mid-kid
/
i3lock-color
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
Browse Source

Slightly refactor border handling code.

master
Sebastian Frysztak 8 years ago
parent
f06dc6cbc4
commit
252999f640
1 changed files with 38 additions and 36 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 74
      blur_simd.c

74
blur_simd.c
View File

@ -63,24 +63,25 @@ void blur_impl_horizontal_pass_sse2(uint32_t *src, uint32_t *dst, float *kernel,
// handle borders // handle borders
int leftBorder = column < HALF_KERNEL; int leftBorder = column < HALF_KERNEL;
int rightBorder = column > width - HALF_KERNEL; int rightBorder = column > width - HALF_KERNEL;
if (leftBorder || rightBorder) { uint32_t _rgbaIn[KERNEL_SIZE] ALIGN16;
uint32_t _rgbaIn[KERNEL_SIZE] ALIGN16; int i = 0;
int i = 0; if (leftBorder) {
if (leftBorder) { // for kernel size 7x7 and column == 0, we have:
// for kernel size 7x7 and column == 0, we have: // x x x P0 P1 P2 P3
// x x x P0 P1 P2 P3 // first loop mirrors P{0..3} to fill x's,
// first loop mirrors P{0..3} to fill x's, // second one loads P{0..3}
// second one loads P{0..3} for (; i < HALF_KERNEL - column; i++)
for (; i < HALF_KERNEL - column; i++) _rgbaIn[i] = *(src + (HALF_KERNEL - i));
_rgbaIn[i] = *(src + (HALF_KERNEL - i)); for (; i < KERNEL_SIZE; i++)
for (; i < KERNEL_SIZE; i++) _rgbaIn[i] = *(src - (HALF_KERNEL - i));
_rgbaIn[i] = *(src - (HALF_KERNEL - i));
} else { for (int k = 0; k < REGISTERS_CNT; k++)
for (; i < width - column; i++) rgbaIn[k] = _mm_load_si128((__m128i*)(_rgbaIn + 4*k));
_rgbaIn[i] = *(src + i); } else if (rightBorder) {
for (int k = 0; i < KERNEL_SIZE; i++, k++) for (; i < width - column; i++)
_rgbaIn[i] = *(src - k); _rgbaIn[i] = *(src + i);
} for (int k = 0; i < KERNEL_SIZE; i++, k++)
_rgbaIn[i] = *(src - k);
for (int k = 0; k < REGISTERS_CNT; k++) for (int k = 0; k < REGISTERS_CNT; k++)
rgbaIn[k] = _mm_load_si128((__m128i*)(_rgbaIn + 4*k)); rgbaIn[k] = _mm_load_si128((__m128i*)(_rgbaIn + 4*k));
@ -166,24 +167,25 @@ void blur_impl_horizontal_pass_avx(uint32_t *src, uint32_t *dst, float *kernel,
// handle borders // handle borders
int leftBorder = column < HALF_KERNEL; int leftBorder = column < HALF_KERNEL;
int rightBorder = column > width - HALF_KERNEL; int rightBorder = column > width - HALF_KERNEL;
if (leftBorder || rightBorder) { uint32_t _rgbaIn[KERNEL_SIZE] ALIGN16;
uint32_t _rgbaIn[KERNEL_SIZE] ALIGN16; int i = 0;
int i = 0; if (leftBorder) {
if (leftBorder) { // for kernel size 7x7 and column == 0, we have:
// for kernel size 7x7 and column == 0, we have: // x x x P0 P1 P2 P3
// x x x P0 P1 P2 P3 // first loop mirrors P{0..3} to fill x's,
// first loop mirrors P{0..3} to fill x's, // second one loads P{0..3}
// second one loads P{0..3} for (; i < HALF_KERNEL - column; i++)
for (; i < HALF_KERNEL - column; i++) _rgbaIn[i] = *(src + (HALF_KERNEL - i));
_rgbaIn[i] = *(src + (HALF_KERNEL - i)); for (; i < KERNEL_SIZE; i++)
for (; i < KERNEL_SIZE; i++) _rgbaIn[i] = *(src - (HALF_KERNEL - i));
_rgbaIn[i] = *(src - (HALF_KERNEL - i));
} else { for (int k = 0; k < REGISTERS_CNT; k++)
for (; i < width - column; i++) rgbaIn[k] = _mm_load_si128((__m128i*)(_rgbaIn + 4*k));
_rgbaIn[i] = *(src + i); } else if (rightBorder) {
for (int k = 0; i < KERNEL_SIZE; i++, k++) for (; i < width - column; i++)
_rgbaIn[i] = *(src - k); _rgbaIn[i] = *(src + i);
} for (int k = 0; i < KERNEL_SIZE; i++, k++)
_rgbaIn[i] = *(src - k);
for (int k = 0; k < REGISTERS_CNT; k++) for (int k = 0; k < REGISTERS_CNT; k++)
rgbaIn[k] = _mm_load_si128((__m128i*)(_rgbaIn + 4*k)); rgbaIn[k] = _mm_load_si128((__m128i*)(_rgbaIn + 4*k));

Write Preview
Loading…
Cancel
Save