void test_vsubQu8 (void)
{
uint8x16_t out_uint8x16_t;
uint8x16_t arg0_uint8x16_t;
uint8x16_t arg1_uint8x16_t;
out_uint8x16_t = vsubq_u8 (arg0_uint8x16_t, arg1_uint8x16_t);
}
static void SubtractGreenFromBlueAndRed(uint32_t* argb_data, int num_pixels) {
const uint32_t* const end = argb_data + (num_pixels & ~3);
const uint8x8_t shuffle = vld1_u8(kGreenShuffle);
for (; argb_data < end; argb_data += 4) {
const uint8x16_t argb = vld1q_u8((uint8_t*)argb_data);
const uint8x16_t greens =
vcombine_u8(vtbl1_u8(vget_low_u8(argb), shuffle),
vtbl1_u8(vget_high_u8(argb), shuffle));
vst1q_u8((uint8_t*)argb_data, vsubq_u8(argb, greens));
}
// fallthrough and finish off with plain-C
VP8LSubtractGreenFromBlueAndRed_C(argb_data, num_pixels & 3);
}
static void SubtractGreenFromBlueAndRed(uint32_t* argb_data, int num_pixels) {
const uint32_t* const end = argb_data + (num_pixels & ~3);
#ifdef USE_VTBLQ
const uint8x16_t shuffle = vld1q_u8(kGreenShuffle);
#else
const uint8x8_t shuffle = vld1_u8(kGreenShuffle);
#endif
for (; argb_data < end; argb_data += 4) {
const uint8x16_t argb = vld1q_u8((uint8_t*)argb_data);
const uint8x16_t greens = DoGreenShuffle(argb, shuffle);
vst1q_u8((uint8_t*)argb_data, vsubq_u8(argb, greens));
}
// fallthrough and finish off with plain-C
VP8LSubtractGreenFromBlueAndRed_C(argb_data, num_pixels & 3);
}
void ar_vsub_u8_neon(uint8_t* res,
const uint8_t* a,
const uint8_t* b,
uint32_t n)
{
uint8x16_t a_loaded;
uint8x16_t b_loaded;
uint8x16_t res_loaded;
for (uint32_t i = 0; i < n; i += 16) {
a_loaded = vld1q_u8(&(a[i]));
b_loaded = vld1q_u8(&(b[i]));
res_loaded = vsubq_u8(a_loaded, b_loaded);
vst1q_u8(&(res[i]),res_loaded);
}
}
/* u8x16 sub */
void mw_neon_mm_sub_u8x16(unsigned char * A, int Row, int Col, unsigned char * B, unsigned char * C)
{
uint8x16_t neon_a, neon_b, neon_c;
int size = Row * Col;
int i = 0;
int k = 0;
for (i = 16; i <= size ; i+=16)
{
k = i - 16;
neon_a = vld1q_u8(A + k);
neon_b = vld1q_u8(B + k);
neon_c = vsubq_u8(neon_a, neon_b);
vst1q_u8(C + k, neon_c);
}
k = i - 16;
for (i = 0; i < size % 16; i++)
{
C[k + i] = A[k + i] - B[k + i];
}
}
inline uint8x16_t vsubq(const uint8x16_t & v0, const uint8x16_t & v1) { return vsubq_u8 (v0, v1); }
static inline uint8x16x4_t
enc_translate (uint8x16x4_t in)
{
uint8x16x4_t mask1, mask2, mask3, mask4, out;
// Translate values 0..63 to the Base64 alphabet. There are five sets:
// # From To Abs Delta Characters
// 0 [0..25] [65..90] +65 +65 ABCDEFGHIJKLMNOPQRSTUVWXYZ
// 1 [26..51] [97..122] +71 +6 abcdefghijklmnopqrstuvwxyz
// 2 [52..61] [48..57] -4 -75 0123456789
// 3 [62] [43] -19 -15 +
// 4 [63] [47] -16 +3 /
// Create cumulative masks for characters in sets [1,2,3,4], [2,3,4],
// [3,4], and [4]:
mask1.val[0] = CMPGT(in.val[0], 25);
mask1.val[1] = CMPGT(in.val[1], 25);
mask1.val[2] = CMPGT(in.val[2], 25);
mask1.val[3] = CMPGT(in.val[3], 25);
mask2.val[0] = CMPGT(in.val[0], 51);
mask2.val[1] = CMPGT(in.val[1], 51);
mask2.val[2] = CMPGT(in.val[2], 51);
mask2.val[3] = CMPGT(in.val[3], 51);
mask3.val[0] = CMPGT(in.val[0], 61);
mask3.val[1] = CMPGT(in.val[1], 61);
mask3.val[2] = CMPGT(in.val[2], 61);
mask3.val[3] = CMPGT(in.val[3], 61);
mask4.val[0] = CMPEQ(in.val[0], 63);
mask4.val[1] = CMPEQ(in.val[1], 63);
mask4.val[2] = CMPEQ(in.val[2], 63);
mask4.val[3] = CMPEQ(in.val[3], 63);
// All characters are at least in cumulative set 0, so add 'A':
out.val[0] = vaddq_u8(in.val[0], vdupq_n_u8(65));
out.val[1] = vaddq_u8(in.val[1], vdupq_n_u8(65));
out.val[2] = vaddq_u8(in.val[2], vdupq_n_u8(65));
out.val[3] = vaddq_u8(in.val[3], vdupq_n_u8(65));
// For inputs which are also in any of the other cumulative sets,
// add delta values against the previous set(s) to correct the shift:
out.val[0] = vaddq_u8(out.val[0], REPLACE(mask1.val[0], 6));
out.val[1] = vaddq_u8(out.val[1], REPLACE(mask1.val[1], 6));
out.val[2] = vaddq_u8(out.val[2], REPLACE(mask1.val[2], 6));
out.val[3] = vaddq_u8(out.val[3], REPLACE(mask1.val[3], 6));
out.val[0] = vsubq_u8(out.val[0], REPLACE(mask2.val[0], 75));
out.val[1] = vsubq_u8(out.val[1], REPLACE(mask2.val[1], 75));
out.val[2] = vsubq_u8(out.val[2], REPLACE(mask2.val[2], 75));
out.val[3] = vsubq_u8(out.val[3], REPLACE(mask2.val[3], 75));
out.val[0] = vsubq_u8(out.val[0], REPLACE(mask3.val[0], 15));
out.val[1] = vsubq_u8(out.val[1], REPLACE(mask3.val[1], 15));
out.val[2] = vsubq_u8(out.val[2], REPLACE(mask3.val[2], 15));
out.val[3] = vsubq_u8(out.val[3], REPLACE(mask3.val[3], 15));
out.val[0] = vaddq_u8(out.val[0], REPLACE(mask4.val[0], 3));
out.val[1] = vaddq_u8(out.val[1], REPLACE(mask4.val[1], 3));
out.val[2] = vaddq_u8(out.val[2], REPLACE(mask4.val[2], 3));
out.val[3] = vaddq_u8(out.val[3], REPLACE(mask4.val[3], 3));
return out;
}