pA = A;
for (k = 0; k < K; k++) {
    pC = C;
    for (i = 0; i < M; i++) {
        pA = A + i * K + k;
        pB = B + k * N;
        for (j = N / 32; j > 0; j--) {
            _asm {
                mov             eax, pC;
                mov             ebx, pA;
                mov             ecx, pB;
                vmovups         ymm0, ymmword ptr[eax];
                vmovss          xmm1, dword ptr[ebx];
                vbroadcastss    ymm4, xmm1;
                vmovups         ymm2, ymmword ptr[ecx];
                vfmadd231ps     ymm0, ymm4, ymm2;
                vmovups         ymmword ptr[eax], ymm0;
            }
            pC += 8; pB += 8;
            _asm {
                mov             eax, pC;
                mov             ebx, pA;
                mov             ecx, pB;
                vmovups         ymm0, ymmword ptr[eax];
                vmovss          xmm1, dword ptr[ebx];
                vbroadcastss    ymm4, xmm1;
                vmovups         ymm2, ymmword ptr[ecx];
                vfmadd231ps     ymm0, ymm4, ymm2;
                vmovups         ymmword ptr[eax], ymm0;
            }
            pC += 8; pB += 8;
            _asm {
                mov             eax, pC;
                mov             ebx, pA;
                mov             ecx, pB;
                vmovups         ymm0, ymmword ptr[eax];
                vmovss          xmm1, dword ptr[ebx];
                vbroadcastss    ymm4, xmm1;
                vmovups         ymm2, ymmword ptr[ecx];
                vfmadd231ps     ymm0, ymm4, ymm2;
                vmovups         ymmword ptr[eax], ymm0;
            }
            pC += 8; pB += 8;
            _asm {
                mov             eax, pC;
                mov             ebx, pA;
                mov             ecx, pB;
                vmovups         ymm0, ymmword ptr[eax];
                vmovss          xmm1, dword ptr[ebx];
                vbroadcastss    ymm4, xmm1;
                vmovups         ymm2, ymmword ptr[ecx];
                vfmadd231ps     ymm0, ymm4, ymm2;
                vmovups         ymmword ptr[eax], ymm0;
            }
            pC += 8; pB += 8;
        }
        for (j = N / 32 * 32; j < N; j++) {
            *pC += *pA * *pB;
            pC += 1; pB += 1;
        }
    }
}