generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

ppc: d45 predictor 16x16

Browse Source 
About 16x faster.

Change-Id: Ie5469fb32d5fd11bb6cb06318cea475d8a5b00b9
This commit is contained in:
Luca Barbato 2017-04-08 22:41:41 +00:00 committed by James Zern
parent c08baa2900
commit 7a7dc9e624
3 changed files with 32 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
test/test_intra_pred_speed.cc
View File

@ -319,8 +319,8 @@ INTRA_PRED_TEST(VSX, TestIntraPred8, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
INTRA_PRED_TEST(VSX, TestIntraPred16, vpx_dc_predictor_16x16_vsx,
vpx_dc_left_predictor_16x16_vsx, vpx_dc_top_predictor_16x16_vsx,
vpx_dc_128_predictor_16x16_vsx, vpx_v_predictor_16x16_vsx,
vpx_h_predictor_16x16_vsx, NULL, NULL, NULL, NULL, NULL, NULL,
vpx_tm_predictor_16x16_vsx)
vpx_h_predictor_16x16_vsx, vpx_d45_predictor_16x16_vsx, NULL,
NULL, NULL, NULL, NULL, vpx_tm_predictor_16x16_vsx)
INTRA_PRED_TEST(VSX, TestIntraPred32, vpx_dc_predictor_32x32_vsx,
vpx_dc_left_predictor_32x32_vsx, vpx_dc_top_predictor_32x32_vsx,

29
vpx_dsp/ppc/intrapred_vsx.c
View File

@ -489,3 +489,32 @@ void vpx_dc_predictor_32x32_vsx(uint8_t *dst, ptrdiff_t stride,
const uint8_t *above, const uint8_t *left) {
dc_fill_predictor_32x32(dst, stride, dc_avg32(above, left));
}
static uint8x16_t avg3(const uint8x16_t a, const uint8x16_t b,
const uint8x16_t c) {
const uint8x16_t ac =
vec_adds(vec_and(a, c), vec_sr(vec_xor(a, c), vec_splat_u8(1)));
return vec_avg(ac, b);
}
// Workaround vec_sld/vec_xxsldi/vec_lsdoi being missing or broken.
static const uint8x16_t sl1 = { 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8,
0x9, 0xA, 0xB, 0xC, 0xD, 0xE, 0xF, 0x10 };
void vpx_d45_predictor_16x16_vsx(uint8_t *dst, ptrdiff_t stride,
const uint8_t *above, const uint8_t *left) {
const uint8x16_t a = vec_vsx_ld(0, above);
const uint8x16_t above_right = vec_splat(a, 15);
const uint8x16_t b = vec_perm(a, above_right, sl1);
const uint8x16_t c = vec_perm(b, above_right, sl1);
uint8x16_t row = avg3(a, b, c);
int i;
(void)left;
for (i = 0; i < 16; i++) {
vec_vsx_st(row, 0, dst);
dst += stride;
row = vec_perm(row, above_right, sl1);
}
}

2
vpx_dsp/vpx_dsp_rtcd_defs.pl
View File

@ -113,7 +113,7 @@ add_proto qw/void vpx_d207_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride,
specialize qw/vpx_d207_predictor_16x16 ssse3/;
add_proto qw/void vpx_d45_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vpx_d45_predictor_16x16 neon ssse3/;
specialize qw/vpx_d45_predictor_16x16 neon ssse3 vsx/;
add_proto qw/void vpx_d63_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vpx_d63_predictor_16x16 ssse3/;