vpxdsp: [x86] add highbd_dc_top_predictor functions

C vs SSE2 speed gains:
_4x4 : ~7.39x
_8x8 : ~11.36x
_16x16 : ~8.68x
_32x32 : ~4.33x

BUG=webm:1411

Change-Id: I7f1487cd1531d4e7f0fbb4596fed3bfb72a59d58
This commit is contained in:
Scott LaVarnway
2017-08-29 11:25:32 -07:00
parent 81f45ff798
commit f783e3a75d
4 changed files with 166 additions and 8 deletions

View File

@@ -481,25 +481,29 @@ HIGHBD_INTRA_PRED_TEST(
#if HAVE_SSE2
HIGHBD_INTRA_PRED_TEST(SSE2, TestHighbdIntraPred4,
vpx_highbd_dc_predictor_4x4_sse2, NULL, NULL, NULL,
vpx_highbd_dc_predictor_4x4_sse2, NULL,
vpx_highbd_dc_top_predictor_4x4_sse2, NULL,
vpx_highbd_v_predictor_4x4_sse2,
vpx_highbd_h_predictor_4x4_sse2, NULL, NULL, NULL, NULL,
NULL, NULL, vpx_highbd_tm_predictor_4x4_c)
HIGHBD_INTRA_PRED_TEST(SSE2, TestHighbdIntraPred8,
vpx_highbd_dc_predictor_8x8_sse2, NULL, NULL, NULL,
vpx_highbd_dc_predictor_8x8_sse2, NULL,
vpx_highbd_dc_top_predictor_8x8_sse2, NULL,
vpx_highbd_v_predictor_8x8_sse2,
vpx_highbd_h_predictor_8x8_sse2, NULL, NULL, NULL, NULL,
NULL, NULL, vpx_highbd_tm_predictor_8x8_sse2)
HIGHBD_INTRA_PRED_TEST(SSE2, TestHighbdIntraPred16,
vpx_highbd_dc_predictor_16x16_sse2, NULL, NULL, NULL,
vpx_highbd_dc_predictor_16x16_sse2, NULL,
vpx_highbd_dc_top_predictor_16x16_sse2, NULL,
vpx_highbd_v_predictor_16x16_sse2,
vpx_highbd_h_predictor_16x16_sse2, NULL, NULL, NULL,
NULL, NULL, NULL, vpx_highbd_tm_predictor_16x16_sse2)
HIGHBD_INTRA_PRED_TEST(SSE2, TestHighbdIntraPred32,
vpx_highbd_dc_predictor_32x32_sse2, NULL, NULL, NULL,
vpx_highbd_dc_predictor_32x32_sse2, NULL,
vpx_highbd_dc_top_predictor_32x32_sse2, NULL,
vpx_highbd_v_predictor_32x32_sse2,
vpx_highbd_h_predictor_32x32_sse2, NULL, NULL, NULL,
NULL, NULL, NULL, vpx_highbd_tm_predictor_32x32_sse2)

View File

@@ -479,6 +479,14 @@ INSTANTIATE_TEST_CASE_P(
&vpx_highbd_dc_predictor_16x16_c, 16, 8),
HighbdIntraPredParam(&vpx_highbd_dc_predictor_32x32_sse2,
&vpx_highbd_dc_predictor_32x32_c, 32, 8),
HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_4x4_sse2,
&vpx_highbd_dc_top_predictor_4x4_c, 4, 8),
HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_8x8_sse2,
&vpx_highbd_dc_top_predictor_8x8_c, 8, 8),
HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_16x16_sse2,
&vpx_highbd_dc_top_predictor_16x16_c, 16, 8),
HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_32x32_sse2,
&vpx_highbd_dc_top_predictor_32x32_c, 32, 8),
HighbdIntraPredParam(&vpx_highbd_tm_predictor_4x4_sse2,
&vpx_highbd_tm_predictor_4x4_c, 4, 8),
HighbdIntraPredParam(&vpx_highbd_tm_predictor_8x8_sse2,
@@ -515,6 +523,14 @@ INSTANTIATE_TEST_CASE_P(
&vpx_highbd_dc_predictor_16x16_c, 16, 10),
HighbdIntraPredParam(&vpx_highbd_dc_predictor_32x32_sse2,
&vpx_highbd_dc_predictor_32x32_c, 32, 10),
HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_4x4_sse2,
&vpx_highbd_dc_top_predictor_4x4_c, 4, 10),
HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_8x8_sse2,
&vpx_highbd_dc_top_predictor_8x8_c, 8, 10),
HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_16x16_sse2,
&vpx_highbd_dc_top_predictor_16x16_c, 16, 10),
HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_32x32_sse2,
&vpx_highbd_dc_top_predictor_32x32_c, 32, 10),
HighbdIntraPredParam(&vpx_highbd_tm_predictor_4x4_sse2,
&vpx_highbd_tm_predictor_4x4_c, 4, 10),
HighbdIntraPredParam(&vpx_highbd_tm_predictor_8x8_sse2,
@@ -551,6 +567,14 @@ INSTANTIATE_TEST_CASE_P(
&vpx_highbd_dc_predictor_16x16_c, 16, 12),
HighbdIntraPredParam(&vpx_highbd_dc_predictor_32x32_sse2,
&vpx_highbd_dc_predictor_32x32_c, 32, 12),
HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_4x4_sse2,
&vpx_highbd_dc_top_predictor_4x4_c, 4, 12),
HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_8x8_sse2,
&vpx_highbd_dc_top_predictor_8x8_c, 8, 12),
HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_16x16_sse2,
&vpx_highbd_dc_top_predictor_16x16_c, 16, 12),
HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_32x32_sse2,
&vpx_highbd_dc_top_predictor_32x32_c, 32, 12),
HighbdIntraPredParam(&vpx_highbd_tm_predictor_4x4_sse2,
&vpx_highbd_tm_predictor_4x4_c, 4, 12),
HighbdIntraPredParam(&vpx_highbd_tm_predictor_8x8_sse2,