//Note: it takes size and offset in units of byte
static inline int compute_ham_similarity_64(unsigned short* ref, unsigned short* circ_array,
int size){
const uint8_t* ref_c=(uint8_t*) ref;
const uint8_t* circ_c=(uint8_t*) circ_array;
register uint8x16_t a,b;
register uint8x16_t c,d,temp;
register uint16x8_t acc;
register uint i=0,count=0;
int j=0;
int shift=size&0xF;
for(i=0;i<=size-16; i+=16){
j++;
a=vld1q_u8(&ref_c[i]);
b=vld1q_u8(&circ_c[i]);
c=veorq_u8(a,b);
acc=vaddq_u16(acc,vpaddlq_u8(vcntq_u8(c)));
}
count=setbits(acc);
a=vld1q_u8(&ref_c[i]);
b=vld1q_u8(&circ_c[i]);
c=veorq_u8(a,b);
c=vcntq_u8(c);
for(i=0;i<shift;i++){
count=count+vgetq_lane_u8 (c,i);
}
return size*8-count;
}
static inline void
desc_to_ptype_v(uint64x2_t descs[4], struct rte_mbuf **rx_pkts,
uint32_t *ptype_tbl)
{
int i;
uint8_t ptype;
uint8x16_t tmp;
for (i = 0; i < 4; i++) {
tmp = vreinterpretq_u8_u64(vshrq_n_u64(descs[i], 30));
ptype = vgetq_lane_u8(tmp, 8);
rx_pkts[i]->packet_type = ptype_tbl[ptype];
}
}
uint8_t test_vgetq_lane_u8(uint8x16_t v1) {
// CHECK: test_vgetq_lane_u8
return vgetq_lane_u8(v1, 15);
// CHECK: umov {{w[0-9]+}}, {{v[0-9]+}}.b[15]
}
uint8_t test_vgetq_lane_u8(uint8x16_t a) {
// CHECK-LABEL: test_vgetq_lane_u8:
// CHECK-NEXT: umov.b w0, v0[15]
// CHECK-NEXT: ret
return vgetq_lane_u8(a, 15);
}
/* u8x16 mv mul */
void mw_neon_mv_mul_u8x16(unsigned char * A, int Row, int T, unsigned char * B, unsigned char * C)
{
int i = 0;
int k = 0;
uint8x16_t neon_b, neon_c;
uint8x16_t neon_a0, neon_a1, neon_a2, neon_a3, neon_a4, neon_a5, neon_a6, neon_a7;
uint8x16_t neon_a8, neon_a9, neon_a10, neon_a11, neon_a12, neon_a13, neon_a14, neon_a15;
uint8x16_t neon_b0, neon_b1, neon_b2, neon_b3, neon_b4, neon_b5, neon_b6, neon_b7;
uint8x16_t neon_b8, neon_b9, neon_b10, neon_b11, neon_b12, neon_b13, neon_b14, neon_b15;
for (i = 0; i < Row; i+=16)
{
neon_c = vmovq_n_u8(0);
for (k = 0; k < T; k+=16)
{
int j = k * T + i;
neon_a0 = vld1q_u8(A + j);
j+=Row;
neon_a1 = vld1q_u8(A + j);
j+=Row;
neon_a2 = vld1q_u8(A + j);
j+=Row;
neon_a3 = vld1q_u8(A + j);
j+=Row;
neon_a4 = vld1q_u8(A + j);
j+=Row;
neon_a5 = vld1q_u8(A + j);
j+=Row;
neon_a6 = vld1q_u8(A + j);
j+=Row;
neon_a7 = vld1q_u8(A + j);
j+=Row;
neon_a8 = vld1q_u8(A + j);
j+=Row;
neon_a9 = vld1q_u8(A + j);
j+=Row;
neon_a10 = vld1q_u8(A + j);
j+=Row;
neon_a11 = vld1q_u8(A + j);
j+=Row;
neon_a12 = vld1q_u8(A + j);
j+=Row;
neon_a13 = vld1q_u8(A + j);
j+=Row;
neon_a14 = vld1q_u8(A + j);
j+=Row;
neon_a15 = vld1q_u8(A + j);
neon_b = vld1q_u8(B + k);
neon_b0 = vdupq_n_u8(vgetq_lane_u8(neon_b, 0));
neon_b1 = vdupq_n_u8(vgetq_lane_u8(neon_b, 1));
neon_b2 = vdupq_n_u8(vgetq_lane_u8(neon_b, 2));
neon_b3 = vdupq_n_u8(vgetq_lane_u8(neon_b, 3));
