Merge "Fix sub8x8 motion search on scaled reference frame"
diff --git a/test/vp9_avg_test.cc b/test/avg_test.cc
similarity index 78%
rename from test/vp9_avg_test.cc
rename to test/avg_test.cc
index cbc667e..7d5380f 100644
--- a/test/vp9_avg_test.cc
+++ b/test/avg_test.cc
@@ -15,9 +15,7 @@
#include "third_party/googletest/src/include/gtest/gtest.h"
#include "./vpx_config.h"
-#if CONFIG_VP9_ENCODER
-#include "./vp9_rtcd.h"
-#endif
+#include "./vpx_dsp_rtcd.h"
#include "test/acm_random.h"
#include "test/clear_system_state.h"
@@ -323,91 +321,91 @@
INSTANTIATE_TEST_CASE_P(
C, AverageTest,
::testing::Values(
- make_tuple(16, 16, 1, 8, &vp9_avg_8x8_c),
- make_tuple(16, 16, 1, 4, &vp9_avg_4x4_c)));
+ make_tuple(16, 16, 1, 8, &vpx_avg_8x8_c),
+ make_tuple(16, 16, 1, 4, &vpx_avg_4x4_c)));
INSTANTIATE_TEST_CASE_P(
C, SatdTest,
::testing::Values(
- make_tuple(16, &vp9_satd_c),
- make_tuple(64, &vp9_satd_c),
- make_tuple(256, &vp9_satd_c),
- make_tuple(1024, &vp9_satd_c)));
+ make_tuple(16, &vpx_satd_c),
+ make_tuple(64, &vpx_satd_c),
+ make_tuple(256, &vpx_satd_c),
+ make_tuple(1024, &vpx_satd_c)));
#if HAVE_SSE2
INSTANTIATE_TEST_CASE_P(
SSE2, AverageTest,
::testing::Values(
- make_tuple(16, 16, 0, 8, &vp9_avg_8x8_sse2),
- make_tuple(16, 16, 5, 8, &vp9_avg_8x8_sse2),
- make_tuple(32, 32, 15, 8, &vp9_avg_8x8_sse2),
- make_tuple(16, 16, 0, 4, &vp9_avg_4x4_sse2),
- make_tuple(16, 16, 5, 4, &vp9_avg_4x4_sse2),
- make_tuple(32, 32, 15, 4, &vp9_avg_4x4_sse2)));
+ make_tuple(16, 16, 0, 8, &vpx_avg_8x8_sse2),
+ make_tuple(16, 16, 5, 8, &vpx_avg_8x8_sse2),
+ make_tuple(32, 32, 15, 8, &vpx_avg_8x8_sse2),
+ make_tuple(16, 16, 0, 4, &vpx_avg_4x4_sse2),
+ make_tuple(16, 16, 5, 4, &vpx_avg_4x4_sse2),
+ make_tuple(32, 32, 15, 4, &vpx_avg_4x4_sse2)));
INSTANTIATE_TEST_CASE_P(
SSE2, IntProRowTest, ::testing::Values(
- make_tuple(16, &vp9_int_pro_row_sse2, &vp9_int_pro_row_c),
- make_tuple(32, &vp9_int_pro_row_sse2, &vp9_int_pro_row_c),
- make_tuple(64, &vp9_int_pro_row_sse2, &vp9_int_pro_row_c)));
+ make_tuple(16, &vpx_int_pro_row_sse2, &vpx_int_pro_row_c),
+ make_tuple(32, &vpx_int_pro_row_sse2, &vpx_int_pro_row_c),
+ make_tuple(64, &vpx_int_pro_row_sse2, &vpx_int_pro_row_c)));
INSTANTIATE_TEST_CASE_P(
SSE2, IntProColTest, ::testing::Values(
- make_tuple(16, &vp9_int_pro_col_sse2, &vp9_int_pro_col_c),
- make_tuple(32, &vp9_int_pro_col_sse2, &vp9_int_pro_col_c),
- make_tuple(64, &vp9_int_pro_col_sse2, &vp9_int_pro_col_c)));
+ make_tuple(16, &vpx_int_pro_col_sse2, &vpx_int_pro_col_c),
+ make_tuple(32, &vpx_int_pro_col_sse2, &vpx_int_pro_col_c),
+ make_tuple(64, &vpx_int_pro_col_sse2, &vpx_int_pro_col_c)));
INSTANTIATE_TEST_CASE_P(
SSE2, SatdTest,
::testing::Values(
- make_tuple(16, &vp9_satd_sse2),
- make_tuple(64, &vp9_satd_sse2),
- make_tuple(256, &vp9_satd_sse2),
- make_tuple(1024, &vp9_satd_sse2)));
+ make_tuple(16, &vpx_satd_sse2),
+ make_tuple(64, &vpx_satd_sse2),
+ make_tuple(256, &vpx_satd_sse2),
+ make_tuple(1024, &vpx_satd_sse2)));
#endif
#if HAVE_NEON
INSTANTIATE_TEST_CASE_P(
NEON, AverageTest,
::testing::Values(
- make_tuple(16, 16, 0, 8, &vp9_avg_8x8_neon),
- make_tuple(16, 16, 5, 8, &vp9_avg_8x8_neon),
- make_tuple(32, 32, 15, 8, &vp9_avg_8x8_neon),
- make_tuple(16, 16, 0, 4, &vp9_avg_4x4_neon),
- make_tuple(16, 16, 5, 4, &vp9_avg_4x4_neon),
- make_tuple(32, 32, 15, 4, &vp9_avg_4x4_neon)));
+ make_tuple(16, 16, 0, 8, &vpx_avg_8x8_neon),
+ make_tuple(16, 16, 5, 8, &vpx_avg_8x8_neon),
+ make_tuple(32, 32, 15, 8, &vpx_avg_8x8_neon),
+ make_tuple(16, 16, 0, 4, &vpx_avg_4x4_neon),
+ make_tuple(16, 16, 5, 4, &vpx_avg_4x4_neon),
+ make_tuple(32, 32, 15, 4, &vpx_avg_4x4_neon)));
INSTANTIATE_TEST_CASE_P(
NEON, IntProRowTest, ::testing::Values(
- make_tuple(16, &vp9_int_pro_row_neon, &vp9_int_pro_row_c),
- make_tuple(32, &vp9_int_pro_row_neon, &vp9_int_pro_row_c),
- make_tuple(64, &vp9_int_pro_row_neon, &vp9_int_pro_row_c)));
+ make_tuple(16, &vpx_int_pro_row_neon, &vpx_int_pro_row_c),
+ make_tuple(32, &vpx_int_pro_row_neon, &vpx_int_pro_row_c),
+ make_tuple(64, &vpx_int_pro_row_neon, &vpx_int_pro_row_c)));
INSTANTIATE_TEST_CASE_P(
NEON, IntProColTest, ::testing::Values(
- make_tuple(16, &vp9_int_pro_col_neon, &vp9_int_pro_col_c),
- make_tuple(32, &vp9_int_pro_col_neon, &vp9_int_pro_col_c),
- make_tuple(64, &vp9_int_pro_col_neon, &vp9_int_pro_col_c)));
+ make_tuple(16, &vpx_int_pro_col_neon, &vpx_int_pro_col_c),
+ make_tuple(32, &vpx_int_pro_col_neon, &vpx_int_pro_col_c),
+ make_tuple(64, &vpx_int_pro_col_neon, &vpx_int_pro_col_c)));
INSTANTIATE_TEST_CASE_P(
NEON, SatdTest,
::testing::Values(
- make_tuple(16, &vp9_satd_neon),
- make_tuple(64, &vp9_satd_neon),
- make_tuple(256, &vp9_satd_neon),
- make_tuple(1024, &vp9_satd_neon)));
+ make_tuple(16, &vpx_satd_neon),
+ make_tuple(64, &vpx_satd_neon),
+ make_tuple(256, &vpx_satd_neon),
+ make_tuple(1024, &vpx_satd_neon)));
#endif
#if HAVE_MSA
INSTANTIATE_TEST_CASE_P(
MSA, AverageTest,
::testing::Values(
- make_tuple(16, 16, 0, 8, &vp9_avg_8x8_msa),
- make_tuple(16, 16, 5, 8, &vp9_avg_8x8_msa),
- make_tuple(32, 32, 15, 8, &vp9_avg_8x8_msa),
- make_tuple(16, 16, 0, 4, &vp9_avg_4x4_msa),
- make_tuple(16, 16, 5, 4, &vp9_avg_4x4_msa),
- make_tuple(32, 32, 15, 4, &vp9_avg_4x4_msa)));
+ make_tuple(16, 16, 0, 8, &vpx_avg_8x8_msa),
+ make_tuple(16, 16, 5, 8, &vpx_avg_8x8_msa),
+ make_tuple(32, 32, 15, 8, &vpx_avg_8x8_msa),
+ make_tuple(16, 16, 0, 4, &vpx_avg_4x4_msa),
+ make_tuple(16, 16, 5, 4, &vpx_avg_4x4_msa),
+ make_tuple(32, 32, 15, 4, &vpx_avg_4x4_msa)));
#endif
} // namespace
diff --git a/test/sad_test.cc b/test/sad_test.cc
index 9dec3cb..a144cfc 100644
--- a/test/sad_test.cc
+++ b/test/sad_test.cc
@@ -702,18 +702,6 @@
#if HAVE_SSE
#if CONFIG_USE_X86INC
-const SadMxNParam sse_tests[] = {
- make_tuple(4, 8, &vpx_sad4x8_sse, -1),
- make_tuple(4, 4, &vpx_sad4x4_sse, -1),
-};
-INSTANTIATE_TEST_CASE_P(SSE, SADTest, ::testing::ValuesIn(sse_tests));
-
-const SadMxNAvgParam avg_sse_tests[] = {
- make_tuple(4, 8, &vpx_sad4x8_avg_sse, -1),
- make_tuple(4, 4, &vpx_sad4x4_avg_sse, -1),
-};
-INSTANTIATE_TEST_CASE_P(SSE, SADavgTest, ::testing::ValuesIn(avg_sse_tests));
-
const SadMxNx4Param x4d_sse_tests[] = {
make_tuple(4, 8, &vpx_sad4x8x4d_sse, -1),
make_tuple(4, 4, &vpx_sad4x4x4d_sse, -1),
@@ -736,6 +724,8 @@
make_tuple(8, 16, &vpx_sad8x16_sse2, -1),
make_tuple(8, 8, &vpx_sad8x8_sse2, -1),
make_tuple(8, 4, &vpx_sad8x4_sse2, -1),
+ make_tuple(4, 8, &vpx_sad4x8_sse2, -1),
+ make_tuple(4, 4, &vpx_sad4x4_sse2, -1),
#if CONFIG_VP9_HIGHBITDEPTH
make_tuple(64, 64, &vpx_highbd_sad64x64_sse2, 8),
make_tuple(64, 32, &vpx_highbd_sad64x32_sse2, 8),
@@ -786,6 +776,8 @@
make_tuple(8, 16, &vpx_sad8x16_avg_sse2, -1),
make_tuple(8, 8, &vpx_sad8x8_avg_sse2, -1),
make_tuple(8, 4, &vpx_sad8x4_avg_sse2, -1),
+ make_tuple(4, 8, &vpx_sad4x8_avg_sse2, -1),
+ make_tuple(4, 4, &vpx_sad4x4_avg_sse2, -1),
#if CONFIG_VP9_HIGHBITDEPTH
make_tuple(64, 64, &vpx_highbd_sad64x64_avg_sse2, 8),
make_tuple(64, 32, &vpx_highbd_sad64x32_avg_sse2, 8),
diff --git a/test/test.mk b/test/test.mk
index 2487bd2..80b57e5 100644
--- a/test/test.mk
+++ b/test/test.mk
@@ -143,7 +143,6 @@
LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += fdct4x4_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += fdct8x8_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += variance_test.cc
-LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_avg_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_error_block_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_quantize_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_subtract_test.cc
@@ -170,6 +169,11 @@
endif # VP10
## Multi-codec / unconditional whitebox tests.
+
+ifeq ($(findstring yes,$(CONFIG_VP9_ENCODER)$(CONFIG_VP10_ENCODER)),yes)
+LIBVPX_TEST_SRCS-yes += avg_test.cc
+endif
+
LIBVPX_TEST_SRCS-$(CONFIG_ENCODERS) += sad_test.cc
TEST_INTRA_PRED_SPEED_SRCS-yes := test_intra_pred_speed.cc
diff --git a/test/test_intra_pred_speed.cc b/test/test_intra_pred_speed.cc
index c270072..3e65fec 100644
--- a/test/test_intra_pred_speed.cc
+++ b/test/test_intra_pred_speed.cc
@@ -337,28 +337,19 @@
vpx_d63_predictor_32x32_c, vpx_tm_predictor_32x32_c)
#if HAVE_SSE2 && CONFIG_USE_X86INC
-#if ARCH_X86_64
INTRA_PRED_TEST(SSE2, TestIntraPred32, vpx_dc_predictor_32x32_sse2,
vpx_dc_left_predictor_32x32_sse2,
vpx_dc_top_predictor_32x32_sse2,
vpx_dc_128_predictor_32x32_sse2, vpx_v_predictor_32x32_sse2,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL,
- vpx_tm_predictor_32x32_sse2)
-#else
-INTRA_PRED_TEST(SSE2, TestIntraPred32, vpx_dc_predictor_32x32_sse2,
- vpx_dc_left_predictor_32x32_sse2,
- vpx_dc_top_predictor_32x32_sse2,
- vpx_dc_128_predictor_32x32_sse2, vpx_v_predictor_32x32_sse2,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL)
-#endif // ARCH_X86_64
+ vpx_h_predictor_32x32_sse2, NULL, NULL, NULL, NULL, NULL,
+ NULL, vpx_tm_predictor_32x32_sse2)
#endif // HAVE_SSE2 && CONFIG_USE_X86INC
#if HAVE_SSSE3 && CONFIG_USE_X86INC
INTRA_PRED_TEST(SSSE3, TestIntraPred32, NULL, NULL, NULL, NULL, NULL,
- vpx_h_predictor_32x32_ssse3, vpx_d45_predictor_32x32_ssse3,
- NULL, NULL, vpx_d153_predictor_32x32_ssse3,
- vpx_d207_predictor_32x32_ssse3, vpx_d63_predictor_32x32_ssse3,
- NULL)
+ NULL, vpx_d45_predictor_32x32_ssse3, NULL, NULL,
+ vpx_d153_predictor_32x32_ssse3, vpx_d207_predictor_32x32_ssse3,
+ vpx_d63_predictor_32x32_ssse3, NULL)
#endif // HAVE_SSSE3 && CONFIG_USE_X86INC
#if HAVE_NEON
diff --git a/test/vp9_intrapred_test.cc b/test/vp9_intrapred_test.cc
index ad3327e..2bebdcb 100644
--- a/test/vp9_intrapred_test.cc
+++ b/test/vp9_intrapred_test.cc
@@ -141,13 +141,13 @@
&vpx_highbd_tm_predictor_16x16_c, 16, 8),
make_tuple(&vpx_highbd_tm_predictor_32x32_sse2,
&vpx_highbd_tm_predictor_32x32_c, 32, 8),
- make_tuple(&vpx_highbd_dc_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_dc_predictor_4x4_sse2,
&vpx_highbd_dc_predictor_4x4_c, 4, 8),
make_tuple(&vpx_highbd_dc_predictor_8x8_sse2,
&vpx_highbd_dc_predictor_8x8_c, 8, 8),
make_tuple(&vpx_highbd_dc_predictor_16x16_sse2,
&vpx_highbd_dc_predictor_16x16_c, 16, 8),
- make_tuple(&vpx_highbd_v_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_v_predictor_4x4_sse2,
&vpx_highbd_v_predictor_4x4_c, 4, 8),
make_tuple(&vpx_highbd_v_predictor_8x8_sse2,
&vpx_highbd_v_predictor_8x8_c, 8, 8),
@@ -155,20 +155,20 @@
&vpx_highbd_v_predictor_16x16_c, 16, 8),
make_tuple(&vpx_highbd_v_predictor_32x32_sse2,
&vpx_highbd_v_predictor_32x32_c, 32, 8),
- make_tuple(&vpx_highbd_tm_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_tm_predictor_4x4_sse2,
&vpx_highbd_tm_predictor_4x4_c, 4, 8),
make_tuple(&vpx_highbd_tm_predictor_8x8_sse2,
&vpx_highbd_tm_predictor_8x8_c, 8, 8)));
#else
INSTANTIATE_TEST_CASE_P(SSE2_TO_C_8, VP9IntraPredTest,
::testing::Values(
- make_tuple(&vpx_highbd_dc_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_dc_predictor_4x4_sse2,
&vpx_highbd_dc_predictor_4x4_c, 4, 8),
make_tuple(&vpx_highbd_dc_predictor_8x8_sse2,
&vpx_highbd_dc_predictor_8x8_c, 8, 8),
make_tuple(&vpx_highbd_dc_predictor_16x16_sse2,
&vpx_highbd_dc_predictor_16x16_c, 16, 8),
- make_tuple(&vpx_highbd_v_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_v_predictor_4x4_sse2,
&vpx_highbd_v_predictor_4x4_c, 4, 8),
make_tuple(&vpx_highbd_v_predictor_8x8_sse2,
&vpx_highbd_v_predictor_8x8_c, 8, 8),
@@ -176,7 +176,7 @@
&vpx_highbd_v_predictor_16x16_c, 16, 8),
make_tuple(&vpx_highbd_v_predictor_32x32_sse2,
&vpx_highbd_v_predictor_32x32_c, 32, 8),
- make_tuple(&vpx_highbd_tm_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_tm_predictor_4x4_sse2,
&vpx_highbd_tm_predictor_4x4_c, 4, 8),
make_tuple(&vpx_highbd_tm_predictor_8x8_sse2,
&vpx_highbd_tm_predictor_8x8_c, 8, 8)));
@@ -194,14 +194,14 @@
make_tuple(&vpx_highbd_tm_predictor_32x32_sse2,
&vpx_highbd_tm_predictor_32x32_c, 32,
10),
- make_tuple(&vpx_highbd_dc_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_dc_predictor_4x4_sse2,
&vpx_highbd_dc_predictor_4x4_c, 4, 10),
make_tuple(&vpx_highbd_dc_predictor_8x8_sse2,
&vpx_highbd_dc_predictor_8x8_c, 8, 10),
make_tuple(&vpx_highbd_dc_predictor_16x16_sse2,
&vpx_highbd_dc_predictor_16x16_c, 16,
10),
- make_tuple(&vpx_highbd_v_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_v_predictor_4x4_sse2,
&vpx_highbd_v_predictor_4x4_c, 4, 10),
make_tuple(&vpx_highbd_v_predictor_8x8_sse2,
&vpx_highbd_v_predictor_8x8_c, 8, 10),
@@ -211,21 +211,21 @@
make_tuple(&vpx_highbd_v_predictor_32x32_sse2,
&vpx_highbd_v_predictor_32x32_c, 32,
10),
- make_tuple(&vpx_highbd_tm_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_tm_predictor_4x4_sse2,
&vpx_highbd_tm_predictor_4x4_c, 4, 10),
make_tuple(&vpx_highbd_tm_predictor_8x8_sse2,
&vpx_highbd_tm_predictor_8x8_c, 8, 10)));
#else
INSTANTIATE_TEST_CASE_P(SSE2_TO_C_10, VP9IntraPredTest,
::testing::Values(
- make_tuple(&vpx_highbd_dc_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_dc_predictor_4x4_sse2,
&vpx_highbd_dc_predictor_4x4_c, 4, 10),
make_tuple(&vpx_highbd_dc_predictor_8x8_sse2,
&vpx_highbd_dc_predictor_8x8_c, 8, 10),
make_tuple(&vpx_highbd_dc_predictor_16x16_sse2,
&vpx_highbd_dc_predictor_16x16_c, 16,
