void
png_read_filter_row_sub4_neon(png_row_infop row_info, png_bytep row,
png_const_bytep prev_row)
{
png_bytep rp = row;
png_bytep rp_stop = row + row_info->rowbytes;
uint8x8x4_t vdest;
vdest.val[3] = vdup_n_u8(0);
png_debug(1, "in png_read_filter_row_sub4_neon");
for (; rp < rp_stop; rp += 16)
{
uint32x2x4_t vtmp = vld4_u32(png_ptr(uint32_t,rp));
uint8x8x4_t *vrpt = png_ptr(uint8x8x4_t,&vtmp);
uint8x8x4_t vrp = *vrpt;
uint32x2x4_t *temp_pointer;
vdest.val[0] = vadd_u8(vdest.val[3], vrp.val[0]);
vdest.val[1] = vadd_u8(vdest.val[0], vrp.val[1]);
vdest.val[2] = vadd_u8(vdest.val[1], vrp.val[2]);
vdest.val[3] = vadd_u8(vdest.val[2], vrp.val[3]);
vst4_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2x4_t,&vdest), 0);
}
PNG_UNUSED(prev_row)
}
void
png_read_filter_row_sub3_neon(png_row_infop row_info, png_bytep row,
png_const_bytep prev_row)
{
png_bytep rp = row;
png_bytep rp_stop = row + row_info->rowbytes;
uint8x16_t vtmp = vld1q_u8(rp);
uint8x8x2_t *vrpt = png_ptr(uint8x8x2_t, &vtmp);
uint8x8x2_t vrp = *vrpt;
uint8x8x4_t vdest;
vdest.val[3] = vdup_n_u8(0);
png_debug(1, "in png_read_filter_row_sub3_neon");
for (; rp < rp_stop;)
{
uint8x8_t vtmp1, vtmp2;
uint32x2_t *temp_pointer;
vtmp1 = vext_u8(vrp.val[0], vrp.val[1], 3);
vdest.val[0] = vadd_u8(vdest.val[3], vrp.val[0]);
vtmp2 = vext_u8(vrp.val[0], vrp.val[1], 6);
vdest.val[1] = vadd_u8(vdest.val[0], vtmp1);
vtmp1 = vext_u8(vrp.val[1], vrp.val[1], 1);
vdest.val[2] = vadd_u8(vdest.val[1], vtmp2);
vdest.val[3] = vadd_u8(vdest.val[2], vtmp1);
vtmp = vld1q_u8(rp + 12);
vrpt = png_ptr(uint8x8x2_t, &vtmp);
vrp = *vrpt;
vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[0]), 0);
rp += 3;
vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[1]), 0);
rp += 3;
vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[2]), 0);
rp += 3;
vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[3]), 0);
rp += 3;
}
PNG_UNUSED(prev_row)
}
void
png_read_filter_row_paeth4_neon(png_row_infop row_info, png_bytep row,
png_const_bytep prev_row)
{
png_bytep rp = row;
png_bytep rp_stop = row + row_info->rowbytes;
png_const_bytep pp = prev_row;
uint8x8_t vlast = vdup_n_u8(0);
uint8x8x4_t vdest;
vdest.val[3] = vdup_n_u8(0);
png_debug(1, "in png_read_filter_row_paeth4_neon");
for (; rp < rp_stop; rp += 16, pp += 16)
{
uint32x2x4_t vtmp;
uint8x8x4_t *vrpt, *vppt;
uint8x8x4_t vrp, vpp;
uint32x2x4_t *temp_pointer;
vtmp = vld4_u32(png_ptr(uint32_t,rp));
vrpt = png_ptr(uint8x8x4_t,&vtmp);
vrp = *vrpt;
vtmp = vld4_u32(png_ptrc(uint32_t,pp));
vppt = png_ptr(uint8x8x4_t,&vtmp);
vpp = *vppt;
vdest.val[0] = paeth(vdest.val[3], vpp.val[0], vlast);
vdest.val[0] = vadd_u8(vdest.val[0], vrp.val[0]);
vdest.val[1] = paeth(vdest.val[0], vpp.val[1], vpp.val[0]);
vdest.val[1] = vadd_u8(vdest.val[1], vrp.val[1]);
vdest.val[2] = paeth(vdest.val[1], vpp.val[2], vpp.val[1]);
vdest.val[2] = vadd_u8(vdest.val[2], vrp.val[2]);
vdest.val[3] = paeth(vdest.val[2], vpp.val[3], vpp.val[2]);
vdest.val[3] = vadd_u8(vdest.val[3], vrp.val[3]);
vlast = vpp.val[3];
vst4_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2x4_t,&vdest), 0);
}
}
void
png_read_filter_row_paeth3_neon(png_row_infop row_info, png_bytep row,
png_const_bytep prev_row)
{
png_bytep rp = row;
png_const_bytep pp = prev_row;
png_bytep rp_stop = row + row_info->rowbytes;
uint8x16_t vtmp;
uint8x8x2_t *vrpt;
uint8x8x2_t vrp;
uint8x8_t vlast = vdup_n_u8(0);
uint8x8x4_t vdest;
vdest.val[3] = vdup_n_u8(0);
vtmp = vld1q_u8(rp);
vrpt = png_ptr(uint8x8x2_t,&vtmp);
vrp = *vrpt;
png_debug(1, "in png_read_filter_row_paeth3_neon");
for (; rp < rp_stop; pp += 12)
{
uint8x8x2_t *vppt;
uint8x8x2_t vpp;
uint8x8_t vtmp1, vtmp2, vtmp3;
uint32x2_t *temp_pointer;
vtmp = vld1q_u8(pp);
vppt = png_ptr(uint8x8x2_t,&vtmp);
vpp = *vppt;
vdest.val[0] = paeth(vdest.val[3], vpp.val[0], vlast);
vdest.val[0] = vadd_u8(vdest.val[0], vrp.val[0]);
vtmp1 = vext_u8(vrp.val[0], vrp.val[1], 3);
vtmp2 = vext_u8(vpp.val[0], vpp.val[1], 3);
vdest.val[1] = paeth(vdest.val[0], vtmp2, vpp.val[0]);
vdest.val[1] = vadd_u8(vdest.val[1], vtmp1);
vtmp1 = vext_u8(vrp.val[0], vrp.val[1], 6);
vtmp3 = vext_u8(vpp.val[0], vpp.val[1], 6);
vdest.val[2] = paeth(vdest.val[1], vtmp3, vtmp2);
vdest.val[2] = vadd_u8(vdest.val[2], vtmp1);
vtmp1 = vext_u8(vrp.val[1], vrp.val[1], 1);
vtmp2 = vext_u8(vpp.val[1], vpp.val[1], 1);
vtmp = vld1q_u8(rp + 12);
vrpt = png_ptr(uint8x8x2_t,&vtmp);
vrp = *vrpt;
vdest.val[3] = paeth(vdest.val[2], vtmp2, vtmp3);
vdest.val[3] = vadd_u8(vdest.val[3], vtmp1);
vlast = vtmp2;
vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[0]), 0);
rp += 3;
vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[1]), 0);
rp += 3;
vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[2]), 0);
rp += 3;
vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[3]), 0);
rp += 3;
}
}
inline uint8x8_t vadd(const uint8x8_t & v0, const uint8x8_t & v1) { return vadd_u8 (v0, v1); }