Merge "vp9: Put lower limit on frames_to_key adjustment to gf interval."
diff --git a/vp10/common/vp10_inv_txfm.c b/vp10/common/vp10_inv_txfm.c
index 00110f3..c31275e6 100644
--- a/vp10/common/vp10_inv_txfm.c
+++ b/vp10/common/vp10_inv_txfm.c
@@ -56,10 +56,10 @@
c1 = e1 - c1;
a1 -= b1;
d1 += c1;
- dest[stride * 0] = clip_pixel_add(dest[stride * 0], a1);
- dest[stride * 1] = clip_pixel_add(dest[stride * 1], b1);
- dest[stride * 2] = clip_pixel_add(dest[stride * 2], c1);
- dest[stride * 3] = clip_pixel_add(dest[stride * 3], d1);
+ dest[stride * 0] = clip_pixel_add(dest[stride * 0], WRAPLOW(a1));
+ dest[stride * 1] = clip_pixel_add(dest[stride * 1], WRAPLOW(b1));
+ dest[stride * 2] = clip_pixel_add(dest[stride * 2], WRAPLOW(c1));
+ dest[stride * 3] = clip_pixel_add(dest[stride * 3], WRAPLOW(d1));
ip++;
dest++;
@@ -268,7 +268,7 @@
s4 = sinpi_1_9 * x2;
s5 = sinpi_2_9 * x3;
s6 = sinpi_4_9 * x3;
- s7 = x0 - x2 + x3;
+ s7 = WRAPLOW(x0 - x2 + x3);
s0 = s0 + s3 + s5;
s1 = s1 - s4 - s6;
@@ -1290,11 +1290,14 @@
c1 = e1 - c1;
a1 -= b1;
d1 += c1;
- dest[stride * 0] = highbd_clip_pixel_add(dest[stride * 0], a1, bd);
- dest[stride * 1] = highbd_clip_pixel_add(dest[stride * 1], b1, bd);
- dest[stride * 2] = highbd_clip_pixel_add(dest[stride * 2], c1, bd);
- dest[stride * 3] = highbd_clip_pixel_add(dest[stride * 3], d1, bd);
-
+ dest[stride * 0] = highbd_clip_pixel_add(dest[stride * 0],
+ HIGHBD_WRAPLOW(a1, bd), bd);
+ dest[stride * 1] = highbd_clip_pixel_add(dest[stride * 1],
+ HIGHBD_WRAPLOW(b1, bd), bd);
+ dest[stride * 2] = highbd_clip_pixel_add(dest[stride * 2],
+ HIGHBD_WRAPLOW(c1, bd), bd);
+ dest[stride * 3] = highbd_clip_pixel_add(dest[stride * 3],
+ HIGHBD_WRAPLOW(d1, bd), bd);
ip++;
dest++;
}
@@ -1510,7 +1513,7 @@
s4 = sinpi_1_9 * x2;
s5 = sinpi_2_9 * x3;
s6 = sinpi_4_9 * x3;
- s7 = (tran_high_t)(x0 - x2 + x3);
+ s7 = (tran_high_t)HIGHBD_WRAPLOW(x0 - x2 + x3, bd);
s0 = s0 + s3 + s5;
s1 = s1 - s4 - s6;
diff --git a/vp9/encoder/x86/vp9_diamond_search_sad_avx.c b/vp9/encoder/x86/vp9_diamond_search_sad_avx.c
index e7c4003..cd3e87e 100644
--- a/vp9/encoder/x86/vp9_diamond_search_sad_avx.c
+++ b/vp9/encoder/x86/vp9_diamond_search_sad_avx.c
@@ -56,7 +56,7 @@
}
/*****************************************************************************
- * This function utilises 3 properties of the cost function lookup tables, *
+ * This function utilizes 3 properties of the cost function lookup tables, *
* constructed in using 'cal_nmvjointsadcost' and 'cal_nmvsadcosts' in *
* vp9_encoder.c. *
* For the joint cost: *
@@ -122,10 +122,7 @@
#endif
unsigned int best_sad;
-
- int i;
- int j;
- int step;
+ int i, j, step;
// Check the prerequisite cost function properties that are easy to check
// in an assert. See the function-level documentation for details on all
@@ -141,11 +138,7 @@
for (i = 0, step = 0; step < tot_steps; step++) {
for (j = 0; j < cfg->searches_per_step; j += 4, i += 4) {
- __m128i v_sad_d;
- __m128i v_cost_d;
- __m128i v_outside_d;
- __m128i v_inside_d;
- __m128i v_diff_mv_w;
+ __m128i v_sad_d, v_cost_d, v_outside_d, v_inside_d, v_diff_mv_w;
#if ARCH_X86_64
__m128i v_blocka[2];
#else
@@ -153,7 +146,7 @@
#endif
// Compute the candidate motion vectors
- const __m128i v_ss_mv_w = _mm_loadu_si128((const __m128i*)&ss_mv[i]);
+ const __m128i v_ss_mv_w = _mm_loadu_si128((const __m128i *)&ss_mv[i]);
const __m128i v_these_mv_w = _mm_add_epi16(v_bmv_w, v_ss_mv_w);
// Clamp them to the search bounds