10),
- make_tuple(&vpx_highbd_v_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_v_predictor_4x4_sse2,
&vpx_highbd_v_predictor_4x4_c, 4, 10),
make_tuple(&vpx_highbd_v_predictor_8x8_sse2,
&vpx_highbd_v_predictor_8x8_c, 8, 10),
@@ -233,7 +233,7 @@
&vpx_highbd_v_predictor_16x16_c, 16, 10),
make_tuple(&vpx_highbd_v_predictor_32x32_sse2,
&vpx_highbd_v_predictor_32x32_c, 32, 10),
- make_tuple(&vpx_highbd_tm_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_tm_predictor_4x4_sse2,
&vpx_highbd_tm_predictor_4x4_c, 4, 10),
make_tuple(&vpx_highbd_tm_predictor_8x8_sse2,
&vpx_highbd_tm_predictor_8x8_c, 8, 10)));
@@ -251,14 +251,14 @@
make_tuple(&vpx_highbd_tm_predictor_32x32_sse2,
&vpx_highbd_tm_predictor_32x32_c, 32,
12),
- make_tuple(&vpx_highbd_dc_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_dc_predictor_4x4_sse2,
&vpx_highbd_dc_predictor_4x4_c, 4, 12),
make_tuple(&vpx_highbd_dc_predictor_8x8_sse2,
&vpx_highbd_dc_predictor_8x8_c, 8, 12),
make_tuple(&vpx_highbd_dc_predictor_16x16_sse2,
&vpx_highbd_dc_predictor_16x16_c, 16,
12),
- make_tuple(&vpx_highbd_v_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_v_predictor_4x4_sse2,
&vpx_highbd_v_predictor_4x4_c, 4, 12),
make_tuple(&vpx_highbd_v_predictor_8x8_sse2,
&vpx_highbd_v_predictor_8x8_c, 8, 12),
@@ -268,21 +268,21 @@
make_tuple(&vpx_highbd_v_predictor_32x32_sse2,
&vpx_highbd_v_predictor_32x32_c, 32,
12),
- make_tuple(&vpx_highbd_tm_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_tm_predictor_4x4_sse2,
&vpx_highbd_tm_predictor_4x4_c, 4, 12),
make_tuple(&vpx_highbd_tm_predictor_8x8_sse2,
&vpx_highbd_tm_predictor_8x8_c, 8, 12)));
#else
INSTANTIATE_TEST_CASE_P(SSE2_TO_C_12, VP9IntraPredTest,
::testing::Values(
- make_tuple(&vpx_highbd_dc_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_dc_predictor_4x4_sse2,
&vpx_highbd_dc_predictor_4x4_c, 4, 12),
make_tuple(&vpx_highbd_dc_predictor_8x8_sse2,
&vpx_highbd_dc_predictor_8x8_c, 8, 12),
make_tuple(&vpx_highbd_dc_predictor_16x16_sse2,
&vpx_highbd_dc_predictor_16x16_c, 16,
12),
- make_tuple(&vpx_highbd_v_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_v_predictor_4x4_sse2,
&vpx_highbd_v_predictor_4x4_c, 4, 12),
make_tuple(&vpx_highbd_v_predictor_8x8_sse2,
&vpx_highbd_v_predictor_8x8_c, 8, 12),
@@ -290,7 +290,7 @@
&vpx_highbd_v_predictor_16x16_c, 16, 12),
make_tuple(&vpx_highbd_v_predictor_32x32_sse2,
&vpx_highbd_v_predictor_32x32_c, 32, 12),
- make_tuple(&vpx_highbd_tm_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_tm_predictor_4x4_sse2,
&vpx_highbd_tm_predictor_4x4_c, 4, 12),
make_tuple(&vpx_highbd_tm_predictor_8x8_sse2,
&vpx_highbd_tm_predictor_8x8_c, 8, 12)));
diff --git a/third_party/x86inc/x86inc.asm b/third_party/x86inc/x86inc.asm
index be59de3..e7d3fa5 100644
--- a/third_party/x86inc/x86inc.asm
+++ b/third_party/x86inc/x86inc.asm
@@ -119,7 +119,7 @@
%if ABI_IS_32BIT
%if CONFIG_PIC=1
%ifidn __OUTPUT_FORMAT__,elf32
- %define GET_GOT_SAVE_ARG 1
+ %define GET_GOT_DEFINED 1
%define WRT_PLT wrt ..plt
%macro GET_GOT 1
extern _GLOBAL_OFFSET_TABLE_
@@ -138,7 +138,7 @@
%define RESTORE_GOT pop %1
%endmacro
%elifidn __OUTPUT_FORMAT__,macho32
- %define GET_GOT_SAVE_ARG 1
+ %define GET_GOT_DEFINED 1
%macro GET_GOT 1
push %1
call %%get_got
@@ -149,6 +149,8 @@
%undef RESTORE_GOT
%define RESTORE_GOT pop %1
%endmacro
+ %else
+ %define GET_GOT_DEFINED 0
%endif
%endif
diff --git a/vp10/common/reconintra.c b/vp10/common/reconintra.c
index b852a65..e9e3949 100644
--- a/vp10/common/reconintra.c
+++ b/vp10/common/reconintra.c
@@ -262,7 +262,7 @@
}
#if CONFIG_MISC_FIXES
-static inline void memset16(uint16_t *dst, int val, int n) {
+static INLINE void memset16(uint16_t *dst, int val, int n) {
while (n--)
*dst++ = val;
}
diff --git a/vp10/common/vp10_rtcd_defs.pl b/vp10/common/vp10_rtcd_defs.pl
index 4404701..9860bae 100644
--- a/vp10/common/vp10_rtcd_defs.pl
+++ b/vp10/common/vp10_rtcd_defs.pl
@@ -351,42 +351,6 @@
#
if (vpx_config("CONFIG_VP10_ENCODER") eq "yes") {
-add_proto qw/unsigned int vp10_avg_8x8/, "const uint8_t *, int p";
-specialize qw/vp10_avg_8x8 sse2 neon msa/;
-
-add_proto qw/unsigned int vp10_avg_4x4/, "const uint8_t *, int p";
-specialize qw/vp10_avg_4x4 sse2 msa/;
-
-add_proto qw/void vp10_minmax_8x8/, "const uint8_t *s, int p, const uint8_t *d, int dp, int *min, int *max";
-specialize qw/vp10_minmax_8x8 sse2/;
-
-add_proto qw/void vp10_hadamard_8x8/, "int16_t const *src_diff, int src_stride, int16_t *coeff";
-specialize qw/vp10_hadamard_8x8 sse2/, "$ssse3_x86_64_x86inc";
-
-add_proto qw/void vp10_hadamard_16x16/, "int16_t const *src_diff, int src_stride, int16_t *coeff";
-specialize qw/vp10_hadamard_16x16 sse2/;
-
-add_proto qw/int16_t vp10_satd/, "const int16_t *coeff, int length";
-specialize qw/vp10_satd sse2/;
-
-add_proto qw/void vp10_int_pro_row/, "int16_t *hbuf, uint8_t const *ref, const int ref_stride, const int height";
-specialize qw/vp10_int_pro_row sse2 neon/;
-
-add_proto qw/int16_t vp10_int_pro_col/, "uint8_t const *ref, const int width";
-specialize qw/vp10_int_pro_col sse2 neon/;
-
-add_proto qw/int vp10_vector_var/, "int16_t const *ref, int16_t const *src, const int bwl";
-specialize qw/vp10_vector_var neon sse2/;
-
-if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
- add_proto qw/unsigned int vp10_highbd_avg_8x8/, "const uint8_t *, int p";
- specialize qw/vp10_highbd_avg_8x8/;
- add_proto qw/unsigned int vp10_highbd_avg_4x4/, "const uint8_t *, int p";
- specialize qw/vp10_highbd_avg_4x4/;
- add_proto qw/void vp10_highbd_minmax_8x8/, "const uint8_t *s, int p, const uint8_t *d, int dp, int *min, int *max";
- specialize qw/vp10_highbd_minmax_8x8/;
-}
-
# ENCODEMB INVOKE
#
diff --git a/vp10/decoder/decodeframe.c b/vp10/decoder/decodeframe.c
index 80f1778..c0fbc49 100644
--- a/vp10/decoder/decodeframe.c
+++ b/vp10/decoder/decodeframe.c
@@ -1143,32 +1143,13 @@
vpx_rb_read_inv_signed_literal(rb, CONFIG_MISC_FIXES ? 6 : 4) : 0;
}
-static void setup_quantization(VP10_COMMON *const cm, MACROBLOCKD *const xd,
+static void setup_quantization(VP10_COMMON *const cm,
struct vpx_read_bit_buffer *rb) {
- int i;
-
cm->base_qindex = vpx_rb_read_literal(rb, QINDEX_BITS);
cm->y_dc_delta_q = read_delta_q(rb);
cm->uv_dc_delta_q = read_delta_q(rb);
cm->uv_ac_delta_q = read_delta_q(rb);
cm->dequant_bit_depth = cm->bit_depth;
- for (i = 0; i < (cm->seg.enabled ? MAX_SEGMENTS : 1); ++i) {
-#if CONFIG_MISC_FIXES
- const int qindex = vp10_get_qindex(&cm->seg, i, cm->base_qindex);
-#endif
- xd->lossless[i] = cm->y_dc_delta_q == 0 &&
-#if CONFIG_MISC_FIXES
- qindex == 0 &&
-#else
- cm->base_qindex == 0 &&
-#endif
- cm->uv_dc_delta_q == 0 &&
- cm->uv_ac_delta_q == 0;
- }
-
-#if CONFIG_VP9_HIGHBITDEPTH
- xd->bd = (int)cm->bit_depth;
-#endif
}
static void setup_segmentation_dequant(VP10_COMMON *const cm) {
@@ -1878,9 +1859,7 @@
static size_t read_uncompressed_header(VP10Decoder *pbi,
struct vpx_read_bit_buffer *rb) {
VP10_COMMON *const cm = &pbi->common;
-#if CONFIG_MISC_FIXES
MACROBLOCKD *const xd = &pbi->mb;
-#endif
BufferPool *const pool = cm->buffer_pool;
RefCntBuffer *const frame_bufs = pool->frame_bufs;
int i, mask, ref_index = 0;
@@ -2108,12 +2087,30 @@
vp10_setup_past_independence(cm);
setup_loopfilter(&cm->lf, rb);
- setup_quantization(cm, &pbi->mb, rb);
+ setup_quantization(cm, rb);
+#if CONFIG_VP9_HIGHBITDEPTH
+ xd->bd = (int)cm->bit_depth;
+#endif
+
setup_segmentation(cm, rb);
+
+ {
+ int i;
+ for (i = 0; i < MAX_SEGMENTS; ++i) {
+ const int qindex = CONFIG_MISC_FIXES && cm->seg.enabled ?
+ vp10_get_qindex(&cm->seg, i, cm->base_qindex) :
+ cm->base_qindex;
+ xd->lossless[i] = qindex == 0 &&
+ cm->y_dc_delta_q == 0 &&
+ cm->uv_dc_delta_q == 0 &&
+ cm->uv_ac_delta_q == 0;
+ }
+ }
+
setup_segmentation_dequant(cm);
#if CONFIG_MISC_FIXES
- cm->tx_mode = (!cm->seg.enabled && xd->lossless[0]) ? ONLY_4X4
- : read_tx_mode(rb);
+ cm->tx_mode = (xd->lossless[0]) ? ONLY_4X4
+ : read_tx_mode(rb);
cm->reference_mode = read_frame_reference_mode(cm, rb);
#endif
diff --git a/vp10/encoder/arm/neon/avg_neon.c b/vp10/encoder/arm/neon/avg_neon.c
deleted file mode 100644
index c4ec5c4..0000000
--- a/vp10/encoder/arm/neon/avg_neon.c
+++ /dev/null
@@ -1,160 +0,0 @@
-/*
- * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include <arm_neon.h>
-#include <assert.h>
-
-#include "./vp10_rtcd.h"
-#include "./vpx_config.h"
-
-#include "vpx/vpx_integer.h"
-
-static INLINE unsigned int horizontal_add_u16x8(const uint16x8_t v_16x8) {
- const uint32x4_t a = vpaddlq_u16(v_16x8);
- const uint64x2_t b = vpaddlq_u32(a);
- const uint32x2_t c = vadd_u32(vreinterpret_u32_u64(vget_low_u64(b)),
- vreinterpret_u32_u64(vget_high_u64(b)));
- return vget_lane_u32(c, 0);
-}
-
-unsigned int vp10_avg_8x8_neon(const uint8_t *s, int p) {
- uint8x8_t v_s0 = vld1_u8(s);
- const uint8x8_t v_s1 = vld1_u8(s + p);
- uint16x8_t v_sum = vaddl_u8(v_s0, v_s1);
-
- v_s0 = vld1_u8(s + 2 * p);
- v_sum = vaddw_u8(v_sum, v_s0);
-
- v_s0 = vld1_u8(s + 3 * p);
- v_sum = vaddw_u8(v_sum, v_s0);
-
- v_s0 = vld1_u8(s + 4 * p);
- v_sum = vaddw_u8(v_sum, v_s0);
-
- v_s0 = vld1_u8(s + 5 * p);
- v_sum = vaddw_u8(v_sum, v_s0);
-
- v_s0 = vld1_u8(s + 6 * p);
- v_sum = vaddw_u8(v_sum, v_s0);
-
- v_s0 = vld1_u8(s + 7 * p);
- v_sum = vaddw_u8(v_sum, v_s0);
-
- return (horizontal_add_u16x8(v_sum) + 32) >> 6;
-}
-
-void vp10_int_pro_row_neon(int16_t hbuf[16], uint8_t const *ref,
- const int ref_stride, const int height) {
- int i;
- uint16x8_t vec_sum_lo = vdupq_n_u16(0);
- uint16x8_t vec_sum_hi = vdupq_n_u16(0);
- const int shift_factor = ((height >> 5) + 3) * -1;
- const int16x8_t vec_shift = vdupq_n_s16(shift_factor);
-
- for (i = 0; i < height; i += 8) {
- const uint8x16_t vec_row1 = vld1q_u8(ref);
- const uint8x16_t vec_row2 = vld1q_u8(ref + ref_stride);
- const uint8x16_t vec_row3 = vld1q_u8(ref + ref_stride * 2);
- const uint8x16_t vec_row4 = vld1q_u8(ref + ref_stride * 3);
- const uint8x16_t vec_row5 = vld1q_u8(ref + ref_stride * 4);
- const uint8x16_t vec_row6 = vld1q_u8(ref + ref_stride * 5);
- const uint8x16_t vec_row7 = vld1q_u8(ref + ref_stride * 6);
- const uint8x16_t vec_row8 = vld1q_u8(ref + ref_stride * 7);
-
- vec_sum_lo = vaddw_u8(vec_sum_lo, vget_low_u8(vec_row1));
- vec_sum_hi = vaddw_u8(vec_sum_hi, vget_high_u8(vec_row1));
-
- vec_sum_lo = vaddw_u8(vec_sum_lo, vget_low_u8(vec_row2));
- vec_sum_hi = vaddw_u8(vec_sum_hi, vget_high_u8(vec_row2));
-
- vec_sum_lo = vaddw_u8(vec_sum_lo, vget_low_u8(vec_row3));
- vec_sum_hi = vaddw_u8(vec_sum_hi, vget_high_u8(vec_row3));
-
- vec_sum_lo = vaddw_u8(vec_sum_lo, vget_low_u8(vec_row4));
- vec_sum_hi = vaddw_u8(vec_sum_hi, vget_high_u8(vec_row4));
-
- vec_sum_lo = vaddw_u8(vec_sum_lo, vget_low_u8(vec_row5));
- vec_sum_hi = vaddw_u8(vec_sum_hi, vget_high_u8(vec_row5));
-
- vec_sum_lo = vaddw_u8(vec_sum_lo, vget_low_u8(vec_row6));
- vec_sum_hi = vaddw_u8(vec_sum_hi, vget_high_u8(vec_row6));
-
- vec_sum_lo = vaddw_u8(vec_sum_lo, vget_low_u8(vec_row7));
- vec_sum_hi = vaddw_u8(vec_sum_hi, vget_high_u8(vec_row7));
-
- vec_sum_lo = vaddw_u8(vec_sum_lo, vget_low_u8(vec_row8));
- vec_sum_hi = vaddw_u8(vec_sum_hi, vget_high_u8(vec_row8));
-
- ref += ref_stride * 8;
- }
-
- vec_sum_lo = vshlq_u16(vec_sum_lo, vec_shift);
- vec_sum_hi = vshlq_u16(vec_sum_hi, vec_shift);
-
- vst1q_s16(hbuf, vreinterpretq_s16_u16(vec_sum_lo));
- hbuf += 8;
- vst1q_s16(hbuf, vreinterpretq_s16_u16(vec_sum_hi));
-}
-
-int16_t vp10_int_pro_col_neon(uint8_t const *ref, const int width) {
- int i;
- uint16x8_t vec_sum = vdupq_n_u16(0);
-
- for (i = 0; i < width; i += 16) {
- const uint8x16_t vec_row = vld1q_u8(ref);
- vec_sum = vaddw_u8(vec_sum, vget_low_u8(vec_row));
- vec_sum = vaddw_u8(vec_sum, vget_high_u8(vec_row));
- ref += 16;
- }
-
- return horizontal_add_u16x8(vec_sum);
-}
-
-// ref, src = [0, 510] - max diff = 16-bits
-// bwl = {2, 3, 4}, width = {16, 32, 64}
-int vp10_vector_var_neon(int16_t const *ref, int16_t const *src, const int bwl) {
- int width = 4 << bwl;
- int32x4_t sse = vdupq_n_s32(0);
- int16x8_t total = vdupq_n_s16(0);
-
- assert(width >= 8);
- assert((width % 8) == 0);
-
- do {
- const int16x8_t r = vld1q_s16(ref);
- const int16x8_t s = vld1q_s16(src);
- const int16x8_t diff = vsubq_s16(r, s); // [-510, 510], 10 bits.
- const int16x4_t diff_lo = vget_low_s16(diff);
- const int16x4_t diff_hi = vget_high_s16(diff);
- sse = vmlal_s16(sse, diff_lo, diff_lo); // dynamic range 26 bits.
- sse = vmlal_s16(sse, diff_hi, diff_hi);
- total = vaddq_s16(total, diff); // dynamic range 16 bits.
-
- ref += 8;
- src += 8;
- width -= 8;
- } while (width != 0);
-
- {
- // Note: 'total''s pairwise addition could be implemented similarly to
- // horizontal_add_u16x8(), but one less vpaddl with 'total' when paired
- // with the summation of 'sse' performed better on a Cortex-A15.
- const int32x4_t t0 = vpaddlq_s16(total); // cascading summation of 'total'
- const int32x2_t t1 = vadd_s32(vget_low_s32(t0), vget_high_s32(t0));
- const int32x2_t t2 = vpadd_s32(t1, t1);
- const int t = vget_lane_s32(t2, 0);
- const int64x2_t s0 = vpaddlq_s32(sse); // cascading summation of 'sse'.
