static void AddGreenToBlueAndRed(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, vaddq_u8(argb, greens));
}
// fallthrough and finish off with plain-C
VP8LAddGreenToBlueAndRed_C(argb_data, num_pixels & 3);
}
static void AddGreenToBlueAndRed(uint32_t* argb_data, int num_pixels) {
const __m128i mask = _mm_set1_epi32(0x0000ff00);
int i;
for (i = 0; i + 4 <= num_pixels; i += 4) {
const __m128i in = _mm_loadu_si128((__m128i*)&argb_data[i]);
const __m128i in_00g0 = _mm_and_si128(in, mask); // 00g0|00g0|...
const __m128i in_0g00 = _mm_slli_epi32(in_00g0, 8); // 0g00|0g00|...
const __m128i in_000g = _mm_srli_epi32(in_00g0, 8); // 000g|000g|...
const __m128i in_0g0g = _mm_or_si128(in_0g00, in_000g);
const __m128i out = _mm_add_epi8(in, in_0g0g);
_mm_storeu_si128((__m128i*)&argb_data[i], out);
}
// fallthrough and finish off with plain-C
VP8LAddGreenToBlueAndRed_C(argb_data + i, num_pixels - i);
}
static void AddGreenToBlueAndRed_SSE2(const uint32_t* const src, int num_pixels,
uint32_t* dst) {
int i;
for (i = 0; i + 4 <= num_pixels; i += 4) {
const __m128i in = _mm_loadu_si128((const __m128i*)&src[i]); // argb
const __m128i A = _mm_srli_epi16(in, 8); // 0 a 0 g
const __m128i B = _mm_shufflelo_epi16(A, _MM_SHUFFLE(2, 2, 0, 0));
const __m128i C = _mm_shufflehi_epi16(B, _MM_SHUFFLE(2, 2, 0, 0)); // 0g0g
const __m128i out = _mm_add_epi8(in, C);
_mm_storeu_si128((__m128i*)&dst[i], out);
}
// fallthrough and finish off with plain-C
if (i != num_pixels) {
VP8LAddGreenToBlueAndRed_C(src + i, num_pixels - i, dst + i);
}
}