__m128i v_these_mv_clamp_w = v_these_mv_w;
@@ -185,8 +178,8 @@
{
#if ARCH_X86_64 // sizeof(intptr_t) == 8
// Load the offsets
- __m128i v_bo10_q = _mm_loadu_si128((const __m128i*)&ss_os[i+0]);
- __m128i v_bo32_q = _mm_loadu_si128((const __m128i*)&ss_os[i+2]);
+ __m128i v_bo10_q = _mm_loadu_si128((const __m128i *)&ss_os[i + 0]);
+ __m128i v_bo32_q = _mm_loadu_si128((const __m128i *)&ss_os[i + 2]);
// Set the ones falling outside to zero
v_bo10_q = _mm_and_si128(v_bo10_q,
_mm_cvtepi32_epi64(v_inside_d));
@@ -196,7 +189,7 @@
v_blocka[0] = _mm_add_epi64(v_ba_q, v_bo10_q);
v_blocka[1] = _mm_add_epi64(v_ba_q, v_bo32_q);
#else // ARCH_X86 // sizeof(intptr_t) == 4
- __m128i v_bo_d = _mm_loadu_si128((const __m128i*)&ss_os[i]);
+ __m128i v_bo_d = _mm_loadu_si128((const __m128i *)&ss_os[i]);
v_bo_d = _mm_and_si128(v_bo_d, v_inside_d);
v_blocka[0] = _mm_add_epi32(v_ba_d, v_bo_d);
#endif
@@ -224,13 +217,10 @@
const uint32_t cost3 = x->nmvsadcost[0][row3] + x->nmvsadcost[0][col3];
__m128i v_cost_10_d, v_cost_32_d;
-
v_cost_10_d = _mm_cvtsi32_si128(cost0);
v_cost_10_d = _mm_insert_epi32(v_cost_10_d, cost1, 1);
-
v_cost_32_d = _mm_cvtsi32_si128(cost2);
v_cost_32_d = _mm_insert_epi32(v_cost_32_d, cost3, 1);
-
v_cost_d = _mm_unpacklo_epi64(v_cost_10_d, v_cost_32_d);
}
@@ -246,7 +236,7 @@
// Multiply by sad_per_bit
v_cost_d = _mm_mullo_epi32(v_cost_d, v_spb_d);
- // ROUND_POWER_OF_TWO(v_cost_d, 8)
+ // ROUND_POWER_OF_TWO(v_cost_d, VP9_PROB_COST_SHIFT)
v_cost_d = _mm_add_epi32(v_cost_d,
_mm_set1_epi32(1 << (VP9_PROB_COST_SHIFT - 1)));
v_cost_d = _mm_srai_epi32(v_cost_d, VP9_PROB_COST_SHIFT);
diff --git a/vpx_dsp/inv_txfm.c b/vpx_dsp/inv_txfm.c
index ad91ead..e18d31d 100644
--- a/vpx_dsp/inv_txfm.c
+++ b/vpx_dsp/inv_txfm.c
@@ -56,10 +56,10 @@
c1 = e1 - c1;
a1 -= b1;
d1 += c1;
- dest[stride * 0] = clip_pixel_add(dest[stride * 0], a1);
- dest[stride * 1] = clip_pixel_add(dest[stride * 1], b1);
- dest[stride * 2] = clip_pixel_add(dest[stride * 2], c1);
- dest[stride * 3] = clip_pixel_add(dest[stride * 3], d1);
+ dest[stride * 0] = clip_pixel_add(dest[stride * 0], WRAPLOW(a1));
+ dest[stride * 1] = clip_pixel_add(dest[stride * 1], WRAPLOW(b1));
+ dest[stride * 2] = clip_pixel_add(dest[stride * 2], WRAPLOW(c1));
+ dest[stride * 3] = clip_pixel_add(dest[stride * 3], WRAPLOW(d1));
ip++;
dest++;
@@ -266,7 +266,7 @@
s4 = sinpi_1_9 * x2;
s5 = sinpi_2_9 * x3;
s6 = sinpi_4_9 * x3;
- s7 = x0 - x2 + x3;
+ s7 = WRAPLOW(x0 - x2 + x3);
s0 = s0 + s3 + s5;
s1 = s1 - s4 - s6;
@@ -1309,10 +1309,14 @@
c1 = e1 - c1;
a1 -= b1;
d1 += c1;
- dest[stride * 0] = highbd_clip_pixel_add(dest[stride * 0], a1, bd);
- dest[stride * 1] = highbd_clip_pixel_add(dest[stride * 1], b1, bd);
- dest[stride * 2] = highbd_clip_pixel_add(dest[stride * 2], c1, bd);
- dest[stride * 3] = highbd_clip_pixel_add(dest[stride * 3], d1, bd);
+ dest[stride * 0] = highbd_clip_pixel_add(dest[stride * 0],
+ HIGHBD_WRAPLOW(a1, bd), bd);
+ dest[stride * 1] = highbd_clip_pixel_add(dest[stride * 1],
+ HIGHBD_WRAPLOW(b1, bd), bd);
+ dest[stride * 2] = highbd_clip_pixel_add(dest[stride * 2],
+ HIGHBD_WRAPLOW(c1, bd), bd);
+ dest[stride * 3] = highbd_clip_pixel_add(dest[stride * 3],
+ HIGHBD_WRAPLOW(d1, bd), bd);
ip++;
dest++;
@@ -1529,7 +1533,7 @@
s4 = sinpi_1_9 * x2;
s5 = sinpi_2_9 * x3;
s6 = sinpi_4_9 * x3;
- s7 = (tran_high_t)(x0 - x2 + x3);
+ s7 = (tran_high_t)HIGHBD_WRAPLOW(x0 - x2 + x3, bd);
s0 = s0 + s3 + s5;
s1 = s1 - s4 - s6;