- const int32x2_t s1 = vadd_s32(vreinterpret_s32_s64(vget_low_s64(s0)),
- vreinterpret_s32_s64(vget_high_s64(s0)));
- const int s = vget_lane_s32(s1, 0);
- const int shift_factor = bwl + 2;
- return s - ((t * t) >> shift_factor);
- }
-}
diff --git a/vp10/encoder/avg.c b/vp10/encoder/avg.c
deleted file mode 100644
index 738c427..0000000
--- a/vp10/encoder/avg.c
+++ /dev/null
@@ -1,230 +0,0 @@
-/*
- * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-#include "./vp10_rtcd.h"
-#include "vp10/common/common.h"
-#include "vpx_ports/mem.h"
-
-unsigned int vp10_avg_8x8_c(const uint8_t *s, int p) {
- int i, j;
- int sum = 0;
- for (i = 0; i < 8; ++i, s+=p)
- for (j = 0; j < 8; sum += s[j], ++j) {}
-
- return (sum + 32) >> 6;
-}
-
-unsigned int vp10_avg_4x4_c(const uint8_t *s, int p) {
- int i, j;
- int sum = 0;
- for (i = 0; i < 4; ++i, s+=p)
- for (j = 0; j < 4; sum += s[j], ++j) {}
-
- return (sum + 8) >> 4;
-}
-
-// src_diff: first pass, 9 bit, dynamic range [-255, 255]
-// second pass, 12 bit, dynamic range [-2040, 2040]
-static void hadamard_col8(const int16_t *src_diff, int src_stride,
- int16_t *coeff) {
- int16_t b0 = src_diff[0 * src_stride] + src_diff[1 * src_stride];
- int16_t b1 = src_diff[0 * src_stride] - src_diff[1 * src_stride];
- int16_t b2 = src_diff[2 * src_stride] + src_diff[3 * src_stride];
- int16_t b3 = src_diff[2 * src_stride] - src_diff[3 * src_stride];
- int16_t b4 = src_diff[4 * src_stride] + src_diff[5 * src_stride];
- int16_t b5 = src_diff[4 * src_stride] - src_diff[5 * src_stride];
- int16_t b6 = src_diff[6 * src_stride] + src_diff[7 * src_stride];
- int16_t b7 = src_diff[6 * src_stride] - src_diff[7 * src_stride];
-
- int16_t c0 = b0 + b2;
- int16_t c1 = b1 + b3;
- int16_t c2 = b0 - b2;
- int16_t c3 = b1 - b3;
- int16_t c4 = b4 + b6;
- int16_t c5 = b5 + b7;
- int16_t c6 = b4 - b6;
- int16_t c7 = b5 - b7;
-
- coeff[0] = c0 + c4;
- coeff[7] = c1 + c5;
- coeff[3] = c2 + c6;
- coeff[4] = c3 + c7;
- coeff[2] = c0 - c4;
- coeff[6] = c1 - c5;
- coeff[1] = c2 - c6;
- coeff[5] = c3 - c7;
-}
-
-void vp10_hadamard_8x8_c(int16_t const *src_diff, int src_stride,
- int16_t *coeff) {
- int idx;
- int16_t buffer[64];
- int16_t *tmp_buf = &buffer[0];
- for (idx = 0; idx < 8; ++idx) {
- hadamard_col8(src_diff, src_stride, tmp_buf); // src_diff: 9 bit
- // dynamic range [-255, 255]
- tmp_buf += 8;
- ++src_diff;
- }
-
- tmp_buf = &buffer[0];
- for (idx = 0; idx < 8; ++idx) {
- hadamard_col8(tmp_buf, 8, coeff); // tmp_buf: 12 bit
- // dynamic range [-2040, 2040]
- coeff += 8; // coeff: 15 bit
- // dynamic range [-16320, 16320]
- ++tmp_buf;
- }
-}
-
-// In place 16x16 2D Hadamard transform
-void vp10_hadamard_16x16_c(int16_t const *src_diff, int src_stride,
- int16_t *coeff) {
- int idx;
- for (idx = 0; idx < 4; ++idx) {
- // src_diff: 9 bit, dynamic range [-255, 255]
- int16_t const *src_ptr = src_diff + (idx >> 1) * 8 * src_stride
- + (idx & 0x01) * 8;
- vp10_hadamard_8x8_c(src_ptr, src_stride, coeff + idx * 64);
- }
-
- // coeff: 15 bit, dynamic range [-16320, 16320]
- for (idx = 0; idx < 64; ++idx) {
- int16_t a0 = coeff[0];
- int16_t a1 = coeff[64];
- int16_t a2 = coeff[128];
- int16_t a3 = coeff[192];
-
- int16_t b0 = (a0 + a1) >> 1; // (a0 + a1): 16 bit, [-32640, 32640]
- int16_t b1 = (a0 - a1) >> 1; // b0-b3: 15 bit, dynamic range
- int16_t b2 = (a2 + a3) >> 1; // [-16320, 16320]
- int16_t b3 = (a2 - a3) >> 1;
-
- coeff[0] = b0 + b2; // 16 bit, [-32640, 32640]
- coeff[64] = b1 + b3;
- coeff[128] = b0 - b2;
- coeff[192] = b1 - b3;
-
- ++coeff;
- }
-}
-
-// coeff: 16 bits, dynamic range [-32640, 32640].
-// length: value range {16, 64, 256, 1024}.
-int16_t vp10_satd_c(const int16_t *coeff, int length) {
- int i;
- int satd = 0;
- for (i = 0; i < length; ++i)
- satd += abs(coeff[i]);
-
- // satd: 26 bits, dynamic range [-32640 * 1024, 32640 * 1024]
- return (int16_t)satd;
-}
-
-// Integer projection onto row vectors.
-// height: value range {16, 32, 64}.
-void vp10_int_pro_row_c(int16_t hbuf[16], uint8_t const *ref,
- const int ref_stride, const int height) {
- int idx;
- const int norm_factor = height >> 1;
- for (idx = 0; idx < 16; ++idx) {
- int i;
- hbuf[idx] = 0;
- // hbuf[idx]: 14 bit, dynamic range [0, 16320].
- for (i = 0; i < height; ++i)
- hbuf[idx] += ref[i * ref_stride];
- // hbuf[idx]: 9 bit, dynamic range [0, 510].
- hbuf[idx] /= norm_factor;
- ++ref;
- }
-}
-
-// width: value range {16, 32, 64}.
-int16_t vp10_int_pro_col_c(uint8_t const *ref, const int width) {
- int idx;
- int16_t sum = 0;
- // sum: 14 bit, dynamic range [0, 16320]
- for (idx = 0; idx < width; ++idx)
- sum += ref[idx];
- return sum;
-}
-
-// ref: [0 - 510]
-// src: [0 - 510]
-// bwl: {2, 3, 4}
-int vp10_vector_var_c(int16_t const *ref, int16_t const *src,
- const int bwl) {
- int i;
- int width = 4 << bwl;
- int sse = 0, mean = 0, var;
-
- for (i = 0; i < width; ++i) {
- int diff = ref[i] - src[i]; // diff: dynamic range [-510, 510], 10 bits.
- mean += diff; // mean: dynamic range 16 bits.
- sse += diff * diff; // sse: dynamic range 26 bits.
- }
-
- // (mean * mean): dynamic range 31 bits.
- var = sse - ((mean * mean) >> (bwl + 2));
- return var;
-}
-
-void vp10_minmax_8x8_c(const uint8_t *s, int p, const uint8_t *d, int dp,
- int *min, int *max) {
- int i, j;
- *min = 255;
- *max = 0;
- for (i = 0; i < 8; ++i, s += p, d += dp) {
- for (j = 0; j < 8; ++j) {
- int diff = abs(s[j]-d[j]);
- *min = diff < *min ? diff : *min;
- *max = diff > *max ? diff : *max;
- }
- }
-}
-
-#if CONFIG_VP9_HIGHBITDEPTH
-unsigned int vp10_highbd_avg_8x8_c(const uint8_t *s8, int p) {
- int i, j;
- int sum = 0;
- const uint16_t* s = CONVERT_TO_SHORTPTR(s8);
- for (i = 0; i < 8; ++i, s+=p)
- for (j = 0; j < 8; sum += s[j], ++j) {}
-
- return (sum + 32) >> 6;
-}
-
-unsigned int vp10_highbd_avg_4x4_c(const uint8_t *s8, int p) {
- int i, j;
- int sum = 0;
- const uint16_t* s = CONVERT_TO_SHORTPTR(s8);
- for (i = 0; i < 4; ++i, s+=p)
- for (j = 0; j < 4; sum += s[j], ++j) {}
-
- return (sum + 8) >> 4;
-}
-
-void vp10_highbd_minmax_8x8_c(const uint8_t *s8, int p, const uint8_t *d8,
- int dp, int *min, int *max) {
- int i, j;
- const uint16_t* s = CONVERT_TO_SHORTPTR(s8);
- const uint16_t* d = CONVERT_TO_SHORTPTR(d8);
- *min = 255;
- *max = 0;
- for (i = 0; i < 8; ++i, s += p, d += dp) {
- for (j = 0; j < 8; ++j) {
- int diff = abs(s[j]-d[j]);
- *min = diff < *min ? diff : *min;
- *max = diff > *max ? diff : *max;
- }
- }
-}
-#endif // CONFIG_VP9_HIGHBITDEPTH
-
-
diff --git a/vp10/encoder/bitstream.c b/vp10/encoder/bitstream.c
index 32275d4..361ac99 100644
--- a/vp10/encoder/bitstream.c
+++ b/vp10/encoder/bitstream.c
@@ -897,18 +897,17 @@
if (mode != TX_MODE_SELECT)
vpx_wb_write_literal(wb, mode, 2);
}
+#else
+static void write_txfm_mode(TX_MODE mode, struct vpx_writer *wb) {
+ vpx_write_literal(wb, VPXMIN(mode, ALLOW_32X32), 2);
+ if (mode >= ALLOW_32X32)
+ vpx_write_bit(wb, mode == TX_MODE_SELECT);
+}
#endif
+
static void update_txfm_probs(VP10_COMMON *cm, vpx_writer *w,
FRAME_COUNTS *counts) {
-#if !CONFIG_MISC_FIXES
- // Mode
- vpx_write_literal(w, VPXMIN(cm->tx_mode, ALLOW_32X32), 2);
- if (cm->tx_mode >= ALLOW_32X32)
- vpx_write_bit(w, cm->tx_mode == TX_MODE_SELECT);
-
- // Probabilities
-#endif
if (cm->tx_mode == TX_MODE_SELECT) {
int i, j;
@@ -1261,7 +1260,7 @@
encode_quantization(cm, wb);
encode_segmentation(cm, xd, wb);
#if CONFIG_MISC_FIXES
- if (!cm->seg.enabled && xd->lossless[0])
+ if (xd->lossless[0])
cm->tx_mode = TX_4X4;
else
write_txfm_mode(cm->tx_mode, wb);
@@ -1291,10 +1290,12 @@
vpx_start_encode(&header_bc, data);
#if !CONFIG_MISC_FIXES
- if (cpi->td.mb.e_mbd.lossless[0])
+ if (cpi->td.mb.e_mbd.lossless[0]) {
cm->tx_mode = TX_4X4;
- else
+ } else {
+ write_txfm_mode(cm->tx_mode, &header_bc);
update_txfm_probs(cm, &header_bc, counts);
+ }
#else
update_txfm_probs(cm, &header_bc, counts);
#endif
@@ -1473,7 +1474,7 @@
assert(n_log2_tiles > 0);
vpx_wb_write_literal(&saved_wb, mag, 2);
if (mag < 3)
- data_sz = remux_tiles(data, data_sz, 1 << n_log2_tiles, mag);
+ data_sz = remux_tiles(data, (int)data_sz, 1 << n_log2_tiles, mag);
} else {
assert(n_log2_tiles == 0);
}
diff --git a/vp10/encoder/encodeframe.c b/vp10/encoder/encodeframe.c
index bcdcff3..9381b65 100644
--- a/vp10/encoder/encodeframe.c
+++ b/vp10/encoder/encodeframe.c
@@ -536,16 +536,16 @@
if (x8_idx < pixels_wide && y8_idx < pixels_high) {
#if CONFIG_VP9_HIGHBITDEPTH
if (highbd_flag & YV12_FLAG_HIGHBITDEPTH) {
- vp10_highbd_minmax_8x8(s + y8_idx * sp + x8_idx, sp,
+ vpx_highbd_minmax_8x8(s + y8_idx * sp + x8_idx, sp,
d + y8_idx * dp + x8_idx, dp,
&min, &max);
} else {
- vp10_minmax_8x8(s + y8_idx * sp + x8_idx, sp,
+ vpx_minmax_8x8(s + y8_idx * sp + x8_idx, sp,
d + y8_idx * dp + x8_idx, dp,
&min, &max);
}
#else
- vp10_minmax_8x8(s + y8_idx * sp + x8_idx, sp,
+ vpx_minmax_8x8(s + y8_idx * sp + x8_idx, sp,
d + y8_idx * dp + x8_idx, dp,
&min, &max);
#endif
@@ -577,18 +577,18 @@
int d_avg = 128;
#if CONFIG_VP9_HIGHBITDEPTH
if (highbd_flag & YV12_FLAG_HIGHBITDEPTH) {
- s_avg = vp10_highbd_avg_4x4(s + y4_idx * sp + x4_idx, sp);
+ s_avg = vpx_highbd_avg_4x4(s + y4_idx * sp + x4_idx, sp);
if (!is_key_frame)
- d_avg = vp10_highbd_avg_4x4(d + y4_idx * dp + x4_idx, dp);
+ d_avg = vpx_highbd_avg_4x4(d + y4_idx * dp + x4_idx, dp);
} else {
- s_avg = vp10_avg_4x4(s + y4_idx * sp + x4_idx, sp);
+ s_avg = vpx_avg_4x4(s + y4_idx * sp + x4_idx, sp);
if (!is_key_frame)
- d_avg = vp10_avg_4x4(d + y4_idx * dp + x4_idx, dp);
+ d_avg = vpx_avg_4x4(d + y4_idx * dp + x4_idx, dp);
}
#else
- s_avg = vp10_avg_4x4(s + y4_idx * sp + x4_idx, sp);
+ s_avg = vpx_avg_4x4(s + y4_idx * sp + x4_idx, sp);
if (!is_key_frame)
- d_avg = vp10_avg_4x4(d + y4_idx * dp + x4_idx, dp);
+ d_avg = vpx_avg_4x4(d + y4_idx * dp + x4_idx, dp);
#endif
sum = s_avg - d_avg;
sse = sum * sum;
@@ -616,18 +616,18 @@
int d_avg = 128;
#if CONFIG_VP9_HIGHBITDEPTH
if (highbd_flag & YV12_FLAG_HIGHBITDEPTH) {
- s_avg = vp10_highbd_avg_8x8(s + y8_idx * sp + x8_idx, sp);
+ s_avg = vpx_highbd_avg_8x8(s + y8_idx * sp + x8_idx, sp);
if (!is_key_frame)
- d_avg = vp10_highbd_avg_8x8(d + y8_idx * dp + x8_idx, dp);
+ d_avg = vpx_highbd_avg_8x8(d + y8_idx * dp + x8_idx, dp);
} else {
- s_avg = vp10_avg_8x8(s + y8_idx * sp + x8_idx, sp);
+ s_avg = vpx_avg_8x8(s + y8_idx * sp + x8_idx, sp);
if (!is_key_frame)
- d_avg = vp10_avg_8x8(d + y8_idx * dp + x8_idx, dp);
+ d_avg = vpx_avg_8x8(d + y8_idx * dp + x8_idx, dp);
}
#else
- s_avg = vp10_avg_8x8(s + y8_idx * sp + x8_idx, sp);
+ s_avg = vpx_avg_8x8(s + y8_idx * sp + x8_idx, sp);
if (!is_key_frame)
- d_avg = vp10_avg_8x8(d + y8_idx * dp + x8_idx, dp);
+ d_avg = vpx_avg_8x8(d + y8_idx * dp + x8_idx, dp);
#endif
sum = s_avg - d_avg;
sse = sum * sum;
@@ -1155,7 +1155,7 @@
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
x->source_variance =
vp10_high_get_sby_perpixel_variance(cpi, &x->plane[0].src,
- bsize, xd->bd);
+ bsize, xd->bd);
} else {
x->source_variance =
vp10_get_sby_perpixel_variance(cpi, &x->plane[0].src, bsize);
@@ -2579,7 +2579,7 @@
}
static TX_MODE select_tx_mode(const VP10_COMP *cpi, MACROBLOCKD *const xd) {
- if (!cpi->common.seg.enabled && xd->lossless[0])
+ if (xd->lossless[0])
return ONLY_4X4;
if (cpi->sf.tx_size_search_method == USE_LARGESTALL)
return ALLOW_32X32;
@@ -2702,16 +2702,12 @@
rdc->m_search_count = 0; // Count of motion search hits.
rdc->ex_search_count = 0; // Exhaustive mesh search hits.
- for (i = 0; i < (cm->seg.enabled ? MAX_SEGMENTS : 1); ++i) {
-#if CONFIG_MISC_FIXES
- const int qindex = vp10_get_qindex(&cm->seg, i, cm->base_qindex);
-#endif
- xd->lossless[i] = cm->y_dc_delta_q == 0 &&
-#if CONFIG_MISC_FIXES
- qindex == 0 &&
-#else
- cm->base_qindex == 0 &&
-#endif
+ for (i = 0; i < MAX_SEGMENTS; ++i) {
+ const int qindex = CONFIG_MISC_FIXES && cm->seg.enabled ?
+ vp10_get_qindex(&cm->seg, i, cm->base_qindex) :
+ cm->base_qindex;
+ xd->lossless[i] = qindex == 0 &&
+ cm->y_dc_delta_q == 0 &&
cm->uv_dc_delta_q == 0 &&
cm->uv_ac_delta_q == 0;
}
diff --git a/vp10/encoder/encoder.c b/vp10/encoder/encoder.c
index 6bba848..175c6d8 100644
--- a/vp10/encoder/encoder.c
+++ b/vp10/encoder/encoder.c
@@ -1422,7 +1422,11 @@
cpi->td.mb.e_mbd.bd = (int)cm->bit_depth;
#endif // CONFIG_VP9_HIGHBITDEPTH
- rc->baseline_gf_interval = (MIN_GF_INTERVAL + MAX_GF_INTERVAL) / 2;
+ if ((oxcf->pass == 0) && (oxcf->rc_mode == VPX_Q)) {
+ rc->baseline_gf_interval = FIXED_GF_INTERVAL;
+ } else {
+ rc->baseline_gf_interval = (MIN_GF_INTERVAL + MAX_GF_INTERVAL) / 2;
+ }
cpi->refresh_golden_frame = 0;
cpi->refresh_last_frame = 1;
diff --git a/vp10/encoder/mcomp.c b/vp10/encoder/mcomp.c
index 04e1daf..2c1c591 100644
--- a/vp10/encoder/mcomp.c
+++ b/vp10/encoder/mcomp.c
@@ -1759,7 +1759,7 @@
int center, offset = 0;
int bw = 4 << bwl; // redundant variable, to be changed in the experiments.
