clean-up and slight speed-up in-loop filtering SSE2

* remove some sign-bit flipping
* turn some macro into inline functions
* fix some 'const' in signatures
* clarify the int8/uint8 usage

Change-Id: Ib04459ac34cb280c57579c5d79a5efd2f8d5e99d
This commit is contained in:
Pascal Massimino
2014-05-19 03:35:05 -07:00
parent 54bfffcabc
commit f60957bfd2
2 changed files with 172 additions and 158 deletions

View File

@ -481,8 +481,8 @@ const VP8PredFunc VP8PredChroma8[NUM_B_DC_MODES] = {
// 4 pixels in, 2 pixels out
static WEBP_INLINE void do_filter2(uint8_t* p, int step) {
const int p1 = p[-2*step], p0 = p[-step], q0 = p[0], q1 = p[step];
const int a = 3 * (q0 - p0) + VP8ksclip1[p1 - q1];
const int a1 = VP8ksclip2[(a + 4) >> 3];
const int a = 3 * (q0 - p0) + VP8ksclip1[p1 - q1]; // in [-893,892]
const int a1 = VP8ksclip2[(a + 4) >> 3]; // in [-16,15]
const int a2 = VP8ksclip2[(a + 3) >> 3];
p[-step] = VP8kclip1[p0 + a2];
p[ 0] = VP8kclip1[q0 - a1];
@ -506,6 +506,7 @@ static WEBP_INLINE void do_filter6(uint8_t* p, int step) {
const int p2 = p[-3*step], p1 = p[-2*step], p0 = p[-step];
const int q0 = p[0], q1 = p[step], q2 = p[2*step];
const int a = VP8ksclip1[3 * (q0 - p0) + VP8ksclip1[p1 - q1]];
// a is in [-128,127], a1 in [-27,27], a2 in [-18,18] and a3 in [-9,9]
const int a1 = (27 * a + 63) >> 7; // eq. to ((3 * a + 7) * 9) >> 7
const int a2 = (18 * a + 63) >> 7; // eq. to ((2 * a + 7) * 9) >> 7
const int a3 = (9 * a + 63) >> 7; // eq. to ((1 * a + 7) * 9) >> 7