integrate parallel_deblocking with CB4x4
this change makes parallel deblocking experiment works with
cb4x4. the inner loop process every 4x4 block.
Change-Id: I86adb3d7b6d67a91ccc12aab29da9bfb8c522cf1
diff --git a/aom_dsp/loopfilter.c b/aom_dsp/loopfilter.c
index 7420174..7ea1e6b 100644
--- a/aom_dsp/loopfilter.c
+++ b/aom_dsp/loopfilter.c
@@ -149,10 +149,15 @@
const uint8_t *blimit, const uint8_t *limit,
const uint8_t *thresh) {
int i;
+#if CONFIG_PARALLEL_DEBLOCKING && CONFIG_CB4X4
+ int count = 4;
+#else
+ int count = 8;
+#endif
// loop filter designed to work using chars so that we can make maximum use
// of 8 bit simd instructions.
- for (i = 0; i < 8; ++i) {
+ for (i = 0; i < count; ++i) {
#if !CONFIG_PARALLEL_DEBLOCKING
const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p];
const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p];
@@ -179,10 +184,15 @@
void aom_lpf_vertical_4_c(uint8_t *s, int pitch, const uint8_t *blimit,
const uint8_t *limit, const uint8_t *thresh) {
int i;
+#if CONFIG_PARALLEL_DEBLOCKING && CONFIG_CB4X4
+ int count = 4;
+#else
+ int count = 8;
+#endif
// loop filter designed to work using chars so that we can make maximum use
// of 8 bit simd instructions.
- for (i = 0; i < 8; ++i) {
+ for (i = 0; i < count; ++i) {
#if !CONFIG_PARALLEL_DEBLOCKING
const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
const uint8_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
@@ -229,10 +239,15 @@
void aom_lpf_horizontal_8_c(uint8_t *s, int p, const uint8_t *blimit,
const uint8_t *limit, const uint8_t *thresh) {
int i;
+#if CONFIG_PARALLEL_DEBLOCKING && CONFIG_CB4X4
+ int count = 4;
+#else
+ int count = 8;
+#endif
// loop filter designed to work using chars so that we can make maximum use
// of 8 bit simd instructions.
- for (i = 0; i < 8; ++i) {
+ for (i = 0; i < count; ++i) {
const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p];
const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p];
@@ -256,8 +271,13 @@
void aom_lpf_vertical_8_c(uint8_t *s, int pitch, const uint8_t *blimit,
const uint8_t *limit, const uint8_t *thresh) {
int i;
+#if CONFIG_PARALLEL_DEBLOCKING && CONFIG_CB4X4
+ int count = 4;
+#else
+ int count = 8;
+#endif
- for (i = 0; i < 8; ++i) {
+ for (i = 0; i < count; ++i) {
const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
const uint8_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
const int8_t mask =
@@ -390,10 +410,15 @@
const uint8_t *limit,
const uint8_t *thresh, int count) {
int i;
+#if CONFIG_PARALLEL_DEBLOCKING && CONFIG_CB4X4
+ int step = 4;
+#else
+ int step = 8;
+#endif
// loop filter designed to work using chars so that we can make maximum use
// of 8 bit simd instructions.
- for (i = 0; i < 8 * count; ++i) {
+ for (i = 0; i < step * count; ++i) {
const uint8_t p7 = s[-8 * p], p6 = s[-7 * p], p5 = s[-6 * p],
p4 = s[-5 * p], p3 = s[-4 * p], p2 = s[-3 * p],
p1 = s[-2 * p], p0 = s[-p];
@@ -436,7 +461,11 @@
void aom_lpf_horizontal_edge_16_c(uint8_t *s, int p, const uint8_t *blimit,
const uint8_t *limit, const uint8_t *thresh) {
+#if CONFIG_PARALLEL_DEBLOCKING && CONFIG_CB4X4
+ mb_lpf_horizontal_edge_w(s, p, blimit, limit, thresh, 1);
+#else
mb_lpf_horizontal_edge_w(s, p, blimit, limit, thresh, 2);
+#endif
}
static void mb_lpf_vertical_edge_w(uint8_t *s, int p, const uint8_t *blimit,
@@ -478,7 +507,11 @@
void aom_lpf_vertical_16_c(uint8_t *s, int p, const uint8_t *blimit,
const uint8_t *limit, const uint8_t *thresh) {
+#if CONFIG_PARALLEL_DEBLOCKING && CONFIG_CB4X4
+ mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 4);
+#else
mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 8);
+#endif
}
void aom_lpf_vertical_16_dual_c(uint8_t *s, int p, const uint8_t *blimit,
@@ -596,10 +629,15 @@
const uint8_t *blimit, const uint8_t *limit,
const uint8_t *thresh, int bd) {
int i;
+#if CONFIG_PARALLEL_DEBLOCKING && CONFIG_CB4X4
+ int count = 4;
+#else
+ int count = 8;
+#endif
// loop filter designed to work using chars so that we can make maximum use
// of 8 bit simd instructions.
