void test_vmovns16 (void)
{
int8x8_t out_int8x8_t;
int16x8_t arg0_int16x8_t;
out_int8x8_t = vmovn_s16 (arg0_int16x8_t);
}
JL_DLLEXPORT struct_aa64_3 test_aa64_vec_2(struct_aa64_3 v1, struct_aa64_4 v2)
{
struct_aa64_3 x = {v1.v1 + vmovn_s16(v2.v1), v1.v2 - v2.v2};
return x;
}
inline int8x8_t vmovn(const int16x8_t & v) { return vmovn_s16(v); }
void qt_blend_argb32_on_argb32_neon(uchar *destPixels, int dbpl,
const uchar *srcPixels, int sbpl,
int w, int h,
int const_alpha)
{
const uint *src = (const uint *) srcPixels;
uint *dst = (uint *) destPixels;
int16x8_t half = vdupq_n_s16(0x80);
int16x8_t full = vdupq_n_s16(0xff);
if (const_alpha == 256) {
for (int y = 0; y < h; ++y) {
int x = 0;
for (; x < w-3; x += 4) {
int32x4_t src32 = vld1q_s32((int32_t *)&src[x]);
if ((src[x] & src[x+1] & src[x+2] & src[x+3]) >= 0xff000000) {
// all opaque
vst1q_s32((int32_t *)&dst[x], src32);
} else if (src[x] | src[x+1] | src[x+2] | src[x+3]) {
int32x4_t dst32 = vld1q_s32((int32_t *)&dst[x]);
const uint8x16_t src8 = vreinterpretq_u8_s32(src32);
const uint8x16_t dst8 = vreinterpretq_u8_s32(dst32);
const uint8x8_t src8_low = vget_low_u8(src8);
const uint8x8_t dst8_low = vget_low_u8(dst8);
const uint8x8_t src8_high = vget_high_u8(src8);
const uint8x8_t dst8_high = vget_high_u8(dst8);
const int16x8_t src16_low = vreinterpretq_s16_u16(vmovl_u8(src8_low));
const int16x8_t dst16_low = vreinterpretq_s16_u16(vmovl_u8(dst8_low));
const int16x8_t src16_high = vreinterpretq_s16_u16(vmovl_u8(src8_high));
const int16x8_t dst16_high = vreinterpretq_s16_u16(vmovl_u8(dst8_high));
const int16x8_t result16_low = qvsource_over_s16(src16_low, dst16_low, half, full);
const int16x8_t result16_high = qvsource_over_s16(src16_high, dst16_high, half, full);
const int32x2_t result32_low = vreinterpret_s32_s8(vmovn_s16(result16_low));
const int32x2_t result32_high = vreinterpret_s32_s8(vmovn_s16(result16_high));
vst1q_s32((int32_t *)&dst[x], vcombine_s32(result32_low, result32_high));
}
}
for (; x<w; ++x) {
uint s = src[x];
if (s >= 0xff000000)
dst[x] = s;
else if (s != 0)
dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s));
}
dst = (quint32 *)(((uchar *) dst) + dbpl);
src = (const quint32 *)(((const uchar *) src) + sbpl);
}
} else if (const_alpha != 0) {
const_alpha = (const_alpha * 255) >> 8;
int16x8_t const_alpha16 = vdupq_n_s16(const_alpha);
for (int y = 0; y < h; ++y) {
int x = 0;
for (; x < w-3; x += 4) {
if (src[x] | src[x+1] | src[x+2] | src[x+3]) {
int32x4_t src32 = vld1q_s32((int32_t *)&src[x]);
int32x4_t dst32 = vld1q_s32((int32_t *)&dst[x]);
const uint8x16_t src8 = vreinterpretq_u8_s32(src32);
const uint8x16_t dst8 = vreinterpretq_u8_s32(dst32);
const uint8x8_t src8_low = vget_low_u8(src8);
const uint8x8_t dst8_low = vget_low_u8(dst8);
const uint8x8_t src8_high = vget_high_u8(src8);
const uint8x8_t dst8_high = vget_high_u8(dst8);
const int16x8_t src16_low = vreinterpretq_s16_u16(vmovl_u8(src8_low));
const int16x8_t dst16_low = vreinterpretq_s16_u16(vmovl_u8(dst8_low));
const int16x8_t src16_high = vreinterpretq_s16_u16(vmovl_u8(src8_high));
const int16x8_t dst16_high = vreinterpretq_s16_u16(vmovl_u8(dst8_high));
const int16x8_t srcalpha16_low = qvbyte_mul_s16(src16_low, const_alpha16, half);
const int16x8_t srcalpha16_high = qvbyte_mul_s16(src16_high, const_alpha16, half);
const int16x8_t result16_low = qvsource_over_s16(srcalpha16_low, dst16_low, half, full);
const int16x8_t result16_high = qvsource_over_s16(srcalpha16_high, dst16_high, half, full);
const int32x2_t result32_low = vreinterpret_s32_s8(vmovn_s16(result16_low));
const int32x2_t result32_high = vreinterpret_s32_s8(vmovn_s16(result16_high));
vst1q_s32((int32_t *)&dst[x], vcombine_s32(result32_low, result32_high));
}
}
for (; x<w; ++x) {
uint s = src[x];
if (s != 0) {
s = BYTE_MUL(s, const_alpha);
dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s));
}
}
dst = (quint32 *)(((uchar *) dst) + dbpl);
src = (const quint32 *)(((const uchar *) src) + sbpl);
}
}