for (d = 0; d <= bw; d += 16) {
- this_sad = vp10_vector_var(&ref[d], src, bwl);
+ this_sad = vpx_vector_var(&ref[d], src, bwl);
if (this_sad < best_sad) {
best_sad = this_sad;
offset = d;
@@ -1772,7 +1772,7 @@
// check limit
if (this_pos < 0 || this_pos > bw)
continue;
- this_sad = vp10_vector_var(&ref[this_pos], src, bwl);
+ this_sad = vpx_vector_var(&ref[this_pos], src, bwl);
if (this_sad < best_sad) {
best_sad = this_sad;
center = this_pos;
@@ -1785,7 +1785,7 @@
// check limit
if (this_pos < 0 || this_pos > bw)
continue;
- this_sad = vp10_vector_var(&ref[this_pos], src, bwl);
+ this_sad = vpx_vector_var(&ref[this_pos], src, bwl);
if (this_sad < best_sad) {
best_sad = this_sad;
center = this_pos;
@@ -1798,7 +1798,7 @@
// check limit
if (this_pos < 0 || this_pos > bw)
continue;
- this_sad = vp10_vector_var(&ref[this_pos], src, bwl);
+ this_sad = vpx_vector_var(&ref[this_pos], src, bwl);
if (this_sad < best_sad) {
best_sad = this_sad;
center = this_pos;
@@ -1811,7 +1811,7 @@
// check limit
if (this_pos < 0 || this_pos > bw)
continue;
- this_sad = vp10_vector_var(&ref[this_pos], src, bwl);
+ this_sad = vpx_vector_var(&ref[this_pos], src, bwl);
if (this_sad < best_sad) {
best_sad = this_sad;
center = this_pos;
@@ -1880,25 +1880,25 @@
// Set up prediction 1-D reference set
ref_buf = xd->plane[0].pre[0].buf - (bw >> 1);
for (idx = 0; idx < search_width; idx += 16) {
- vp10_int_pro_row(&hbuf[idx], ref_buf, ref_stride, bh);
+ vpx_int_pro_row(&hbuf[idx], ref_buf, ref_stride, bh);
ref_buf += 16;
}
ref_buf = xd->plane[0].pre[0].buf - (bh >> 1) * ref_stride;
for (idx = 0; idx < search_height; ++idx) {
- vbuf[idx] = vp10_int_pro_col(ref_buf, bw) >> norm_factor;
+ vbuf[idx] = vpx_int_pro_col(ref_buf, bw) >> norm_factor;
ref_buf += ref_stride;
}
// Set up src 1-D reference set
for (idx = 0; idx < bw; idx += 16) {
src_buf = x->plane[0].src.buf + idx;
- vp10_int_pro_row(&src_hbuf[idx], src_buf, src_stride, bh);
+ vpx_int_pro_row(&src_hbuf[idx], src_buf, src_stride, bh);
}
src_buf = x->plane[0].src.buf;
for (idx = 0; idx < bh; ++idx) {
- src_vbuf[idx] = vp10_int_pro_col(src_buf, bw) >> norm_factor;
+ src_vbuf[idx] = vpx_int_pro_col(src_buf, bw) >> norm_factor;
src_buf += src_stride;
}
diff --git a/vp10/encoder/mips/msa/avg_msa.c b/vp10/encoder/mips/msa/avg_msa.c
deleted file mode 100644
index e8cfd53..0000000
--- a/vp10/encoder/mips/msa/avg_msa.c
+++ /dev/null
@@ -1,56 +0,0 @@
-/*
- * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include "./vp10_rtcd.h"
-#include "vpx_dsp/mips/macros_msa.h"
-
-uint32_t vp10_avg_8x8_msa(const uint8_t *src, int32_t src_stride) {
- uint32_t sum_out;
- v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
- v8u16 sum0, sum1, sum2, sum3, sum4, sum5, sum6, sum7;
- v4u32 sum = { 0 };
-
- LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
- HADD_UB4_UH(src0, src1, src2, src3, sum0, sum1, sum2, sum3);
- HADD_UB4_UH(src4, src5, src6, src7, sum4, sum5, sum6, sum7);
- ADD4(sum0, sum1, sum2, sum3, sum4, sum5, sum6, sum7, sum0, sum2, sum4, sum6);
- ADD2(sum0, sum2, sum4, sum6, sum0, sum4);
- sum0 += sum4;
-
- sum = __msa_hadd_u_w(sum0, sum0);
- sum0 = (v8u16)__msa_pckev_h((v8i16)sum, (v8i16)sum);
- sum = __msa_hadd_u_w(sum0, sum0);
- sum = (v4u32)__msa_srari_w((v4i32)sum, 6);
- sum_out = __msa_copy_u_w((v4i32)sum, 0);
-
- return sum_out;
-}
-
-uint32_t vp10_avg_4x4_msa(const uint8_t *src, int32_t src_stride) {
- uint32_t sum_out;
- uint32_t src0, src1, src2, src3;
- v16u8 vec = { 0 };
- v8u16 sum0;
- v4u32 sum1;
- v2u64 sum2;
-
- LW4(src, src_stride, src0, src1, src2, src3);
- INSERT_W4_UB(src0, src1, src2, src3, vec);
-
- sum0 = __msa_hadd_u_h(vec, vec);
- sum1 = __msa_hadd_u_w(sum0, sum0);
- sum0 = (v8u16)__msa_pckev_h((v8i16)sum1, (v8i16)sum1);
- sum1 = __msa_hadd_u_w(sum0, sum0);
- sum2 = __msa_hadd_u_d(sum1, sum1);
- sum1 = (v4u32)__msa_srari_w((v4i32)sum2, 4);
- sum_out = __msa_copy_u_w((v4i32)sum1, 0);
-
- return sum_out;
-}
diff --git a/vp10/encoder/ratectrl.c b/vp10/encoder/ratectrl.c
index 3ff2476..6068775 100644
--- a/vp10/encoder/ratectrl.c
+++ b/vp10/encoder/ratectrl.c
@@ -794,16 +794,18 @@
ASSIGN_MINQ_TABLE(cm->bit_depth, inter_minq);
if (frame_is_intra_only(cm)) {
-
- // Handle the special case for key frames forced when we have reached
- // the maximum key frame interval. Here force the Q to a range
- // based on the ambient Q to reduce the risk of popping.
- if (rc->this_key_frame_forced) {
+ if (oxcf->rc_mode == VPX_Q) {
+ int qindex = cq_level;
+ double q = vp10_convert_qindex_to_q(qindex, cm->bit_depth);
+ int delta_qindex = vp10_compute_qdelta(rc, q, q * 0.25,
+ cm->bit_depth);
+ active_best_quality = VPXMAX(qindex + delta_qindex, rc->best_quality);
+ } else if (rc->this_key_frame_forced) {
int qindex = rc->last_boosted_qindex;
double last_boosted_q = vp10_convert_qindex_to_q(qindex, cm->bit_depth);
int delta_qindex = vp10_compute_qdelta(rc, last_boosted_q,
- last_boosted_q * 0.75,
- cm->bit_depth);
+ last_boosted_q * 0.75,
+ cm->bit_depth);
active_best_quality = VPXMAX(qindex + delta_qindex, rc->best_quality);
} else {
// not first frame of one pass and kf_boost is set
@@ -823,8 +825,8 @@
// on active_best_quality.
q_val = vp10_convert_qindex_to_q(active_best_quality, cm->bit_depth);
active_best_quality += vp10_compute_qdelta(rc, q_val,
- q_val * q_adj_factor,
- cm->bit_depth);
+ q_val * q_adj_factor,
+ cm->bit_depth);
}
} else if (!rc->is_src_frame_alt_ref &&
(cpi->refresh_golden_frame || cpi->refresh_alt_ref_frame)) {
@@ -848,17 +850,28 @@
active_best_quality = active_best_quality * 15 / 16;
} else if (oxcf->rc_mode == VPX_Q) {
- if (!cpi->refresh_alt_ref_frame) {
- active_best_quality = cq_level;
- } else {
- active_best_quality = get_gf_active_quality(rc, q, cm->bit_depth);
- }
+ int qindex = cq_level;
+ double q = vp10_convert_qindex_to_q(qindex, cm->bit_depth);
+ int delta_qindex;
+ if (cpi->refresh_alt_ref_frame)
+ delta_qindex = vp10_compute_qdelta(rc, q, q * 0.40, cm->bit_depth);
+ else
+ delta_qindex = vp10_compute_qdelta(rc, q, q * 0.50, cm->bit_depth);
+ active_best_quality = VPXMAX(qindex + delta_qindex, rc->best_quality);
} else {
active_best_quality = get_gf_active_quality(rc, q, cm->bit_depth);
}
} else {
if (oxcf->rc_mode == VPX_Q) {
- active_best_quality = cq_level;
+ int qindex = cq_level;
+ double q = vp10_convert_qindex_to_q(qindex, cm->bit_depth);
+ double delta_rate[FIXED_GF_INTERVAL] =
+ {0.50, 1.0, 0.85, 1.0, 0.70, 1.0, 0.85, 1.0};
+ int delta_qindex =
+ vp10_compute_qdelta(rc, q,
+ q * delta_rate[cm->current_video_frame %
+ FIXED_GF_INTERVAL], cm->bit_depth);
+ active_best_quality = VPXMAX(qindex + delta_qindex, rc->best_quality);
} else {
// Use the lower of active_worst_quality and recent/average Q.
if (cm->current_video_frame > 1)
@@ -1254,7 +1267,7 @@
rc->avg_frame_qindex[KEY_FRAME] =
ROUND_POWER_OF_TWO(3 * rc->avg_frame_qindex[KEY_FRAME] + qindex, 2);
} else {
- if (rc->is_src_frame_alt_ref ||
+ if (!rc->is_src_frame_alt_ref &&
!(cpi->refresh_golden_frame || cpi->refresh_alt_ref_frame)) {
rc->last_q[INTER_FRAME] = qindex;
rc->avg_frame_qindex[INTER_FRAME] =
@@ -1562,29 +1575,36 @@
RATE_CONTROL *const rc) {
const VP10EncoderConfig *const oxcf = &cpi->oxcf;
- // Set Maximum gf/arf interval
- rc->max_gf_interval = oxcf->max_gf_interval;
- rc->min_gf_interval = oxcf->min_gf_interval;
- if (rc->min_gf_interval == 0)
- rc->min_gf_interval = vp10_rc_get_default_min_gf_interval(
- oxcf->width, oxcf->height, cpi->framerate);
- if (rc->max_gf_interval == 0)
- rc->max_gf_interval = vp10_rc_get_default_max_gf_interval(
- cpi->framerate, rc->min_gf_interval);
+ // Special case code for 1 pass fixed Q mode tests
+ if ((oxcf->pass == 0) && (oxcf->rc_mode == VPX_Q)) {
+ rc->max_gf_interval = FIXED_GF_INTERVAL;
+ rc->min_gf_interval = FIXED_GF_INTERVAL;
+ rc->static_scene_max_gf_interval = FIXED_GF_INTERVAL;
+ } else {
+ // Set Maximum gf/arf interval
+ rc->max_gf_interval = oxcf->max_gf_interval;
+ rc->min_gf_interval = oxcf->min_gf_interval;
+ if (rc->min_gf_interval == 0)
+ rc->min_gf_interval = vp10_rc_get_default_min_gf_interval(
+ oxcf->width, oxcf->height, cpi->framerate);
+ if (rc->max_gf_interval == 0)
+ rc->max_gf_interval = vp10_rc_get_default_max_gf_interval(
+ cpi->framerate, rc->min_gf_interval);
- // Extended interval for genuinely static scenes
- rc->static_scene_max_gf_interval = MAX_LAG_BUFFERS * 2;
+ // Extended interval for genuinely static scenes
+ rc->static_scene_max_gf_interval = MAX_LAG_BUFFERS * 2;
- if (is_altref_enabled(cpi)) {
- if (rc->static_scene_max_gf_interval > oxcf->lag_in_frames - 1)
- rc->static_scene_max_gf_interval = oxcf->lag_in_frames - 1;
+ if (is_altref_enabled(cpi)) {
+ if (rc->static_scene_max_gf_interval > oxcf->lag_in_frames - 1)
+ rc->static_scene_max_gf_interval = oxcf->lag_in_frames - 1;
+ }
+
+ if (rc->max_gf_interval > rc->static_scene_max_gf_interval)
+ rc->max_gf_interval = rc->static_scene_max_gf_interval;
+
+ // Clamp min to max
+ rc->min_gf_interval = VPXMIN(rc->min_gf_interval, rc->max_gf_interval);
}
-
- if (rc->max_gf_interval > rc->static_scene_max_gf_interval)
- rc->max_gf_interval = rc->static_scene_max_gf_interval;
-
- // Clamp min to max
- rc->min_gf_interval = VPXMIN(rc->min_gf_interval, rc->max_gf_interval);
}
void vp10_rc_update_framerate(VP10_COMP *cpi) {
diff --git a/vp10/encoder/ratectrl.h b/vp10/encoder/ratectrl.h
index 8008c16..0b9fd45 100644
--- a/vp10/encoder/ratectrl.h
+++ b/vp10/encoder/ratectrl.h
@@ -26,6 +26,7 @@
#define MIN_GF_INTERVAL 4
#define MAX_GF_INTERVAL 16
+#define FIXED_GF_INTERVAL 8 // Used in some testing modes only
typedef enum {
INTER_NORMAL = 0,
diff --git a/vp10/encoder/rdopt.c b/vp10/encoder/rdopt.c
index bba2171..bbddc1d 100644
--- a/vp10/encoder/rdopt.c
+++ b/vp10/encoder/rdopt.c
@@ -725,7 +725,7 @@
assert(bs == xd->mi[0]->mbmi.sb_type);
- if (CONFIG_MISC_FIXES && xd->lossless[xd->mi[0]->mbmi.segment_id]) {
+ if (CONFIG_MISC_FIXES && xd->lossless[0]) {
choose_smallest_tx_size(cpi, x, rate, distortion, skip, ret_sse,
ref_best_rd, bs);
} else if (cpi->sf.tx_size_search_method == USE_LARGESTALL ||
diff --git a/vp10/encoder/x86/avg_intrin_sse2.c b/vp10/encoder/x86/avg_intrin_sse2.c
deleted file mode 100644
index cf23013..0000000
--- a/vp10/encoder/x86/avg_intrin_sse2.c
+++ /dev/null
@@ -1,424 +0,0 @@
-/*
- * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include <emmintrin.h>
-
-#include "./vp10_rtcd.h"
-#include "vpx_ports/mem.h"
-
-void vp10_minmax_8x8_sse2(const uint8_t *s, int p, const uint8_t *d, int dp,
- int *min, int *max) {
- __m128i u0, s0, d0, diff, maxabsdiff, minabsdiff, negdiff, absdiff0, absdiff;
- u0 = _mm_setzero_si128();
- // Row 0
- s0 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(s)), u0);
- d0 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(d)), u0);
- diff = _mm_subs_epi16(s0, d0);
- negdiff = _mm_subs_epi16(u0, diff);
- absdiff0 = _mm_max_epi16(diff, negdiff);
- // Row 1
- s0 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(s + p)), u0);
- d0 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(d + dp)), u0);
- diff = _mm_subs_epi16(s0, d0);
- negdiff = _mm_subs_epi16(u0, diff);
- absdiff = _mm_max_epi16(diff, negdiff);
- maxabsdiff = _mm_max_epi16(absdiff0, absdiff);
- minabsdiff = _mm_min_epi16(absdiff0, absdiff);
- // Row 2
- s0 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(s + 2 * p)), u0);
- d0 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(d + 2 * dp)), u0);
- diff = _mm_subs_epi16(s0, d0);
- negdiff = _mm_subs_epi16(u0, diff);
- absdiff = _mm_max_epi16(diff, negdiff);
- maxabsdiff = _mm_max_epi16(maxabsdiff, absdiff);
- minabsdiff = _mm_min_epi16(minabsdiff, absdiff);
- // Row 3
- s0 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(s + 3 * p)), u0);
- d0 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(d + 3 * dp)), u0);
- diff = _mm_subs_epi16(s0, d0);
- negdiff = _mm_subs_epi16(u0, diff);
- absdiff = _mm_max_epi16(diff, negdiff);
- maxabsdiff = _mm_max_epi16(maxabsdiff, absdiff);
- minabsdiff = _mm_min_epi16(minabsdiff, absdiff);
- // Row 4
- s0 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(s + 4 * p)), u0);
- d0 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(d + 4 * dp)), u0);
- diff = _mm_subs_epi16(s0, d0);
- negdiff = _mm_subs_epi16(u0, diff);
- absdiff = _mm_max_epi16(diff, negdiff);
- maxabsdiff = _mm_max_epi16(maxabsdiff, absdiff);
- minabsdiff = _mm_min_epi16(minabsdiff, absdiff);
- // Row 5
- s0 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(s + 5 * p)), u0);
- d0 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(d + 5 * dp)), u0);
- diff = _mm_subs_epi16(s0, d0);
- negdiff = _mm_subs_epi16(u0, diff);
- absdiff = _mm_max_epi16(diff, negdiff);
- maxabsdiff = _mm_max_epi16(maxabsdiff, absdiff);
- minabsdiff = _mm_min_epi16(minabsdiff, absdiff);
- // Row 6
- s0 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(s + 6 * p)), u0);
- d0 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(d + 6 * dp)), u0);
- diff = _mm_subs_epi16(s0, d0);
- negdiff = _mm_subs_epi16(u0, diff);
- absdiff = _mm_max_epi16(diff, negdiff);
- maxabsdiff = _mm_max_epi16(maxabsdiff, absdiff);
- minabsdiff = _mm_min_epi16(minabsdiff, absdiff);
- // Row 7
- s0 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(s + 7 * p)), u0);
- d0 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(d + 7 * dp)), u0);
- diff = _mm_subs_epi16(s0, d0);
- negdiff = _mm_subs_epi16(u0, diff);
- absdiff = _mm_max_epi16(diff, negdiff);
- maxabsdiff = _mm_max_epi16(maxabsdiff, absdiff);
- minabsdiff = _mm_min_epi16(minabsdiff, absdiff);
-
- maxabsdiff = _mm_max_epi16(maxabsdiff, _mm_srli_si128(maxabsdiff, 8));
- maxabsdiff = _mm_max_epi16(maxabsdiff, _mm_srli_epi64(maxabsdiff, 32));
- maxabsdiff = _mm_max_epi16(maxabsdiff, _mm_srli_epi64(maxabsdiff, 16));
- *max = _mm_extract_epi16(maxabsdiff, 0);
-
- minabsdiff = _mm_min_epi16(minabsdiff, _mm_srli_si128(minabsdiff, 8));
- minabsdiff = _mm_min_epi16(minabsdiff, _mm_srli_epi64(minabsdiff, 32));
- minabsdiff = _mm_min_epi16(minabsdiff, _mm_srli_epi64(minabsdiff, 16));
- *min = _mm_extract_epi16(minabsdiff, 0);
-}
-
-unsigned int vp10_avg_8x8_sse2(const uint8_t *s, int p) {
- __m128i s0, s1, u0;
- unsigned int avg = 0;
- u0 = _mm_setzero_si128();
- s0 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(s)), u0);
- s1 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(s + p)), u0);
- s0 = _mm_adds_epu16(s0, s1);
- s1 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(s + 2 * p)), u0);
- s0 = _mm_adds_epu16(s0, s1);
- s1 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(s + 3 * p)), u0);
- s0 = _mm_adds_epu16(s0, s1);
- s1 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(s + 4 * p)), u0);
- s0 = _mm_adds_epu16(s0, s1);
- s1 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(s + 5 * p)), u0);
- s0 = _mm_adds_epu16(s0, s1);
- s1 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(s + 6 * p)), u0);
- s0 = _mm_adds_epu16(s0, s1);
- s1 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(s + 7 * p)), u0);
- s0 = _mm_adds_epu16(s0, s1);
-
- s0 = _mm_adds_epu16(s0, _mm_srli_si128(s0, 8));
- s0 = _mm_adds_epu16(s0, _mm_srli_epi64(s0, 32));
- s0 = _mm_adds_epu16(s0, _mm_srli_epi64(s0, 16));
- avg = _mm_extract_epi16(s0, 0);
- return (avg + 32) >> 6;
-}
-
-unsigned int vp10_avg_4x4_sse2(const uint8_t *s, int p) {
- __m128i s0, s1, u0;
- unsigned int avg = 0;
- u0 = _mm_setzero_si128();
- s0 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(s)), u0);
- s1 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(s + p)), u0);
- s0 = _mm_adds_epu16(s0, s1);
- s1 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(s + 2 * p)), u0);
- s0 = _mm_adds_epu16(s0, s1);
