#[cfg(target_arch = "x86")]
use std::arch::x86::*;
#[cfg(target_arch = "x86_64")]
use std::arch::x86_64::*;
use crate::EdgeMode;
use crate::edge_mode::clamp_edge;
use crate::sse::{load_f32_f16, store_f32_f16};
use crate::unsafe_slice::UnsafeSlice;
use core::f16;
pub(crate) fn fg_horizontal_pass_sse_f16<const CN: usize>(
undefined_slice: &UnsafeSlice<f16>,
stride: u32,
width: u32,
height: u32,
radius: u32,
start: u32,
end: u32,
edge_mode: EdgeMode,
) {
unsafe {
fast_gaussian_horizontal_pass_sse_f16_impl::<CN>(
undefined_slice,
stride,
width,
height,
radius,
start,
end,
edge_mode,
);
}
}
#[inline]
#[target_feature(enable = "sse4.1,f16c")]
fn fast_gaussian_horizontal_pass_sse_f16_impl<const CN: usize>(
bytes: &UnsafeSlice<f16>,
stride: u32,
width: u32,
height: u32,
radius: u32,
start: u32,
end: u32,
edge_mode: EdgeMode,
) {
unsafe {
let mut buffer: [[f32; 4]; 1024] = [[0.; 4]; 1024];
let radius_64 = radius as i64;
let width_wide = width as i64;
let v_half = _mm_set1_ps(2.);
let v_weight = _mm_set1_ps(1f32 / (radius as f32 * radius as f32));
for y in start..std::cmp::min(height, end) {
let mut diffs = _mm_setzero_ps();
let mut summs = _mm_setzero_ps();
let current_y = ((y as i64) * (stride as i64)) as usize;
let start_x = 0 - 2 * radius_64;
for x in start_x..(width as i64) {
if x >= 0 {
let current_px = x as usize * CN;
let pixel = _mm_mul_ps(summs, v_weight);
let bytes_offset = current_y + current_px;
let dst_ptr = bytes.get_ptr(bytes_offset);
store_f32_f16::<CN>(dst_ptr, pixel);
let arr_index = ((x - radius_64) & 1023) as usize;
let d_arr_index = (x & 1023) as usize;
let d_buf_ptr = buffer.get_unchecked(d_arr_index).as_ptr();
let mut d_stored = _mm_loadu_ps(d_buf_ptr);
d_stored = _mm_mul_ps(d_stored, v_half);
let buf_ptr = buffer.get_unchecked(arr_index).as_ptr();
let a_stored = _mm_loadu_ps(buf_ptr);
diffs = _mm_add_ps(diffs, _mm_sub_ps(a_stored, d_stored));
} else if x + radius_64 >= 0 {
let arr_index = (x & 1023) as usize;
let buf_ptr = buffer.get_unchecked(arr_index).as_ptr();
let mut stored = _mm_loadu_ps(buf_ptr);
stored = _mm_mul_ps(stored, v_half);
diffs = _mm_sub_ps(diffs, stored);
}
let next_row_y = (y as usize) * (stride as usize);
let next_row_x = clamp_edge!(edge_mode, x + radius_64, 0, width_wide);
let next_row_px = next_row_x * CN;
let s_ptr = bytes.get_ptr(next_row_y + next_row_px);
let pixel_color = load_f32_f16::<CN>(s_ptr);
let arr_index = ((x + radius_64) & 1023) as usize;
let buf_ptr = buffer.get_unchecked_mut(arr_index).as_mut_ptr();
diffs = _mm_add_ps(diffs, pixel_color);
summs = _mm_add_ps(summs, diffs);
_mm_storeu_ps(buf_ptr, pixel_color);
}
}
}
}
pub(crate) fn fg_vertical_pass_sse_f16<T, const CN: usize>(
undefined_slice: &UnsafeSlice<T>,
stride: u32,
width: u32,
height: u32,
radius: u32,
start: u32,
end: u32,
edge_mode: EdgeMode,
) {
unsafe {
fast_gaussian_vertical_pass_sse_f16_impl::<T, CN>(
undefined_slice,
stride,
width,
height,
radius,
start,
end,
edge_mode,
);
}
}
#[inline]
#[target_feature(enable = "sse4.1,f16c")]
fn fast_gaussian_vertical_pass_sse_f16_impl<T, const CN: usize>(
undefined_slice: &UnsafeSlice<T>,
stride: u32,
width: u32,
height: u32,
radius: u32,
start: u32,
end: u32,
edge_mode: EdgeMode,
) {
unsafe {
let bytes: &UnsafeSlice<'_, f16> = std::mem::transmute(undefined_slice);
let mut buffer: [[f32; 4]; 1024] = [[0.; 4]; 1024];
let v_half = _mm_set1_ps(2.);
let v_weight = _mm_set1_ps(1f32 / (radius as f32 * radius as f32));
let height_wide = height as i64;
let radius_64 = radius as i64;
for x in start..std::cmp::min(width, end) {
let mut diffs = _mm_setzero_ps();
let mut summs = _mm_setzero_ps();
let start_y = 0 - 2 * radius as i64;
for y in start_y..height_wide {
let current_y = (y * (stride as i64)) as usize;
if y >= 0 {
let current_px = ((std::cmp::max(x, 0)) * CN as u32) as usize;
let pixel = _mm_mul_ps(summs, v_weight);
let bytes_offset = current_y + current_px;
store_f32_f16::<CN>(bytes.get_ptr(bytes_offset), pixel);
let arr_index = ((y - radius_64) & 1023) as usize;
let d_arr_index = (y & 1023) as usize;
let d_buf_ptr = buffer.get_unchecked(d_arr_index).as_ptr();
let mut d_stored = _mm_loadu_ps(d_buf_ptr);
d_stored = _mm_mul_ps(d_stored, v_half);
let buf_ptr = buffer.get_unchecked(arr_index).as_ptr();
let a_stored = _mm_loadu_ps(buf_ptr);
diffs = _mm_add_ps(diffs, _mm_sub_ps(a_stored, d_stored));
} else if y + radius_64 >= 0 {
let arr_index = (y & 1023) as usize;
let buf_ptr = buffer.get_unchecked(arr_index).as_ptr();
let mut stored = _mm_loadu_ps(buf_ptr);
stored = _mm_mul_ps(stored, v_half);
diffs = _mm_sub_ps(diffs, stored);
}
let next_row_y =
clamp_edge!(edge_mode, y + radius_64, 0, height_wide) * (stride as usize);
let next_row_x = (x * CN as u32) as usize;
let s_ptr = bytes.get_ptr(next_row_y + next_row_x);
let pixel_color = load_f32_f16::<CN>(s_ptr);
let arr_index = ((y + radius_64) & 1023) as usize;
let buf_ptr = buffer.get_unchecked_mut(arr_index).as_mut_ptr();
diffs = _mm_add_ps(diffs, pixel_color);
summs = _mm_add_ps(summs, diffs);
_mm_storeu_ps(buf_ptr, pixel_color);
}
}
}
}