void dmz_deinterleave_RGBA_to_R(uint8_t *source, uint8_t *dest, int size) {
#if DMZ_HAS_NEON_COMPILETIME
if (dmz_has_neon_runtime()) {
assert(size >= 16); // required for the vectorized handling of leftover_bytes; also, a reasonable expectation!
for (int offset = 0; offset + 15 < size; offset += 16) {
uint8x16x4_t r1 = vld4q_u8(&source[offset * 4]);
vst1q_u8(&dest[offset], r1.val[0]);
}
// use "overlapping" to process the remaining bytes
// See http://community.arm.com/groups/processors/blog/2010/05/10/coding-for-neon--part-2-dealing-with-leftovers
if (size % 16 > 0) {
int offset = size - 16;
uint8x16x4_t r1 = vld4q_u8(&source[offset * 4]);
vst1q_u8(&dest[offset], r1.val[0]);
}
}
else
#endif
{
for (int offset = 0; offset + 7 < size; offset += 8) {
int bufferOffset = offset * 4;
dest[offset] = source[bufferOffset];
dest[offset + 1] = source[bufferOffset + (1 * 4)];
dest[offset + 2] = source[bufferOffset + (2 * 4)];
dest[offset + 3] = source[bufferOffset + (3 * 4)];
dest[offset + 4] = source[bufferOffset + (4 * 4)];
dest[offset + 5] = source[bufferOffset + (5 * 4)];
dest[offset + 6] = source[bufferOffset + (6 * 4)];
dest[offset + 7] = source[bufferOffset + (7 * 4)];
}
int leftover_bytes = size % 8; // each RGBA pixel is 4 bytes, so can assume size % 4 == 0
if (leftover_bytes > 0) {
for (int offset = size - leftover_bytes; offset < size; offset += 4) {
int bufferOffset = offset * 4;
dest[offset] = source[bufferOffset];
dest[offset + 1] = source[bufferOffset + (1 * 4)];
dest[offset + 2] = source[bufferOffset + (2 * 4)];
dest[offset + 3] = source[bufferOffset + (3 * 4)];
}
}
}
}
extern "C" jboolean JNICALL Java_io_card_payment_CardScanner_nUseNeon() {
return (dmz_has_neon_runtime());
}