- s1 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)(s + 3 * p)), u0);
- s0 = _mm_adds_epu16(s0, s1);
-
- s0 = _mm_adds_epu16(s0, _mm_srli_si128(s0, 4));
- s0 = _mm_adds_epu16(s0, _mm_srli_epi64(s0, 16));
- avg = _mm_extract_epi16(s0, 0);
- return (avg + 8) >> 4;
-}
-
-static void hadamard_col8_sse2(__m128i *in, int iter) {
- __m128i a0 = in[0];
- __m128i a1 = in[1];
- __m128i a2 = in[2];
- __m128i a3 = in[3];
- __m128i a4 = in[4];
- __m128i a5 = in[5];
- __m128i a6 = in[6];
- __m128i a7 = in[7];
-
- __m128i b0 = _mm_add_epi16(a0, a1);
- __m128i b1 = _mm_sub_epi16(a0, a1);
- __m128i b2 = _mm_add_epi16(a2, a3);
- __m128i b3 = _mm_sub_epi16(a2, a3);
- __m128i b4 = _mm_add_epi16(a4, a5);
- __m128i b5 = _mm_sub_epi16(a4, a5);
- __m128i b6 = _mm_add_epi16(a6, a7);
- __m128i b7 = _mm_sub_epi16(a6, a7);
-
- a0 = _mm_add_epi16(b0, b2);
- a1 = _mm_add_epi16(b1, b3);
- a2 = _mm_sub_epi16(b0, b2);
- a3 = _mm_sub_epi16(b1, b3);
- a4 = _mm_add_epi16(b4, b6);
- a5 = _mm_add_epi16(b5, b7);
- a6 = _mm_sub_epi16(b4, b6);
- a7 = _mm_sub_epi16(b5, b7);
-
- if (iter == 0) {
- b0 = _mm_add_epi16(a0, a4);
- b7 = _mm_add_epi16(a1, a5);
- b3 = _mm_add_epi16(a2, a6);
- b4 = _mm_add_epi16(a3, a7);
- b2 = _mm_sub_epi16(a0, a4);
- b6 = _mm_sub_epi16(a1, a5);
- b1 = _mm_sub_epi16(a2, a6);
- b5 = _mm_sub_epi16(a3, a7);
-
- a0 = _mm_unpacklo_epi16(b0, b1);
- a1 = _mm_unpacklo_epi16(b2, b3);
- a2 = _mm_unpackhi_epi16(b0, b1);
- a3 = _mm_unpackhi_epi16(b2, b3);
- a4 = _mm_unpacklo_epi16(b4, b5);
- a5 = _mm_unpacklo_epi16(b6, b7);
- a6 = _mm_unpackhi_epi16(b4, b5);
- a7 = _mm_unpackhi_epi16(b6, b7);
-
- b0 = _mm_unpacklo_epi32(a0, a1);
- b1 = _mm_unpacklo_epi32(a4, a5);
- b2 = _mm_unpackhi_epi32(a0, a1);
- b3 = _mm_unpackhi_epi32(a4, a5);
- b4 = _mm_unpacklo_epi32(a2, a3);
- b5 = _mm_unpacklo_epi32(a6, a7);
- b6 = _mm_unpackhi_epi32(a2, a3);
- b7 = _mm_unpackhi_epi32(a6, a7);
-
- in[0] = _mm_unpacklo_epi64(b0, b1);
- in[1] = _mm_unpackhi_epi64(b0, b1);
- in[2] = _mm_unpacklo_epi64(b2, b3);
- in[3] = _mm_unpackhi_epi64(b2, b3);
- in[4] = _mm_unpacklo_epi64(b4, b5);
- in[5] = _mm_unpackhi_epi64(b4, b5);
- in[6] = _mm_unpacklo_epi64(b6, b7);
- in[7] = _mm_unpackhi_epi64(b6, b7);
- } else {
- in[0] = _mm_add_epi16(a0, a4);
- in[7] = _mm_add_epi16(a1, a5);
- in[3] = _mm_add_epi16(a2, a6);
- in[4] = _mm_add_epi16(a3, a7);
- in[2] = _mm_sub_epi16(a0, a4);
- in[6] = _mm_sub_epi16(a1, a5);
- in[1] = _mm_sub_epi16(a2, a6);
- in[5] = _mm_sub_epi16(a3, a7);
- }
-}
-
-void vp10_hadamard_8x8_sse2(int16_t const *src_diff, int src_stride,
- int16_t *coeff) {
- __m128i src[8];
- src[0] = _mm_load_si128((const __m128i *)src_diff);
- src[1] = _mm_load_si128((const __m128i *)(src_diff += src_stride));
- src[2] = _mm_load_si128((const __m128i *)(src_diff += src_stride));
- src[3] = _mm_load_si128((const __m128i *)(src_diff += src_stride));
- src[4] = _mm_load_si128((const __m128i *)(src_diff += src_stride));
- src[5] = _mm_load_si128((const __m128i *)(src_diff += src_stride));
- src[6] = _mm_load_si128((const __m128i *)(src_diff += src_stride));
- src[7] = _mm_load_si128((const __m128i *)(src_diff += src_stride));
-
- hadamard_col8_sse2(src, 0);
- hadamard_col8_sse2(src, 1);
-
- _mm_store_si128((__m128i *)coeff, src[0]);
- coeff += 8;
- _mm_store_si128((__m128i *)coeff, src[1]);
- coeff += 8;
- _mm_store_si128((__m128i *)coeff, src[2]);
- coeff += 8;
- _mm_store_si128((__m128i *)coeff, src[3]);
- coeff += 8;
- _mm_store_si128((__m128i *)coeff, src[4]);
- coeff += 8;
- _mm_store_si128((__m128i *)coeff, src[5]);
- coeff += 8;
- _mm_store_si128((__m128i *)coeff, src[6]);
- coeff += 8;
- _mm_store_si128((__m128i *)coeff, src[7]);
-}
-
-void vp10_hadamard_16x16_sse2(int16_t const *src_diff, int src_stride,
- int16_t *coeff) {
- int idx;
- for (idx = 0; idx < 4; ++idx) {
- int16_t const *src_ptr = src_diff + (idx >> 1) * 8 * src_stride
- + (idx & 0x01) * 8;
- vp10_hadamard_8x8_sse2(src_ptr, src_stride, coeff + idx * 64);
- }
-
- for (idx = 0; idx < 64; idx += 8) {
- __m128i coeff0 = _mm_load_si128((const __m128i *)coeff);
- __m128i coeff1 = _mm_load_si128((const __m128i *)(coeff + 64));
- __m128i coeff2 = _mm_load_si128((const __m128i *)(coeff + 128));
- __m128i coeff3 = _mm_load_si128((const __m128i *)(coeff + 192));
-
- __m128i b0 = _mm_add_epi16(coeff0, coeff1);
- __m128i b1 = _mm_sub_epi16(coeff0, coeff1);
- __m128i b2 = _mm_add_epi16(coeff2, coeff3);
- __m128i b3 = _mm_sub_epi16(coeff2, coeff3);
-
- b0 = _mm_srai_epi16(b0, 1);
- b1 = _mm_srai_epi16(b1, 1);
- b2 = _mm_srai_epi16(b2, 1);
- b3 = _mm_srai_epi16(b3, 1);
-
- coeff0 = _mm_add_epi16(b0, b2);
- coeff1 = _mm_add_epi16(b1, b3);
- _mm_store_si128((__m128i *)coeff, coeff0);
- _mm_store_si128((__m128i *)(coeff + 64), coeff1);
-
- coeff2 = _mm_sub_epi16(b0, b2);
- coeff3 = _mm_sub_epi16(b1, b3);
- _mm_store_si128((__m128i *)(coeff + 128), coeff2);
- _mm_store_si128((__m128i *)(coeff + 192), coeff3);
-
- coeff += 8;
- }
-}
-
-int16_t vp10_satd_sse2(const int16_t *coeff, int length) {
- int i;
- __m128i sum = _mm_load_si128((const __m128i *)coeff);
- __m128i sign = _mm_srai_epi16(sum, 15);
- __m128i val = _mm_xor_si128(sum, sign);
- sum = _mm_sub_epi16(val, sign);
- coeff += 8;
-
- for (i = 8; i < length; i += 8) {
- __m128i src_line = _mm_load_si128((const __m128i *)coeff);
- sign = _mm_srai_epi16(src_line, 15);
- val = _mm_xor_si128(src_line, sign);
- val = _mm_sub_epi16(val, sign);
- sum = _mm_add_epi16(sum, val);
- coeff += 8;
- }
-
- val = _mm_srli_si128(sum, 8);
- sum = _mm_add_epi16(sum, val);
- val = _mm_srli_epi64(sum, 32);
- sum = _mm_add_epi16(sum, val);
- val = _mm_srli_epi32(sum, 16);
- sum = _mm_add_epi16(sum, val);
-
- return _mm_extract_epi16(sum, 0);
-}
-
-void vp10_int_pro_row_sse2(int16_t *hbuf, uint8_t const*ref,
- const int ref_stride, const int height) {
- int idx;
- __m128i zero = _mm_setzero_si128();
- __m128i src_line = _mm_loadu_si128((const __m128i *)ref);
- __m128i s0 = _mm_unpacklo_epi8(src_line, zero);
- __m128i s1 = _mm_unpackhi_epi8(src_line, zero);
- __m128i t0, t1;
- int height_1 = height - 1;
- ref += ref_stride;
-
- for (idx = 1; idx < height_1; idx += 2) {
- src_line = _mm_loadu_si128((const __m128i *)ref);
- t0 = _mm_unpacklo_epi8(src_line, zero);
- t1 = _mm_unpackhi_epi8(src_line, zero);
- s0 = _mm_adds_epu16(s0, t0);
- s1 = _mm_adds_epu16(s1, t1);
- ref += ref_stride;
-
- src_line = _mm_loadu_si128((const __m128i *)ref);
- t0 = _mm_unpacklo_epi8(src_line, zero);
- t1 = _mm_unpackhi_epi8(src_line, zero);
- s0 = _mm_adds_epu16(s0, t0);
- s1 = _mm_adds_epu16(s1, t1);
- ref += ref_stride;
- }
-
- src_line = _mm_loadu_si128((const __m128i *)ref);
- t0 = _mm_unpacklo_epi8(src_line, zero);
- t1 = _mm_unpackhi_epi8(src_line, zero);
- s0 = _mm_adds_epu16(s0, t0);
- s1 = _mm_adds_epu16(s1, t1);
-
- if (height == 64) {
- s0 = _mm_srai_epi16(s0, 5);
- s1 = _mm_srai_epi16(s1, 5);
- } else if (height == 32) {
- s0 = _mm_srai_epi16(s0, 4);
- s1 = _mm_srai_epi16(s1, 4);
- } else {
- s0 = _mm_srai_epi16(s0, 3);
- s1 = _mm_srai_epi16(s1, 3);
- }
-
- _mm_storeu_si128((__m128i *)hbuf, s0);
- hbuf += 8;
- _mm_storeu_si128((__m128i *)hbuf, s1);
-}
-
-int16_t vp10_int_pro_col_sse2(uint8_t const *ref, const int width) {
- __m128i zero = _mm_setzero_si128();
- __m128i src_line = _mm_load_si128((const __m128i *)ref);
- __m128i s0 = _mm_sad_epu8(src_line, zero);
- __m128i s1;
- int i;
-
- for (i = 16; i < width; i += 16) {
- ref += 16;
- src_line = _mm_load_si128((const __m128i *)ref);
- s1 = _mm_sad_epu8(src_line, zero);
- s0 = _mm_adds_epu16(s0, s1);
- }
-
- s1 = _mm_srli_si128(s0, 8);
- s0 = _mm_adds_epu16(s0, s1);
-
- return _mm_extract_epi16(s0, 0);
-}
-
-int vp10_vector_var_sse2(int16_t const *ref, int16_t const *src,
- const int bwl) {
- int idx;
- int width = 4 << bwl;
- int16_t mean;
- __m128i v0 = _mm_loadu_si128((const __m128i *)ref);
- __m128i v1 = _mm_load_si128((const __m128i *)src);
- __m128i diff = _mm_subs_epi16(v0, v1);
- __m128i sum = diff;
- __m128i sse = _mm_madd_epi16(diff, diff);
-
- ref += 8;
- src += 8;
-
- for (idx = 8; idx < width; idx += 8) {
- v0 = _mm_loadu_si128((const __m128i *)ref);
- v1 = _mm_load_si128((const __m128i *)src);
- diff = _mm_subs_epi16(v0, v1);
-
- sum = _mm_add_epi16(sum, diff);
- v0 = _mm_madd_epi16(diff, diff);
- sse = _mm_add_epi32(sse, v0);
-
- ref += 8;
- src += 8;
- }
-
- v0 = _mm_srli_si128(sum, 8);
- sum = _mm_add_epi16(sum, v0);
- v0 = _mm_srli_epi64(sum, 32);
- sum = _mm_add_epi16(sum, v0);
- v0 = _mm_srli_epi32(sum, 16);
- sum = _mm_add_epi16(sum, v0);
-
- v1 = _mm_srli_si128(sse, 8);
- sse = _mm_add_epi32(sse, v1);
- v1 = _mm_srli_epi64(sse, 32);
- sse = _mm_add_epi32(sse, v1);
-
- mean = _mm_extract_epi16(sum, 0);
-
- return _mm_cvtsi128_si32(sse) - ((mean * mean) >> (bwl + 2));
-}
diff --git a/vp10/encoder/x86/dct_ssse3_x86_64.asm b/vp10/encoder/x86/dct_ssse3_x86_64.asm
deleted file mode 100644
index 5e8adab..0000000
--- a/vp10/encoder/x86/dct_ssse3_x86_64.asm
+++ /dev/null
@@ -1,121 +0,0 @@
-;
-; Copyright (c) 2014 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
-%define private_prefix vp10
-
-%include "third_party/x86inc/x86inc.asm"
-
-; This file provides SSSE3 version of the forward transformation. Part
-; of the macro definitions are originally derived from the ffmpeg project.
-; The current version applies to x86 64-bit only.
-
-SECTION .text
-
-%if ARCH_X86_64
-; matrix transpose
-%macro INTERLEAVE_2X 4
- punpckh%1 m%4, m%2, m%3
- punpckl%1 m%2, m%3
- SWAP %3, %4
-%endmacro
-
-%macro TRANSPOSE8X8 9
- INTERLEAVE_2X wd, %1, %2, %9
- INTERLEAVE_2X wd, %3, %4, %9
- INTERLEAVE_2X wd, %5, %6, %9
- INTERLEAVE_2X wd, %7, %8, %9
-
- INTERLEAVE_2X dq, %1, %3, %9
- INTERLEAVE_2X dq, %2, %4, %9
- INTERLEAVE_2X dq, %5, %7, %9
- INTERLEAVE_2X dq, %6, %8, %9
-
- INTERLEAVE_2X qdq, %1, %5, %9
- INTERLEAVE_2X qdq, %3, %7, %9
- INTERLEAVE_2X qdq, %2, %6, %9
- INTERLEAVE_2X qdq, %4, %8, %9
-
- SWAP %2, %5
- SWAP %4, %7
-%endmacro
-
-%macro HMD8_1D 0
- psubw m8, m0, m1
- psubw m9, m2, m3
- paddw m0, m1
- paddw m2, m3
- SWAP 1, 8
- SWAP 3, 9
- psubw m8, m4, m5
- psubw m9, m6, m7
- paddw m4, m5
- paddw m6, m7
- SWAP 5, 8
- SWAP 7, 9
-
- psubw m8, m0, m2
- psubw m9, m1, m3
- paddw m0, m2
- paddw m1, m3
- SWAP 2, 8
- SWAP 3, 9
- psubw m8, m4, m6
- psubw m9, m5, m7
- paddw m4, m6
- paddw m5, m7
- SWAP 6, 8
- SWAP 7, 9
-
- psubw m8, m0, m4
- psubw m9, m1, m5
- paddw m0, m4
- paddw m1, m5
- SWAP 4, 8
- SWAP 5, 9
- psubw m8, m2, m6
- psubw m9, m3, m7
- paddw m2, m6
- paddw m3, m7
- SWAP 6, 8
- SWAP 7, 9
-%endmacro
-
-INIT_XMM ssse3
-cglobal hadamard_8x8, 3, 5, 10, input, stride, output
- lea r3, [2 * strideq]
- lea r4, [4 * strideq]
-
- mova m0, [inputq]
- mova m1, [inputq + r3]
- lea inputq, [inputq + r4]
- mova m2, [inputq]
- mova m3, [inputq + r3]
- lea inputq, [inputq + r4]
- mova m4, [inputq]
- mova m5, [inputq + r3]
- lea inputq, [inputq + r4]
- mova m6, [inputq]
- mova m7, [inputq + r3]
-
- HMD8_1D
- TRANSPOSE8X8 0, 1, 2, 3, 4, 5, 6, 7, 9
- HMD8_1D
-
- mova [outputq + 0], m0
- mova [outputq + 16], m1
- mova [outputq + 32], m2
- mova [outputq + 48], m3
- mova [outputq + 64], m4
- mova [outputq + 80], m5
- mova [outputq + 96], m6
- mova [outputq + 112], m7
-
- RET
-%endif
diff --git a/vp10/vp10cx.mk b/vp10/vp10cx.mk
index ead993a..dc3b271 100644
--- a/vp10/vp10cx.mk
+++ b/vp10/vp10cx.mk
@@ -17,7 +17,6 @@
VP10_CX_SRCS-yes += vp10_cx_iface.c
-VP10_CX_SRCS-yes += encoder/avg.c
VP10_CX_SRCS-yes += encoder/bitstream.c
VP10_CX_SRCS-yes += encoder/context_tree.c
VP10_CX_SRCS-yes += encoder/context_tree.h
@@ -87,7 +86,6 @@
VP10_CX_SRCS-yes += encoder/mbgraph.c
VP10_CX_SRCS-yes += encoder/mbgraph.h
-VP10_CX_SRCS-$(HAVE_SSE2) += encoder/x86/avg_intrin_sse2.c
VP10_CX_SRCS-$(HAVE_SSE2) += encoder/x86/temporal_filter_apply_sse2.asm
VP10_CX_SRCS-$(HAVE_SSE2) += encoder/x86/quantize_sse2.c
ifeq ($(CONFIG_VP9_HIGHBITDEPTH),yes)
@@ -102,7 +100,6 @@
ifeq ($(ARCH_X86_64),yes)
ifeq ($(CONFIG_USE_X86INC),yes)
VP10_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/quantize_ssse3_x86_64.asm
-VP10_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/dct_ssse3_x86_64.asm
endif
endif
@@ -119,10 +116,8 @@
VP10_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/dct_neon.c
VP10_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/error_neon.c
endif
-VP10_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/avg_neon.c
VP10_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/quantize_neon.c
-VP10_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/avg_msa.c
VP10_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/error_msa.c
VP10_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/fdct4x4_msa.c
VP10_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/fdct8x8_msa.c
diff --git a/vp9/common/vp9_rtcd_defs.pl b/vp9/common/vp9_rtcd_defs.pl
index d166bbf..d6c86fe 100644
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -194,42 +194,6 @@
#
if (vpx_config("CONFIG_VP9_ENCODER") eq "yes") {
-add_proto qw/unsigned int vp9_avg_8x8/, "const uint8_t *, int p";
-specialize qw/vp9_avg_8x8 sse2 neon msa/;
-
-add_proto qw/unsigned int vp9_avg_4x4/, "const uint8_t *, int p";
-specialize qw/vp9_avg_4x4 sse2 neon msa/;
-
-add_proto qw/void vp9_minmax_8x8/, "const uint8_t *s, int p, const uint8_t *d, int dp, int *min, int *max";
-specialize qw/vp9_minmax_8x8 sse2/;
-
-add_proto qw/void vp9_hadamard_8x8/, "int16_t const *src_diff, int src_stride, int16_t *coeff";
-specialize qw/vp9_hadamard_8x8 sse2/, "$ssse3_x86_64_x86inc";
-
-add_proto qw/void vp9_hadamard_16x16/, "int16_t const *src_diff, int src_stride, int16_t *coeff";
-specialize qw/vp9_hadamard_16x16 sse2/;
-
-add_proto qw/int vp9_satd/, "const int16_t *coeff, int length";
-specialize qw/vp9_satd sse2 neon/;
-
-add_proto qw/void vp9_int_pro_row/, "int16_t *hbuf, uint8_t const *ref, const int ref_stride, const int height";
-specialize qw/vp9_int_pro_row sse2 neon/;
-
-add_proto qw/int16_t vp9_int_pro_col/, "uint8_t const *ref, const int width";
-specialize qw/vp9_int_pro_col sse2 neon/;
-
-add_proto qw/int vp9_vector_var/, "int16_t const *ref, int16_t const *src, const int bwl";
-specialize qw/vp9_vector_var neon sse2/;
-
-if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
- add_proto qw/unsigned int vp9_highbd_avg_8x8/, "const uint8_t *, int p";
- specialize qw/vp9_highbd_avg_8x8/;
- add_proto qw/unsigned int vp9_highbd_avg_4x4/, "const uint8_t *, int p";
- specialize qw/vp9_highbd_avg_4x4/;
- add_proto qw/void vp9_highbd_minmax_8x8/, "const uint8_t *s, int p, const uint8_t *d, int dp, int *min, int *max";
- specialize qw/vp9_highbd_minmax_8x8/;
-}
-
# ENCODEMB INVOKE
#
diff --git a/vp9/encoder/vp9_aq_cyclicrefresh.c b/vp9/encoder/vp9_aq_cyclicrefresh.c
index 0def2cf..7c71d9d 100644
--- a/vp9/encoder/vp9_aq_cyclicrefresh.c
+++ b/vp9/encoder/vp9_aq_cyclicrefresh.c
@@ -191,7 +191,8 @@
BLOCK_SIZE bsize,
int64_t rate,
int64_t dist,
- int skip) {
+ int skip,
+ struct macroblock_plane *const p) {
const VP9_COMMON *const cm = &cpi->common;
CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
const int bw = num_8x8_blocks_wide_lookup[bsize];
@@ -199,12 +200,33 @@
const int xmis = VPXMIN(cm->mi_cols - mi_col, bw);
const int ymis = VPXMIN(cm->mi_rows - mi_row, bh);
const int block_index = mi_row * cm->mi_cols + mi_col;
- const int refresh_this_block = candidate_refresh_aq(cr, mbmi, rate, dist,
- bsize);
+ int refresh_this_block = candidate_refresh_aq(cr, mbmi, rate, dist, bsize);
// Default is to not update the refresh map.
int new_map_value = cr->map[block_index];
int x = 0; int y = 0;
+ int is_skin = 0;
+ if (refresh_this_block == 0 &&
+ bsize <= BLOCK_16X16 &&
+ cpi->oxcf.content != VP9E_CONTENT_SCREEN) {
+ // Take center pixel in block to determine is_skin.