- for (i = 0; i < 8; ++i) {
+ for (i = 0; i < count; ++i) {
#if !CONFIG_PARALLEL_DEBLOCKING
const uint16_t p3 = s[-4 * p];
const uint16_t p2 = s[-3 * p];
@@ -636,10 +674,15 @@
const uint8_t *limit, const uint8_t *thresh,
int bd) {
int i;
+#if CONFIG_PARALLEL_DEBLOCKING && CONFIG_CB4X4
+ int count = 4;
+#else
+ int count = 8;
+#endif
// loop filter designed to work using chars so that we can make maximum use
// of 8 bit simd instructions.
- for (i = 0; i < 8; ++i) {
+ for (i = 0; i < count; ++i) {
#if !CONFIG_PARALLEL_DEBLOCKING
const uint16_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
const uint16_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
@@ -689,10 +732,15 @@
const uint8_t *limit, const uint8_t *thresh,
int bd) {
int i;
+#if CONFIG_PARALLEL_DEBLOCKING && CONFIG_CB4X4
+ int count = 4;
+#else
+ int count = 8;
+#endif
// loop filter designed to work using chars so that we can make maximum use
// of 8 bit simd instructions.
- for (i = 0; i < 8; ++i) {
+ for (i = 0; i < count; ++i) {
const uint16_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p];
const uint16_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p];
@@ -718,8 +766,13 @@
const uint8_t *limit, const uint8_t *thresh,
int bd) {
int i;
+#if CONFIG_PARALLEL_DEBLOCKING && CONFIG_CB4X4
+ int count = 4;
+#else
+ int count = 8;
+#endif
- for (i = 0; i < 8; ++i) {
+ for (i = 0; i < count; ++i) {
const uint16_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
const uint16_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
const int8_t mask =
@@ -813,10 +866,15 @@
const uint8_t *thresh, int count,
int bd) {
int i;
+#if CONFIG_PARALLEL_DEBLOCKING && CONFIG_CB4X4
+ int step = 4;
+#else
+ int step = 8;
+#endif
// loop filter designed to work using chars so that we can make maximum use
// of 8 bit simd instructions.
- for (i = 0; i < 8 * count; ++i) {
+ for (i = 0; i < step * count; ++i) {
const uint16_t p3 = s[-4 * p];
const uint16_t p2 = s[-3 * p];
const uint16_t p1 = s[-2 * p];
@@ -852,7 +910,11 @@
const uint8_t *blimit,
const uint8_t *limit,
const uint8_t *thresh, int bd) {
+#if CONFIG_PARALLEL_DEBLOCKING && CONFIG_CB4X4
+ highbd_mb_lpf_horizontal_edge_w(s, p, blimit, limit, thresh, 1, bd);
+#else
highbd_mb_lpf_horizontal_edge_w(s, p, blimit, limit, thresh, 2, bd);
+#endif
}
static void highbd_mb_lpf_vertical_edge_w(uint16_t *s, int p,
@@ -888,13 +950,21 @@
void aom_highbd_lpf_vertical_16_c(uint16_t *s, int p, const uint8_t *blimit,
const uint8_t *limit, const uint8_t *thresh,
int bd) {
+#if CONFIG_PARALLEL_DEBLOCKING && CONFIG_CB4X4
+ highbd_mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 4, bd);
+#else
highbd_mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 8, bd);
+#endif
}
void aom_highbd_lpf_vertical_16_dual_c(uint16_t *s, int p,
const uint8_t *blimit,
const uint8_t *limit,
const uint8_t *thresh, int bd) {
+#if CONFIG_PARALLEL_DEBLOCKING && CONFIG_CB4X4
+ highbd_mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 8, bd);
+#else
highbd_mb_lpf_vertical_edge_w(s, p, blimit, limit, thresh, 16, bd);
+#endif
}
#endif // CONFIG_HIGHBITDEPTH