neon_b4 = vdupq_n_u8(vgetq_lane_u8(neon_b, 4));
neon_b5 = vdupq_n_u8(vgetq_lane_u8(neon_b, 5));
neon_b6 = vdupq_n_u8(vgetq_lane_u8(neon_b, 6));
neon_b7 = vdupq_n_u8(vgetq_lane_u8(neon_b, 7));
neon_b8 = vdupq_n_u8(vgetq_lane_u8(neon_b, 8));
neon_b9 = vdupq_n_u8(vgetq_lane_u8(neon_b, 9));
neon_b10 = vdupq_n_u8(vgetq_lane_u8(neon_b, 10));
neon_b11 = vdupq_n_u8(vgetq_lane_u8(neon_b, 11));
neon_b12 = vdupq_n_u8(vgetq_lane_u8(neon_b, 12));
neon_b13 = vdupq_n_u8(vgetq_lane_u8(neon_b, 13));
neon_b14 = vdupq_n_u8(vgetq_lane_u8(neon_b, 14));
neon_b15 = vdupq_n_u8(vgetq_lane_u8(neon_b, 15));
neon_c = vaddq_u8(vmulq_u8(neon_a0, neon_b0), neon_c);
neon_c = vaddq_u8(vmulq_u8(neon_a1, neon_b1), neon_c);
neon_c = vaddq_u8(vmulq_u8(neon_a2, neon_b2), neon_c);
neon_c = vaddq_u8(vmulq_u8(neon_a3, neon_b3), neon_c);
neon_c = vaddq_u8(vmulq_u8(neon_a4, neon_b4), neon_c);
neon_c = vaddq_u8(vmulq_u8(neon_a5, neon_b5), neon_c);
neon_c = vaddq_u8(vmulq_u8(neon_a6, neon_b6), neon_c);
neon_c = vaddq_u8(vmulq_u8(neon_a7, neon_b7), neon_c);
neon_c = vaddq_u8(vmulq_u8(neon_a8, neon_b8), neon_c);
neon_c = vaddq_u8(vmulq_u8(neon_a9, neon_b9), neon_c);
neon_c = vaddq_u8(vmulq_u8(neon_a10, neon_b10), neon_c);
neon_c = vaddq_u8(vmulq_u8(neon_a11, neon_b11), neon_c);
neon_c = vaddq_u8(vmulq_u8(neon_a12, neon_b12), neon_c);
neon_c = vaddq_u8(vmulq_u8(neon_a13, neon_b13), neon_c);
neon_c = vaddq_u8(vmulq_u8(neon_a14, neon_b14), neon_c);
neon_c = vaddq_u8(vmulq_u8(neon_a15, neon_b15), neon_c);
}
vst1q_u8(C + i, neon_c);
}
}
// CHECK-LABEL: define i8 @test_vgetq_lane_u8(<16 x i8> %a) #0 {
// CHECK: [[VGETQ_LANE:%.*]] = extractelement <16 x i8> %a, i32 15
// CHECK: ret i8 [[VGETQ_LANE]]
uint8_t test_vgetq_lane_u8(uint8x16_t a) {
return vgetq_lane_u8(a, 15);
}
static void ShiftRows(void)
{
uint8x16_t state_dup;
state_dup = *state;
temp = vgetq_lane_u8(state_dup, 1);
vst1q_lane_u8(&temp, *state, 13);
temp = vgetq_lane_u8(state_dup, 2);
vst1q_lane_u8(&temp, *state, 10);
temp = vgetq_lane_u8(state_dup, 3);
vst1q_lane_u8(&temp, *state, 7);
temp = vgetq_lane_u8(state_dup, 5);
vst1q_lane_u8(&temp, *state, 1);
temp = vgetq_lane_u8(state_dup, 6);
vst1q_lane_u8(&temp, *state, 14);
temp = vgetq_lane_u8(state_dup, 7);
vst1q_lane_u8(&temp, *state, 11);
temp = vgetq_lane_u8(state_dup, 9);
vst1q_lane_u8(&temp, *state, 5);
temp = vgetq_lane_u8(state_dup, 10);
vst1q_lane_u8(&temp, *state, 2);
temp = vgetq_lane_u8(state_dup, 11);
vst1q_lane_u8(&temp, *state, 15);
temp = vgetq_lane_u8(state_dup, 13);
vst1q_lane_u8(&temp, *state, 9);
temp = vgetq_lane_u8(state_dup, 14);
vst1q_lane_u8(&temp, *state, 6);
temp = vgetq_lane_u8(state_dup, 15);
vst1q_lane_u8(&temp, *state, 3);
}