+ const int y_width_shift = (4 << b_width_log2_lookup[bsize]) >> 1;
+ const int y_height_shift = (4 << b_height_log2_lookup[bsize]) >> 1;
+ const int uv_width_shift = y_width_shift >> 1;
+ const int uv_height_shift = y_height_shift >> 1;
+ const int stride = p[0].src.stride;
+ const int strideuv = p[1].src.stride;
+ const uint8_t ysource =
+ p[0].src.buf[y_height_shift * stride + y_width_shift];
+ const uint8_t usource =
+ p[1].src.buf[uv_height_shift * strideuv + uv_width_shift];
+ const uint8_t vsource =
+ p[2].src.buf[uv_height_shift * strideuv + uv_width_shift];
+ is_skin = vp9_skin_pixel(ysource, usource, vsource);
+ if (is_skin)
+ refresh_this_block = 1;
+ }
+
// If this block is labeled for refresh, check if we should reset the
// segment_id.
if (cyclic_refresh_segment_id_boosted(mbmi->segment_id)) {
diff --git a/vp9/encoder/vp9_aq_cyclicrefresh.h b/vp9/encoder/vp9_aq_cyclicrefresh.h
index a5b3813..edf0a97 100644
--- a/vp9/encoder/vp9_aq_cyclicrefresh.h
+++ b/vp9/encoder/vp9_aq_cyclicrefresh.h
@@ -14,6 +14,8 @@
#include "vpx/vpx_integer.h"
#include "vp9/common/vp9_blockd.h"
+#include "vp9/encoder/vp9_block.h"
+#include "vp9/encoder/vp9_skin_detection.h"
#ifdef __cplusplus
extern "C" {
@@ -93,7 +95,8 @@
void vp9_cyclic_refresh_update_segment(struct VP9_COMP *const cpi,
MB_MODE_INFO *const mbmi,
int mi_row, int mi_col, BLOCK_SIZE bsize,
- int64_t rate, int64_t dist, int skip);
+ int64_t rate, int64_t dist, int skip,
+ struct macroblock_plane *const p);
void vp9_cyclic_refresh_update_sb_postencode(struct VP9_COMP *const cpi,
const MB_MODE_INFO *const mbmi,
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index c643b18..c07eee9 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -496,6 +496,8 @@
threshold_base = 3 * threshold_base;
else if (noise_level == kMedium)
threshold_base = threshold_base << 1;
+ else if (noise_level < kLow)
+ threshold_base = (7 * threshold_base) >> 3;
}
if (cm->width <= 352 && cm->height <= 288) {
thresholds[0] = threshold_base >> 3;
@@ -556,16 +558,16 @@
if (x8_idx < pixels_wide && y8_idx < pixels_high) {
#if CONFIG_VP9_HIGHBITDEPTH
if (highbd_flag & YV12_FLAG_HIGHBITDEPTH) {
- vp9_highbd_minmax_8x8(s + y8_idx * sp + x8_idx, sp,
+ vpx_highbd_minmax_8x8(s + y8_idx * sp + x8_idx, sp,
d + y8_idx * dp + x8_idx, dp,
&min, &max);
} else {
- vp9_minmax_8x8(s + y8_idx * sp + x8_idx, sp,
+ vpx_minmax_8x8(s + y8_idx * sp + x8_idx, sp,
d + y8_idx * dp + x8_idx, dp,
&min, &max);
}
#else
- vp9_minmax_8x8(s + y8_idx * sp + x8_idx, sp,
+ vpx_minmax_8x8(s + y8_idx * sp + x8_idx, sp,
d + y8_idx * dp + x8_idx, dp,
&min, &max);
#endif
@@ -597,18 +599,18 @@
int d_avg = 128;
#if CONFIG_VP9_HIGHBITDEPTH
if (highbd_flag & YV12_FLAG_HIGHBITDEPTH) {
- s_avg = vp9_highbd_avg_4x4(s + y4_idx * sp + x4_idx, sp);
+ s_avg = vpx_highbd_avg_4x4(s + y4_idx * sp + x4_idx, sp);
if (!is_key_frame)
- d_avg = vp9_highbd_avg_4x4(d + y4_idx * dp + x4_idx, dp);
+ d_avg = vpx_highbd_avg_4x4(d + y4_idx * dp + x4_idx, dp);
} else {
- s_avg = vp9_avg_4x4(s + y4_idx * sp + x4_idx, sp);
+ s_avg = vpx_avg_4x4(s + y4_idx * sp + x4_idx, sp);
if (!is_key_frame)
- d_avg = vp9_avg_4x4(d + y4_idx * dp + x4_idx, dp);
+ d_avg = vpx_avg_4x4(d + y4_idx * dp + x4_idx, dp);
}
#else
- s_avg = vp9_avg_4x4(s + y4_idx * sp + x4_idx, sp);
+ s_avg = vpx_avg_4x4(s + y4_idx * sp + x4_idx, sp);
if (!is_key_frame)
- d_avg = vp9_avg_4x4(d + y4_idx * dp + x4_idx, dp);
+ d_avg = vpx_avg_4x4(d + y4_idx * dp + x4_idx, dp);
#endif
sum = s_avg - d_avg;
sse = sum * sum;
@@ -636,18 +638,18 @@
int d_avg = 128;
#if CONFIG_VP9_HIGHBITDEPTH
if (highbd_flag & YV12_FLAG_HIGHBITDEPTH) {
- s_avg = vp9_highbd_avg_8x8(s + y8_idx * sp + x8_idx, sp);
+ s_avg = vpx_highbd_avg_8x8(s + y8_idx * sp + x8_idx, sp);
if (!is_key_frame)
- d_avg = vp9_highbd_avg_8x8(d + y8_idx * dp + x8_idx, dp);
+ d_avg = vpx_highbd_avg_8x8(d + y8_idx * dp + x8_idx, dp);
} else {
- s_avg = vp9_avg_8x8(s + y8_idx * sp + x8_idx, sp);
+ s_avg = vpx_avg_8x8(s + y8_idx * sp + x8_idx, sp);
if (!is_key_frame)
- d_avg = vp9_avg_8x8(d + y8_idx * dp + x8_idx, dp);
+ d_avg = vpx_avg_8x8(d + y8_idx * dp + x8_idx, dp);
}
#else
- s_avg = vp9_avg_8x8(s + y8_idx * sp + x8_idx, sp);
+ s_avg = vpx_avg_8x8(s + y8_idx * sp + x8_idx, sp);
if (!is_key_frame)
- d_avg = vp9_avg_8x8(d + y8_idx * dp + x8_idx, dp);
+ d_avg = vpx_avg_8x8(d + y8_idx * dp + x8_idx, dp);
#endif
sum = s_avg - d_avg;
sse = sum * sum;
@@ -668,6 +670,8 @@
v64x64 vt;
v16x16 vt2[16];
int force_split[21];
+ int avg_32x32;
+ int avg_16x16[4];
uint8_t *s;
const uint8_t *d;
int sp;
@@ -676,9 +680,13 @@
int64_t thresholds[4] = {cpi->vbp_thresholds[0], cpi->vbp_thresholds[1],
cpi->vbp_thresholds[2], cpi->vbp_thresholds[3]};
+ // For the variance computation under SVC mode, we treat the frame as key if
+ // the reference (base layer frame) is key frame (i.e., is_key_frame == 1).
+ const int is_key_frame = (cm->frame_type == KEY_FRAME ||
+ (is_one_pass_cbr_svc(cpi) &&
+ cpi->svc.layer_context[cpi->svc.temporal_layer_id].is_key_frame));
// Always use 4x4 partition for key frame.
- const int is_key_frame = (cm->frame_type == KEY_FRAME);
- const int use_4x4_partition = is_key_frame;
+ const int use_4x4_partition = cm->frame_type == KEY_FRAME;
const int low_res = (cm->width <= 352 && cm->height <= 288);
int variance4x4downsample[16];
@@ -704,8 +712,7 @@
s = x->plane[0].src.buf;
sp = x->plane[0].src.stride;
- if (!is_key_frame && !(is_one_pass_cbr_svc(cpi) &&
- cpi->svc.layer_context[cpi->svc.temporal_layer_id].is_key_frame)) {
+ if (!is_key_frame) {
// In the case of spatial/temporal scalable coding, the assumption here is
// that the temporal reference frame will always be of type LAST_FRAME.
// TODO(marpan): If that assumption is broken, we need to revisit this code.
@@ -819,6 +826,7 @@
const int y32_idx = ((i >> 1) << 5);
const int i2 = i << 2;
force_split[i + 1] = 0;
+ avg_16x16[i] = 0;
for (j = 0; j < 4; j++) {
const int x16_idx = x32_idx + ((j & 1) << 4);
const int y16_idx = y32_idx + ((j >> 1) << 4);
@@ -836,6 +844,7 @@
is_key_frame);
fill_variance_tree(&vt.split[i].split[j], BLOCK_16X16);
get_variance(&vt.split[i].split[j].part_variances.none);
+ avg_16x16[i] += vt.split[i].split[j].part_variances.none.variance;
if (vt.split[i].split[j].part_variances.none.variance >
thresholds[2]) {
// 16X16 variance is above threshold for split, so force split to 8x8
@@ -862,9 +871,7 @@
}
}
}
- // TODO(marpan): There is an issue with variance based on 4x4 average in
- // svc mode, don't allow it for now.
- if (is_key_frame || (low_res && !cpi->use_svc &&
+ if (is_key_frame || (low_res &&
vt.split[i].split[j].part_variances.none.variance >
(thresholds[1] << 1))) {
force_split[split_index] = 0;
@@ -886,8 +893,8 @@
}
}
}
-
// Fill the rest of the variance tree by summing split partition values.
+ avg_32x32 = 0;
for (i = 0; i < 4; i++) {
const int i2 = i << 2;
for (j = 0; j < 4; j++) {
@@ -908,19 +915,30 @@
}
}
fill_variance_tree(&vt.split[i], BLOCK_32X32);
- // If variance of this 32x32 block is above the threshold, force the block
- // to split. This also forces a split on the upper (64x64) level.
+ // If variance of this 32x32 block is above the threshold, or if its above
+ // (some threshold of) the average variance over the sub-16x16 blocks, then
+ // force this block to split. This also forces a split on the upper
+ // (64x64) level.
if (!force_split[i + 1]) {
get_variance(&vt.split[i].part_variances.none);
- if (vt.split[i].part_variances.none.variance > thresholds[1]) {
+ if (vt.split[i].part_variances.none.variance > thresholds[1] ||
+ (!is_key_frame &&
+ vt.split[i].part_variances.none.variance > (thresholds[1] >> 1) &&
+ vt.split[i].part_variances.none.variance > (avg_16x16[i] >> 1))) {
force_split[i + 1] = 1;
force_split[0] = 1;
}
+ avg_32x32 += vt.split[i].part_variances.none.variance;
}
}
if (!force_split[0]) {
fill_variance_tree(&vt, BLOCK_64X64);
get_variance(&vt.part_variances.none);
+ // If variance of this 64x64 block is above (some threshold of) the average
+ // variance over the sub-32x32 blocks, then force this block to split.
+ if (!is_key_frame &&
+ vt.part_variances.none.variance > (5 * avg_32x32) >> 4)
+ force_split[0] = 1;
}
// Now go through the entire structure, splitting every block size until
@@ -1027,7 +1045,7 @@
if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ) {
vp9_cyclic_refresh_update_segment(cpi, &xd->mi[0]->mbmi, mi_row,
mi_col, bsize, ctx->rate, ctx->dist,
- x->skip);
+ x->skip, p);
}
}
@@ -1687,6 +1705,7 @@
MACROBLOCKD *const xd = &x->e_mbd;
MODE_INFO *const mi = xd->mi[0];
MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+ struct macroblock_plane *const p = x->plane;
const struct segmentation *const seg = &cm->seg;
const int bw = num_8x8_blocks_wide_lookup[mi->mbmi.sb_type];
const int bh = num_8x8_blocks_high_lookup[mi->mbmi.sb_type];
@@ -1707,7 +1726,7 @@
} else {
// Setting segmentation map for cyclic_refresh.
vp9_cyclic_refresh_update_segment(cpi, mbmi, mi_row, mi_col, bsize,
- ctx->rate, ctx->dist, x->skip);
+ ctx->rate, ctx->dist, x->skip, p);
}
vp9_init_plane_quantizers(cpi, x);
}
diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index 72fa828..e4681f6 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -1478,7 +1478,11 @@
cpi->td.mb.e_mbd.bd = (int)cm->bit_depth;
#endif // CONFIG_VP9_HIGHBITDEPTH
- rc->baseline_gf_interval = (MIN_GF_INTERVAL + MAX_GF_INTERVAL) / 2;
+ if ((oxcf->pass == 0) && (oxcf->rc_mode == VPX_Q)) {
+ rc->baseline_gf_interval = FIXED_GF_INTERVAL;
+ } else {
+ rc->baseline_gf_interval = (MIN_GF_INTERVAL + MAX_GF_INTERVAL) / 2;
+ }
cpi->refresh_golden_frame = 0;
cpi->refresh_last_frame = 1;
@@ -2793,6 +2797,22 @@
cpi->resize_pending);
}
#endif
+ if (is_one_pass_cbr_svc(cpi)) {
+ // Keep track of frame index for each reference frame.
+ SVC *const svc = &cpi->svc;
+ if (cm->frame_type == KEY_FRAME) {
+ svc->ref_frame_index[cpi->lst_fb_idx] = svc->current_superframe;
+ svc->ref_frame_index[cpi->gld_fb_idx] = svc->current_superframe;
+ svc->ref_frame_index[cpi->alt_fb_idx] = svc->current_superframe;
+ } else {
+ if (cpi->refresh_last_frame)
+ svc->ref_frame_index[cpi->lst_fb_idx] = svc->current_superframe;
+ if (cpi->refresh_golden_frame)
+ svc->ref_frame_index[cpi->gld_fb_idx] = svc->current_superframe;
+ if (cpi->refresh_alt_ref_frame)
+ svc->ref_frame_index[cpi->alt_fb_idx] = svc->current_superframe;
+ }
+ }
}
static void loopfilter_frame(VP9_COMP *cpi, VP9_COMMON *cm) {
diff --git a/vp9/encoder/vp9_mcomp.c b/vp9/encoder/vp9_mcomp.c
index 327ac19..a84202b 100644
--- a/vp9/encoder/vp9_mcomp.c
+++ b/vp9/encoder/vp9_mcomp.c
@@ -1755,7 +1755,7 @@
int center, offset = 0;
int bw = 4 << bwl; // redundant variable, to be changed in the experiments.
for (d = 0; d <= bw; d += 16) {
- this_sad = vp9_vector_var(&ref[d], src, bwl);
+ this_sad = vpx_vector_var(&ref[d], src, bwl);
if (this_sad < best_sad) {
best_sad = this_sad;
offset = d;
@@ -1768,7 +1768,7 @@
// check limit
if (this_pos < 0 || this_pos > bw)
continue;
- this_sad = vp9_vector_var(&ref[this_pos], src, bwl);
+ this_sad = vpx_vector_var(&ref[this_pos], src, bwl);
if (this_sad < best_sad) {
best_sad = this_sad;
center = this_pos;
@@ -1781,7 +1781,7 @@
// check limit
if (this_pos < 0 || this_pos > bw)
continue;
- this_sad = vp9_vector_var(&ref[this_pos], src, bwl);
+ this_sad = vpx_vector_var(&ref[this_pos], src, bwl);
if (this_sad < best_sad) {
best_sad = this_sad;
center = this_pos;
@@ -1794,7 +1794,7 @@
// check limit
if (this_pos < 0 || this_pos > bw)
continue;
- this_sad = vp9_vector_var(&ref[this_pos], src, bwl);
+ this_sad = vpx_vector_var(&ref[this_pos], src, bwl);
if (this_sad < best_sad) {
best_sad = this_sad;
center = this_pos;
@@ -1807,7 +1807,7 @@
// check limit
if (this_pos < 0 || this_pos > bw)
continue;
- this_sad = vp9_vector_var(&ref[this_pos], src, bwl);
+ this_sad = vpx_vector_var(&ref[this_pos], src, bwl);
if (this_sad < best_sad) {
best_sad = this_sad;
center = this_pos;
@@ -1876,25 +1876,25 @@
// Set up prediction 1-D reference set
ref_buf = xd->plane[0].pre[0].buf - (bw >> 1);
for (idx = 0; idx < search_width; idx += 16) {
- vp9_int_pro_row(&hbuf[idx], ref_buf, ref_stride, bh);
+ vpx_int_pro_row(&hbuf[idx], ref_buf, ref_stride, bh);
ref_buf += 16;
}
ref_buf = xd->plane[0].pre[0].buf - (bh >> 1) * ref_stride;
for (idx = 0; idx < search_height; ++idx) {
- vbuf[idx] = vp9_int_pro_col(ref_buf, bw) >> norm_factor;
+ vbuf[idx] = vpx_int_pro_col(ref_buf, bw) >> norm_factor;
ref_buf += ref_stride;
}
// Set up src 1-D reference set
for (idx = 0; idx < bw; idx += 16) {
src_buf = x->plane[0].src.buf + idx;
- vp9_int_pro_row(&src_hbuf[idx], src_buf, src_stride, bh);
+ vpx_int_pro_row(&src_hbuf[idx], src_buf, src_stride, bh);
}
src_buf = x->plane[0].src.buf;
for (idx = 0; idx < bh; ++idx) {
- src_vbuf[idx] = vp9_int_pro_col(src_buf, bw) >> norm_factor;
+ src_vbuf[idx] = vpx_int_pro_col(src_buf, bw) >> norm_factor;
src_buf += src_stride;
}
diff --git a/vp9/encoder/vp9_pickmode.c b/vp9/encoder/vp9_pickmode.c
index 90650db..b929758 100644
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -619,14 +619,14 @@
scan_order->scan, scan_order->iscan);
break;
case TX_16X16:
- vp9_hadamard_16x16(src_diff, diff_stride, (int16_t *)coeff);
+ vpx_hadamard_16x16(src_diff, diff_stride, (int16_t *)coeff);
vp9_quantize_fp(coeff, 256, x->skip_block, p->zbin, p->round_fp,
p->quant_fp, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
scan_order->scan, scan_order->iscan);
break;
case TX_8X8:
- vp9_hadamard_8x8(src_diff, diff_stride, (int16_t *)coeff);
+ vpx_hadamard_8x8(src_diff, diff_stride, (int16_t *)coeff);
vp9_quantize_fp(coeff, 64, x->skip_block, p->zbin, p->round_fp,
p->quant_fp, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
@@ -673,7 +673,7 @@
if (*eob == 1)
*rate += (int)abs(qcoeff[0]);
else if (*eob > 1)
- *rate += vp9_satd((const int16_t *)qcoeff, step << 4);
+ *rate += vpx_satd((const int16_t *)qcoeff, step << 4);
*dist += vp9_block_error_fp(coeff, dqcoeff, step << 4) >> shift;
}
@@ -1094,6 +1094,7 @@
BLOCK_SIZE bsize, PICK_MODE_CONTEXT *ctx) {
VP9_COMMON *const cm = &cpi->common;
SPEED_FEATURES *const sf = &cpi->sf;
+ const SVC *const svc = &cpi->svc;
TileInfo *const tile_info = &tile_data->tile_info;
MACROBLOCKD *const xd = &x->e_mbd;
MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
@@ -1143,6 +1144,7 @@
int best_pred_sad = INT_MAX;
int best_early_term = 0;
int ref_frame_cost[MAX_REF_FRAMES];
+ int svc_force_zero_mode[3] = {0};
#if CONFIG_VP9_TEMPORAL_DENOISING
int64_t zero_last_cost_orig = INT64_MAX;
#endif
@@ -1196,6 +1198,17 @@
} else {
usable_ref_frame = GOLDEN_FRAME;
}
+
+ // If the reference is temporally aligned with current superframe
+ // (e.g., spatial reference within superframe), constrain the inter mode:
+ // for now only test zero motion.
+ if (cpi->use_svc && svc ->force_zero_mode_spatial_ref) {
+ if (svc->ref_frame_index[cpi->lst_fb_idx] == svc->current_superframe)
+ svc_force_zero_mode[LAST_FRAME - 1] = 1;
+ if (svc->ref_frame_index[cpi->gld_fb_idx] == svc->current_superframe)
+ svc_force_zero_mode[GOLDEN_FRAME - 1] = 1;
+ }
+
for (ref_frame = LAST_FRAME; ref_frame <= usable_ref_frame; ++ref_frame) {
const YV12_BUFFER_CONFIG *yv12 = get_ref_frame_buffer(cpi, ref_frame);
@@ -1248,8 +1261,13 @@
continue;
ref_frame = ref_mode_set[idx].ref_frame;
- if (cpi->use_svc)
+ if (cpi->use_svc) {
ref_frame = ref_mode_set_svc[idx].ref_frame;
+ if (svc_force_zero_mode[ref_frame - 1] &&
+ frame_mv[this_mode][ref_frame].as_int != 0)
+ continue;
+ }
+
if (!(cpi->ref_frame_flags & flag_list[ref_frame]))
continue;
if (const_motion[ref_frame] && this_mode == NEARMV)
diff --git a/vp9/encoder/vp9_ratectrl.c b/vp9/encoder/vp9_ratectrl.c
index 8ab51cd..2579c60 100644
--- a/vp9/encoder/vp9_ratectrl.c
+++ b/vp9/encoder/vp9_ratectrl.c
@@ -833,10 +833,16 @@
ASSIGN_MINQ_TABLE(cm->bit_depth, inter_minq);
if (frame_is_intra_only(cm)) {
- // Handle the special case for key frames forced when we have reached
- // the maximum key frame interval. Here force the Q to a range
- // based on the ambient Q to reduce the risk of popping.
- if (rc->this_key_frame_forced) {
+ if (oxcf->rc_mode == VPX_Q) {
+ int qindex = cq_level;
+ double q = vp9_convert_qindex_to_q(qindex, cm->bit_depth);
+ int delta_qindex = vp9_compute_qdelta(rc, q, q * 0.25,
+ cm->bit_depth);
+ active_best_quality = VPXMAX(qindex + delta_qindex, rc->best_quality);
+ } else if (rc->this_key_frame_forced) {
+ // Handle the special case for key frames forced when we have reached
+ // the maximum key frame interval. Here force the Q to a range
+ // based on the ambient Q to reduce the risk of popping.
int qindex = rc->last_boosted_qindex;
double last_boosted_q = vp9_convert_qindex_to_q(qindex, cm->bit_depth);
int delta_qindex = vp9_compute_qdelta(rc, last_boosted_q,
@@ -886,17 +892,28 @@
active_best_quality = active_best_quality * 15 / 16;
} else if (oxcf->rc_mode == VPX_Q) {
- if (!cpi->refresh_alt_ref_frame) {
- active_best_quality = cq_level;
- } else {
- active_best_quality = get_gf_active_quality(rc, q, cm->bit_depth);
- }
+ int qindex = cq_level;
+ double q = vp9_convert_qindex_to_q(qindex, cm->bit_depth);
+ int delta_qindex;
+ if (cpi->refresh_alt_ref_frame)
+ delta_qindex = vp9_compute_qdelta(rc, q, q * 0.40, cm->bit_depth);
+ else
+ delta_qindex = vp9_compute_qdelta(rc, q, q * 0.50, cm->bit_depth);
+ active_best_quality = VPXMAX(qindex + delta_qindex, rc->best_quality);
} else {
active_best_quality = get_gf_active_quality(rc, q, cm->bit_depth);
}
} else {
if (oxcf->rc_mode == VPX_Q) {
- active_best_quality = cq_level;
+ int qindex = cq_level;
+ double q = vp9_convert_qindex_to_q(qindex, cm->bit_depth);
+ double delta_rate[FIXED_GF_INTERVAL] =
+ {0.50, 1.0, 0.85, 1.0, 0.70, 1.0, 0.85, 1.0};
+ int delta_qindex =
+ vp9_compute_qdelta(rc, q,
+ q * delta_rate[cm->current_video_frame %
+ FIXED_GF_INTERVAL], cm->bit_depth);
+ active_best_quality = VPXMAX(qindex + delta_qindex, rc->best_quality);
} else {
// Use the lower of active_worst_quality and recent/average Q.
if (cm->current_video_frame > 1)
@@ -1313,9 +1330,9 @@
}
}
} else {
- if (rc->is_src_frame_alt_ref ||
- !(cpi->refresh_golden_frame || cpi->refresh_alt_ref_frame) ||
- (cpi->use_svc && oxcf->rc_mode == VPX_CBR)) {
+ if ((cpi->use_svc && oxcf->rc_mode == VPX_CBR) ||
+ (!rc->is_src_frame_alt_ref &&
+ !(cpi->refresh_golden_frame || cpi->refresh_alt_ref_frame))) {
rc->last_q[INTER_FRAME] = qindex;
rc->avg_frame_qindex[INTER_FRAME] =
ROUND_POWER_OF_TWO(3 * rc->avg_frame_qindex[INTER_FRAME] + qindex, 2);
@@ -1722,29 +1739,36 @@
RATE_CONTROL *const rc) {
const VP9EncoderConfig *const oxcf = &cpi->oxcf;
- // Set Maximum gf/arf interval
- rc->max_gf_interval = oxcf->max_gf_interval;
- rc->min_gf_interval = oxcf->min_gf_interval;
- if (rc->min_gf_interval == 0)
- rc->min_gf_interval = vp9_rc_get_default_min_gf_interval(
- oxcf->width, oxcf->height, cpi->framerate);
- if (rc->max_gf_interval == 0)
- rc->max_gf_interval = vp9_rc_get_default_max_gf_interval(
- cpi->framerate, rc->min_gf_interval);
+ // Special case code for 1 pass fixed Q mode tests
+ if ((oxcf->pass == 0) && (oxcf->rc_mode == VPX_Q)) {
+ rc->max_gf_interval = FIXED_GF_INTERVAL;
+ rc->min_gf_interval = FIXED_GF_INTERVAL;
+ rc->static_scene_max_gf_interval = FIXED_GF_INTERVAL;
+ } else {
+ // Set Maximum gf/arf interval
+ rc->max_gf_interval = oxcf->max_gf_interval;
+ rc->min_gf_interval = oxcf->min_gf_interval;
+ if (rc->min_gf_interval == 0)
+ rc->min_gf_interval = vp9_rc_get_default_min_gf_interval(
+ oxcf->width, oxcf->height, cpi->framerate);
+ if (rc->max_gf_interval == 0)
+ rc->max_gf_interval = vp9_rc_get_default_max_gf_interval(
+ cpi->framerate, rc->min_gf_interval);
- // Extended interval for genuinely static scenes
- rc->static_scene_max_gf_interval = MAX_LAG_BUFFERS * 2;
+ // Extended interval for genuinely static scenes
+ rc->static_scene_max_gf_interval = MAX_LAG_BUFFERS * 2;
- if (is_altref_enabled(cpi)) {
- if (rc->static_scene_max_gf_interval > oxcf->lag_in_frames - 1)
- rc->static_scene_max_gf_interval = oxcf->lag_in_frames - 1;
+ if (is_altref_enabled(cpi)) {
+ if (rc->static_scene_max_gf_interval > oxcf->lag_in_frames - 1)
+ rc->static_scene_max_gf_interval = oxcf->lag_in_frames - 1;
+ }
+
+ if (rc->max_gf_interval > rc->static_scene_max_gf_interval)
+ rc->max_gf_interval = rc->static_scene_max_gf_interval;
+
+ // Clamp min to max
+ rc->min_gf_interval = VPXMIN(rc->min_gf_interval, rc->max_gf_interval);
}
-
- if (rc->max_gf_interval > rc->static_scene_max_gf_interval)
- rc->max_gf_interval = rc->static_scene_max_gf_interval;
-
- // Clamp min to max
- rc->min_gf_interval = VPXMIN(rc->min_gf_interval, rc->max_gf_interval);
}
void vp9_rc_update_framerate(VP9_COMP *cpi) {
diff --git a/vp9/encoder/vp9_ratectrl.h b/vp9/encoder/vp9_ratectrl.h
index 136fd3e..3df909c 100644
--- a/vp9/encoder/vp9_ratectrl.h
+++ b/vp9/encoder/vp9_ratectrl.h
@@ -26,6 +26,7 @@
#define MIN_GF_INTERVAL 4
#define MAX_GF_INTERVAL 16
+#define FIXED_GF_INTERVAL 8 // Used in some testing modes only
#define ONEHALFONLY_RESIZE 0
typedef enum {
diff --git a/vp9/encoder/vp9_speed_features.c b/vp9/encoder/vp9_speed_features.c
index 318d810..c5f0bad 100644
--- a/vp9/encoder/vp9_speed_features.c
+++ b/vp9/encoder/vp9_speed_features.c
@@ -394,7 +394,7 @@
sf->intra_y_mode_bsize_mask[i] = INTRA_DC_TM_H_V;
} else {
for (i = 0; i < BLOCK_SIZES; ++i)
- if (i >= BLOCK_16X16)
+ if (i > BLOCK_16X16)
sf->intra_y_mode_bsize_mask[i] = INTRA_DC;
else
// Use H and V intra mode for block sizes <= 16X16.
diff --git a/vp9/encoder/vp9_svc_layercontext.c b/vp9/encoder/vp9_svc_layercontext.c
index b0617c1..a4e7eb1 100644
--- a/vp9/encoder/vp9_svc_layercontext.c
+++ b/vp9/encoder/vp9_svc_layercontext.c
@@ -25,13 +25,17 @@
const VP9EncoderConfig *const oxcf = &cpi->oxcf;
int mi_rows = cpi->common.mi_rows;
int mi_cols = cpi->common.mi_cols;
- int sl, tl;
+ int sl, tl, i;
int alt_ref_idx = svc->number_spatial_layers;
svc->spatial_layer_id = 0;
svc->temporal_layer_id = 0;
svc->first_spatial_layer_to_encode = 0;
svc->rc_drop_superframe = 0;
+ svc->force_zero_mode_spatial_ref = 0;
+ svc->current_superframe = 0;
+ for (i = 0; i < REF_FRAMES; ++i)
+ svc->ref_frame_index[i] = -1;
if (cpi->oxcf.error_resilient_mode == 0 && cpi->oxcf.pass == 2) {
if (vpx_realloc_frame_buffer(&cpi->svc.empty_frame.img,
@@ -353,6 +357,8 @@
cpi->svc.number_temporal_layers];
++lc->current_video_frame_in_layer;
++lc->frames_from_key_frame;
+ if (cpi->svc.spatial_layer_id == cpi->svc.number_spatial_layers - 1)
+ ++cpi->svc.current_superframe;
}
int vp9_is_upper_layer_key_frame(const VP9_COMP *const cpi) {
@@ -542,6 +548,7 @@
int vp9_one_pass_cbr_svc_start_layer(VP9_COMP *const cpi) {
int width = 0, height = 0;
LAYER_CONTEXT *lc = NULL;
+ cpi->svc.force_zero_mode_spatial_ref = 1;
if (cpi->svc.temporal_layering_mode == VP9E_TEMPORAL_LAYERING_MODE_0212) {
set_flags_and_fb_idx_for_temporal_mode3(cpi);
diff --git a/vp9/encoder/vp9_svc_layercontext.h b/vp9/encoder/vp9_svc_layercontext.h
index 5dbf9b4..1f446d7 100644
--- a/vp9/encoder/vp9_svc_layercontext.h
+++ b/vp9/encoder/vp9_svc_layercontext.h
@@ -83,6 +83,9 @@
int ext_lst_fb_idx[VPX_MAX_LAYERS];
int ext_gld_fb_idx[VPX_MAX_LAYERS];
int ext_alt_fb_idx[VPX_MAX_LAYERS];
+ int ref_frame_index[REF_FRAMES];
+ int force_zero_mode_spatial_ref;
+ int current_superframe;
} SVC;
struct VP9_COMP;
diff --git a/vp9/encoder/vp9_temporal_filter.c b/vp9/encoder/vp9_temporal_filter.c
index 7e4c61f..015dbc0 100644
--- a/vp9/encoder/vp9_temporal_filter.c
+++ b/vp9/encoder/vp9_temporal_filter.c
@@ -205,15 +205,34 @@
for (i = 0, k = 0; i < block_height; i++) {
for (j = 0; j < block_width; j++, k++) {
- int src_byte = frame1[byte];
- int pixel_value = *frame2++;
+ int pixel_value = *frame2;
+ int diff_sse[9] = { 0 };
+ int idx, idy, index = 0;
- modifier = src_byte - pixel_value;
- // This is an integer approximation of:
- // float coeff = (3.0 * modifer * modifier) / pow(2, strength);
- // modifier = (int)roundf(coeff > 16 ? 0 : 16-coeff);
- modifier *= modifier;
+ for (idy = -1; idy <= 1; ++idy) {
+ for (idx = -1; idx <= 1; ++idx) {
+ int row = i + idy;
+ int col = j + idx;
+
+ if (row >= 0 && row < (int)block_height &&
+ col >= 0 && col < (int)block_width) {
+ int diff = frame1[byte + idy * (int)stride + idx] -
+ frame2[idy * (int)block_width + idx];
+ diff_sse[index] = diff * diff;
+ ++index;
+ }
+ }
+ }
+ assert(index > 0);
+
+ modifier = 0;
+ for (idx = 0; idx < 9; ++idx)
+ modifier += diff_sse[idx];
+
modifier *= 3;
+ modifier /= index;
+
+ ++frame2;
modifier += rounding;
modifier >>= strength;
@@ -406,56 +425,58 @@
if (mbd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
int adj_strength = strength + 2 * (mbd->bd - 8);
// Apply the filter (YUV)
- vp9_highbd_temporal_filter_apply(f->y_buffer + mb_y_offset,
- f->y_stride,
- predictor, 16, 16, adj_strength,
- filter_weight,
- accumulator, count);
- vp9_highbd_temporal_filter_apply(f->u_buffer + mb_uv_offset,
- f->uv_stride, predictor + 256,
- mb_uv_width, mb_uv_height,
- adj_strength,
- filter_weight, accumulator + 256,
- count + 256);
- vp9_highbd_temporal_filter_apply(f->v_buffer + mb_uv_offset,
- f->uv_stride, predictor + 512,
- mb_uv_width, mb_uv_height,
- adj_strength, filter_weight,
- accumulator + 512, count + 512);
+ vp9_highbd_temporal_filter_apply_c(f->y_buffer + mb_y_offset,
+ f->y_stride,
+ predictor, 16, 16, adj_strength,
+ filter_weight,
+ accumulator, count);
+ vp9_highbd_temporal_filter_apply_c(f->u_buffer + mb_uv_offset,
+ f->uv_stride, predictor + 256,
+ mb_uv_width, mb_uv_height,
+ adj_strength,
+ filter_weight, accumulator + 256,
+ count + 256);
+ vp9_highbd_temporal_filter_apply_c(f->v_buffer + mb_uv_offset,
+ f->uv_stride, predictor + 512,
+ mb_uv_width, mb_uv_height,
+ adj_strength, filter_weight,
+ accumulator + 512, count + 512);
} else {
// Apply the filter (YUV)
- vp9_temporal_filter_apply(f->y_buffer + mb_y_offset, f->y_stride,
- predictor, 16, 16,
- strength, filter_weight,
- accumulator, count);
- vp9_temporal_filter_apply(f->u_buffer + mb_uv_offset, f->uv_stride,
- predictor + 256,
- mb_uv_width, mb_uv_height, strength,
- filter_weight, accumulator + 256,
- count + 256);
- vp9_temporal_filter_apply(f->v_buffer + mb_uv_offset, f->uv_stride,
- predictor + 512,
- mb_uv_width, mb_uv_height, strength,
- filter_weight, accumulator + 512,
- count + 512);
+ vp9_temporal_filter_apply_c(f->y_buffer + mb_y_offset, f->y_stride,
+ predictor, 16, 16,
+ strength, filter_weight,
+ accumulator, count);
+ vp9_temporal_filter_apply_c(f->u_buffer + mb_uv_offset,
+ f->uv_stride,
+ predictor + 256,
+ mb_uv_width, mb_uv_height, strength,
+ filter_weight, accumulator + 256,
+ count + 256);
+ vp9_temporal_filter_apply_c(f->v_buffer + mb_uv_offset,
+ f->uv_stride,
+ predictor + 512,
+ mb_uv_width, mb_uv_height, strength,
+ filter_weight, accumulator + 512,
+ count + 512);
}
#else
// Apply the filter (YUV)
// TODO(jingning): Need SIMD optimization for this.
vp9_temporal_filter_apply_c(f->y_buffer + mb_y_offset, f->y_stride,
- predictor, 16, 16,
- strength, filter_weight,
- accumulator, count);
+ predictor, 16, 16,
+ strength, filter_weight,
+ accumulator, count);
vp9_temporal_filter_apply_c(f->u_buffer + mb_uv_offset, f->uv_stride,
- predictor + 256,
- mb_uv_width, mb_uv_height, strength,
- filter_weight, accumulator + 256,
- count + 256);
+ predictor + 256,
+ mb_uv_width, mb_uv_height, strength,
+ filter_weight, accumulator + 256,
+ count + 256);
vp9_temporal_filter_apply_c(f->v_buffer + mb_uv_offset, f->uv_stride,
- predictor + 512,
- mb_uv_width, mb_uv_height, strength,
- filter_weight, accumulator + 512,
- count + 512);
+ predictor + 512,
+ mb_uv_width, mb_uv_height, strength,
+ filter_weight, accumulator + 512,
+ count + 512);
#endif // CONFIG_VP9_HIGHBITDEPTH
}
}
diff --git a/vp9/vp9cx.mk b/vp9/vp9cx.mk
index 5918240..de688bf 100644
--- a/vp9/vp9cx.mk
+++ b/vp9/vp9cx.mk
@@ -17,7 +17,6 @@
VP9_CX_SRCS-yes += vp9_cx_iface.c
-VP9_CX_SRCS-yes += encoder/vp9_avg.c
VP9_CX_SRCS-yes += encoder/vp9_bitstream.c
VP9_CX_SRCS-yes += encoder/vp9_context_tree.c
VP9_CX_SRCS-yes += encoder/vp9_context_tree.h
@@ -93,7 +92,6 @@
VP9_CX_SRCS-yes += encoder/vp9_mbgraph.c
VP9_CX_SRCS-yes += encoder/vp9_mbgraph.h
-VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_avg_intrin_sse2.c
VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_temporal_filter_apply_sse2.asm
VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_quantize_sse2.c
VP9_CX_SRCS-$(HAVE_AVX) += encoder/x86/vp9_diamond_search_sad_avx.c
@@ -114,7 +112,6 @@
ifeq ($(ARCH_X86_64),yes)
ifeq ($(CONFIG_USE_X86INC),yes)
VP9_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/vp9_quantize_ssse3_x86_64.asm
-VP9_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/vp9_dct_ssse3_x86_64.asm
endif
endif
@@ -131,10 +128,8 @@
VP9_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/vp9_dct_neon.c
VP9_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/vp9_error_neon.c
endif
-VP9_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/vp9_avg_neon.c
VP9_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/vp9_quantize_neon.c
-VP9_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/vp9_avg_msa.c
VP9_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/vp9_error_msa.c
VP9_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/vp9_fdct4x4_msa.c
VP9_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/vp9_fdct8x8_msa.c
diff --git a/vp9/encoder/arm/neon/vp9_avg_neon.c b/vpx_dsp/arm/avg_neon.c
similarity index 93%
rename from vp9/encoder/arm/neon/vp9_avg_neon.c
rename to vpx_dsp/arm/avg_neon.c
index 78467ce..d054c41 100644
--- a/vp9/encoder/arm/neon/vp9_avg_neon.c
+++ b/vpx_dsp/arm/avg_neon.c
@@ -11,7 +11,7 @@
#include <arm_neon.h>
#include <assert.h>
-#include "./vp9_rtcd.h"
+#include "./vpx_dsp_rtcd.h"
#include "./vpx_config.h"
#include "vpx/vpx_integer.h"
@@ -24,7 +24,7 @@
return vget_lane_u32(c, 0);
}
-unsigned int vp9_avg_4x4_neon(const uint8_t *s, int p) {
+unsigned int vpx_avg_4x4_neon(const uint8_t *s, int p) {
uint16x8_t v_sum;
uint32x2_t v_s0 = vdup_n_u32(0);
uint32x2_t v_s1 = vdup_n_u32(0);
@@ -36,7 +36,7 @@
return (horizontal_add_u16x8(v_sum) + 8) >> 4;
}
-unsigned int vp9_avg_8x8_neon(const uint8_t *s, int p) {
+unsigned int vpx_avg_8x8_neon(const uint8_t *s, int p) {
uint8x8_t v_s0 = vld1_u8(s);
const uint8x8_t v_s1 = vld1_u8(s + p);
uint16x8_t v_sum = vaddl_u8(v_s0, v_s1);
@@ -64,7 +64,7 @@
// coeff: 16 bits, dynamic range [-32640, 32640].
// length: value range {16, 64, 256, 1024}.
-int vp9_satd_neon(const int16_t *coeff, int length) {
+int vpx_satd_neon(const int16_t *coeff, int length) {
const int16x4_t zero = vdup_n_s16(0);
int32x4_t accum = vdupq_n_s32(0);
@@ -89,7 +89,7 @@
}
}
-void vp9_int_pro_row_neon(int16_t hbuf[16], uint8_t const *ref,
+void vpx_int_pro_row_neon(int16_t hbuf[16], uint8_t const *ref,
const int ref_stride, const int height) {
int i;
uint16x8_t vec_sum_lo = vdupq_n_u16(0);
@@ -142,7 +142,7 @@
vst1q_s16(hbuf, vreinterpretq_s16_u16(vec_sum_hi));
}
-int16_t vp9_int_pro_col_neon(uint8_t const *ref, const int width) {
+int16_t vpx_int_pro_col_neon(uint8_t const *ref, const int width) {
int i;
uint16x8_t vec_sum = vdupq_n_u16(0);
@@ -158,7 +158,7 @@
// ref, src = [0, 510] - max diff = 16-bits
// bwl = {2, 3, 4}, width = {16, 32, 64}
-int vp9_vector_var_neon(int16_t const *ref, int16_t const *src, const int bwl) {
+int vpx_vector_var_neon(int16_t const *ref, int16_t const *src, const int bwl) {
int width = 4 << bwl;
int32x4_t sse = vdupq_n_s32(0);
int16x8_t total = vdupq_n_s16(0);
diff --git a/vp9/encoder/vp9_avg.c b/vpx_dsp/avg.c
similarity index 86%
rename from vp9/encoder/vp9_avg.c
rename to vpx_dsp/avg.c
index 7baa09a..26fe785 100644
--- a/vp9/encoder/vp9_avg.c
+++ b/vpx_dsp/avg.c
@@ -7,11 +7,12 @@
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
-#include "./vp9_rtcd.h"
-#include "vp9/common/vp9_common.h"
+#include <stdlib.h>
+
+#include "./vpx_dsp_rtcd.h"
#include "vpx_ports/mem.h"
-unsigned int vp9_avg_8x8_c(const uint8_t *s, int p) {
+unsigned int vpx_avg_8x8_c(const uint8_t *s, int p) {
int i, j;
int sum = 0;
for (i = 0; i < 8; ++i, s+=p)
@@ -20,7 +21,7 @@
return (sum + 32) >> 6;
}
-unsigned int vp9_avg_4x4_c(const uint8_t *s, int p) {
+unsigned int vpx_avg_4x4_c(const uint8_t *s, int p) {
int i, j;
int sum = 0;
for (i = 0; i < 4; ++i, s+=p)
@@ -61,7 +62,7 @@
coeff[5] = c3 - c7;
}
-void vp9_hadamard_8x8_c(int16_t const *src_diff, int src_stride,
+void vpx_hadamard_8x8_c(int16_t const *src_diff, int src_stride,
int16_t *coeff) {
int idx;
int16_t buffer[64];
@@ -84,14 +85,14 @@
}
// In place 16x16 2D Hadamard transform
-void vp9_hadamard_16x16_c(int16_t const *src_diff, int src_stride,
+void vpx_hadamard_16x16_c(int16_t const *src_diff, int src_stride,
int16_t *coeff) {
int idx;
for (idx = 0; idx < 4; ++idx) {
// src_diff: 9 bit, dynamic range [-255, 255]
int16_t const *src_ptr = src_diff + (idx >> 1) * 8 * src_stride
+ (idx & 0x01) * 8;
- vp9_hadamard_8x8_c(src_ptr, src_stride, coeff + idx * 64);
+ vpx_hadamard_8x8_c(src_ptr, src_stride, coeff + idx * 64);
}
// coeff: 15 bit, dynamic range [-16320, 16320]
@@ -117,7 +118,7 @@
// coeff: 16 bits, dynamic range [-32640, 32640].
// length: value range {16, 64, 256, 1024}.
-int vp9_satd_c(const int16_t *coeff, int length) {
+int vpx_satd_c(const int16_t *coeff, int length) {
int i;
int satd = 0;
for (i = 0; i < length; ++i)
@@ -129,7 +130,7 @@
// Integer projection onto row vectors.
// height: value range {16, 32, 64}.
-void vp9_int_pro_row_c(int16_t hbuf[16], uint8_t const *ref,
+void vpx_int_pro_row_c(int16_t hbuf[16], uint8_t const *ref,
const int ref_stride, const int height) {
int idx;
const int norm_factor = height >> 1;
@@ -146,7 +147,7 @@
}
// width: value range {16, 32, 64}.
-int16_t vp9_int_pro_col_c(uint8_t const *ref, const int width) {
+int16_t vpx_int_pro_col_c(uint8_t const *ref, const int width) {
int idx;
int16_t sum = 0;
// sum: 14 bit, dynamic range [0, 16320]
@@ -158,7 +159,7 @@
// ref: [0 - 510]
// src: [0 - 510]
// bwl: {2, 3, 4}
-int vp9_vector_var_c(int16_t const *ref, int16_t const *src,
+int vpx_vector_var_c(int16_t const *ref, int16_t const *src,
const int bwl) {
int i;
int width = 4 << bwl;
@@ -175,7 +176,7 @@
return var;
}
-void vp9_minmax_8x8_c(const uint8_t *s, int p, const uint8_t *d, int dp,
+void vpx_minmax_8x8_c(const uint8_t *s, int p, const uint8_t *d, int dp,
int *min, int *max) {
int i, j;
*min = 255;
@@ -190,7 +191,7 @@
}
#if CONFIG_VP9_HIGHBITDEPTH
-unsigned int vp9_highbd_avg_8x8_c(const uint8_t *s8, int p) {
+unsigned int vpx_highbd_avg_8x8_c(const uint8_t *s8, int p) {
int i, j;
int sum = 0;
const uint16_t* s = CONVERT_TO_SHORTPTR(s8);
@@ -200,7 +201,7 @@
return (sum + 32) >> 6;
}
-unsigned int vp9_highbd_avg_4x4_c(const uint8_t *s8, int p) {
+unsigned int vpx_highbd_avg_4x4_c(const uint8_t *s8, int p) {
int i, j;
int sum = 0;
const uint16_t* s = CONVERT_TO_SHORTPTR(s8);
@@ -210,7 +211,7 @@
return (sum + 8) >> 4;
}
-void vp9_highbd_minmax_8x8_c(const uint8_t *s8, int p, const uint8_t *d8,
+void vpx_highbd_minmax_8x8_c(const uint8_t *s8, int p, const uint8_t *d8,
int dp, int *min, int *max) {
int i, j;
const uint16_t* s = CONVERT_TO_SHORTPTR(s8);
diff --git a/vp9/encoder/mips/msa/vp9_avg_msa.c b/vpx_dsp/mips/avg_msa.c
similarity index 91%
rename from vp9/encoder/mips/msa/vp9_avg_msa.c
rename to vpx_dsp/mips/avg_msa.c
index 611adb1..52a24ed 100644
--- a/vp9/encoder/mips/msa/vp9_avg_msa.c
+++ b/vpx_dsp/mips/avg_msa.c
@@ -8,10 +8,10 @@
* be found in the AUTHORS file in the root of the source tree.
*/
-#include "./vp9_rtcd.h"
+#include "./vpx_dsp_rtcd.h"
#include "vpx_dsp/mips/macros_msa.h"
-uint32_t vp9_avg_8x8_msa(const uint8_t *src, int32_t src_stride) {
+uint32_t vpx_avg_8x8_msa(const uint8_t *src, int32_t src_stride) {
uint32_t sum_out;
v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
v8u16 sum0, sum1, sum2, sum3, sum4, sum5, sum6, sum7;
@@ -33,7 +33,7 @@
return sum_out;
}
-uint32_t vp9_avg_4x4_msa(const uint8_t *src, int32_t src_stride) {
+uint32_t vpx_avg_4x4_msa(const uint8_t *src, int32_t src_stride) {
uint32_t sum_out;
uint32_t src0, src1, src2, src3;
v16u8 vec = { 0 };
diff --git a/vpx_dsp/vpx_dsp.mk b/vpx_dsp/vpx_dsp.mk
index 9620eaa..e394688 100644
--- a/vpx_dsp/vpx_dsp.mk
+++ b/vpx_dsp/vpx_dsp.mk
@@ -252,6 +252,18 @@
DSP_SRCS-$(HAVE_AVX) += x86/quantize_avx_x86_64.asm
endif
endif
+
+# avg
+DSP_SRCS-yes += avg.c
+DSP_SRCS-$(HAVE_SSE2) += x86/avg_intrin_sse2.c
+DSP_SRCS-$(HAVE_NEON) += arm/avg_neon.c
+DSP_SRCS-$(HAVE_MSA) += mips/avg_msa.c
+ifeq ($(ARCH_X86_64),yes)
+ifeq ($(CONFIG_USE_X86INC),yes)
+DSP_SRCS-$(HAVE_SSSE3) += x86/avg_ssse3_x86_64.asm
+endif
+endif
+
endif # CONFIG_VP9_ENCODER || CONFIG_VP10_ENCODER
ifeq ($(CONFIG_ENCODERS),yes)
diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl
index 5ce7134..a2a0674 100644
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -226,7 +226,7 @@
specialize qw/vpx_d63e_predictor_32x32/;
add_proto qw/void vpx_h_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_h_predictor_32x32 neon msa/, "$ssse3_x86inc";
+specialize qw/vpx_h_predictor_32x32 neon msa/, "$sse2_x86inc";
add_proto qw/void vpx_d117_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vpx_d117_predictor_32x32/;
@@ -241,7 +241,7 @@
specialize qw/vpx_v_predictor_32x32 neon msa/, "$sse2_x86inc";
add_proto qw/void vpx_tm_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_tm_predictor_32x32 neon msa/, "$sse2_x86_64_x86inc";
+specialize qw/vpx_tm_predictor_32x32 neon msa/, "$sse2_x86inc";
add_proto qw/void vpx_dc_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vpx_dc_predictor_32x32 msa neon/, "$sse2_x86inc";
@@ -288,13 +288,13 @@
specialize qw/vpx_highbd_d153_predictor_4x4/;
add_proto qw/void vpx_highbd_v_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vpx_highbd_v_predictor_4x4/, "$sse_x86inc";
+ specialize qw/vpx_highbd_v_predictor_4x4/, "$sse2_x86inc";
add_proto qw/void vpx_highbd_tm_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vpx_highbd_tm_predictor_4x4/, "$sse_x86inc";
+ specialize qw/vpx_highbd_tm_predictor_4x4/, "$sse2_x86inc";
add_proto qw/void vpx_highbd_dc_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vpx_highbd_dc_predictor_4x4/, "$sse_x86inc";
+ specialize qw/vpx_highbd_dc_predictor_4x4/, "$sse2_x86inc";
add_proto qw/void vpx_highbd_dc_top_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
specialize qw/vpx_highbd_dc_top_predictor_4x4/;
@@ -990,14 +990,43 @@
specialize qw/vpx_sad8x4 msa/, "$sse2_x86inc";
add_proto qw/unsigned int vpx_sad4x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
-specialize qw/vpx_sad4x8 msa/, "$sse_x86inc";
+specialize qw/vpx_sad4x8 msa/, "$sse2_x86inc";
add_proto qw/unsigned int vpx_sad4x4/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
-specialize qw/vpx_sad4x4 mmx neon msa/, "$sse_x86inc";
+specialize qw/vpx_sad4x4 mmx neon msa/, "$sse2_x86inc";
#
# Avg
#
+if ((vpx_config("CONFIG_VP9_ENCODER") eq "yes") || (vpx_config("CONFIG_VP10_ENCODER") eq "yes")) {
+ add_proto qw/unsigned int vpx_avg_8x8/, "const uint8_t *, int p";
+ specialize qw/vpx_avg_8x8 sse2 neon msa/;
+
+ add_proto qw/unsigned int vpx_avg_4x4/, "const uint8_t *, int p";
+ specialize qw/vpx_avg_4x4 sse2 neon msa/;
+
+ add_proto qw/void vpx_minmax_8x8/, "const uint8_t *s, int p, const uint8_t *d, int dp, int *min, int *max";
+ specialize qw/vpx_minmax_8x8 sse2/;
+
+ add_proto qw/void vpx_hadamard_8x8/, "int16_t const *src_diff, int src_stride, int16_t *coeff";
+ specialize qw/vpx_hadamard_8x8 sse2/, "$ssse3_x86_64_x86inc";
+
+ add_proto qw/void vpx_hadamard_16x16/, "int16_t const *src_diff, int src_stride, int16_t *coeff";
+ specialize qw/vpx_hadamard_16x16 sse2/;
+
+ add_proto qw/int vpx_satd/, "const int16_t *coeff, int length";
+ specialize qw/vpx_satd sse2 neon/;
+
+ add_proto qw/void vpx_int_pro_row/, "int16_t *hbuf, uint8_t const *ref, const int ref_stride, const int height";
+ specialize qw/vpx_int_pro_row sse2 neon/;
+
+ add_proto qw/int16_t vpx_int_pro_col/, "uint8_t const *ref, const int width";
+ specialize qw/vpx_int_pro_col sse2 neon/;
+
+ add_proto qw/int vpx_vector_var/, "int16_t const *ref, int16_t const *src, const int bwl";
+ specialize qw/vpx_vector_var neon sse2/;
+} # CONFIG_VP9_ENCODER || CONFIG_VP10_ENCODER
+
add_proto qw/unsigned int vpx_sad64x64_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
specialize qw/vpx_sad64x64_avg avx2 msa/, "$sse2_x86inc";
@@ -1032,10 +1061,10 @@
specialize qw/vpx_sad8x4_avg msa/, "$sse2_x86inc";
add_proto qw/unsigned int vpx_sad4x8_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
-specialize qw/vpx_sad4x8_avg msa/, "$sse_x86inc";
+specialize qw/vpx_sad4x8_avg msa/, "$sse2_x86inc";
add_proto qw/unsigned int vpx_sad4x4_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
-specialize qw/vpx_sad4x4_avg msa/, "$sse_x86inc";
+specialize qw/vpx_sad4x4_avg msa/, "$sse2_x86inc";
#
# Multi-block SAD, comparing a reference to N blocks 1 pixel apart horizontally
@@ -1195,6 +1224,13 @@
#
# Avg
#
+ add_proto qw/unsigned int vpx_highbd_avg_8x8/, "const uint8_t *, int p";
+ specialize qw/vpx_highbd_avg_8x8/;
+ add_proto qw/unsigned int vpx_highbd_avg_4x4/, "const uint8_t *, int p";
+ specialize qw/vpx_highbd_avg_4x4/;
+ add_proto qw/void vpx_highbd_minmax_8x8/, "const uint8_t *s, int p, const uint8_t *d, int dp, int *min, int *max";
+ specialize qw/vpx_highbd_minmax_8x8/;
+
add_proto qw/unsigned int vpx_highbd_sad64x64_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
specialize qw/vpx_highbd_sad64x64_avg/, "$sse2_x86inc";
diff --git a/vp9/encoder/x86/vp9_avg_intrin_sse2.c b/vpx_dsp/x86/avg_intrin_sse2.c
similarity index 95%
rename from vp9/encoder/x86/vp9_avg_intrin_sse2.c
rename to vpx_dsp/x86/avg_intrin_sse2.c
index 4414871..f9af6cf 100644
--- a/vp9/encoder/x86/vp9_avg_intrin_sse2.c
+++ b/vpx_dsp/x86/avg_intrin_sse2.c
@@ -10,10 +10,10 @@
#include <emmintrin.h>
-#include "./vp9_rtcd.h"
+#include "./vpx_dsp_rtcd.h"
#include "vpx_ports/mem.h"
-void vp9_minmax_8x8_sse2(const uint8_t *s, int p, const uint8_t *d, int dp,
+void vpx_minmax_8x8_sse2(const uint8_t *s, int p, const uint8_t *d, int dp,
int *min, int *max) {
__m128i u0, s0, d0, diff, maxabsdiff, minabsdiff, negdiff, absdiff0, absdiff;
u0 = _mm_setzero_si128();
@@ -91,7 +91,7 @@
*min = _mm_extract_epi16(minabsdiff, 0);
}
-unsigned int vp9_avg_8x8_sse2(const uint8_t *s, int p) {
+unsigned int vpx_avg_8x8_sse2(const uint8_t *s, int p) {
__m128i s0, s1, u0;
unsigned int avg = 0;
u0 = _mm_setzero_si128();
@@ -118,7 +118,7 @@
return (avg + 32) >> 6;
}
-unsigned int vp9_avg_4x4_sse2(const uint8_t *s, int p) {
+unsigned int vpx_avg_4x4_sse2(const uint8_t *s, int p) {
__m128i s0, s1, u0;
unsigned int avg = 0;
u0 = _mm_setzero_si128();
@@ -212,7 +212,7 @@
}
}
-void vp9_hadamard_8x8_sse2(int16_t const *src_diff, int src_stride,
+void vpx_hadamard_8x8_sse2(int16_t const *src_diff, int src_stride,
int16_t *coeff) {
__m128i src[8];
src[0] = _mm_load_si128((const __m128i *)src_diff);
@@ -244,13 +244,13 @@
_mm_store_si128((__m128i *)coeff, src[7]);
}
-void vp9_hadamard_16x16_sse2(int16_t const *src_diff, int src_stride,
+void vpx_hadamard_16x16_sse2(int16_t const *src_diff, int src_stride,
int16_t *coeff) {
int idx;
for (idx = 0; idx < 4; ++idx) {
int16_t const *src_ptr = src_diff + (idx >> 1) * 8 * src_stride
+ (idx & 0x01) * 8;
- vp9_hadamard_8x8_sse2(src_ptr, src_stride, coeff + idx * 64);
+ vpx_hadamard_8x8_sse2(src_ptr, src_stride, coeff + idx * 64);
}
for (idx = 0; idx < 64; idx += 8) {
@@ -283,7 +283,7 @@
}
}
-int vp9_satd_sse2(const int16_t *coeff, int length) {
+int vpx_satd_sse2(const int16_t *coeff, int length) {
int i;
const __m128i zero = _mm_setzero_si128();
__m128i accum = zero;
@@ -309,7 +309,7 @@
return _mm_cvtsi128_si32(accum);
}
-void vp9_int_pro_row_sse2(int16_t *hbuf, uint8_t const*ref,
+void vpx_int_pro_row_sse2(int16_t *hbuf, uint8_t const*ref,
const int ref_stride, const int height) {
int idx;
__m128i zero = _mm_setzero_si128();
@@ -358,7 +358,7 @@
_mm_storeu_si128((__m128i *)hbuf, s1);
}
-int16_t vp9_int_pro_col_sse2(uint8_t const *ref, const int width) {
+int16_t vpx_int_pro_col_sse2(uint8_t const *ref, const int width) {
__m128i zero = _mm_setzero_si128();
__m128i src_line = _mm_load_si128((const __m128i *)ref);
__m128i s0 = _mm_sad_epu8(src_line, zero);
@@ -378,7 +378,7 @@
return _mm_extract_epi16(s0, 0);
}
-int vp9_vector_var_sse2(int16_t const *ref, int16_t const *src,
+int vpx_vector_var_sse2(int16_t const *ref, int16_t const *src,
const int bwl) {
int idx;
int width = 4 << bwl;
diff --git a/vp9/encoder/x86/vp9_dct_ssse3_x86_64.asm b/vpx_dsp/x86/avg_ssse3_x86_64.asm
similarity index 96%
rename from vp9/encoder/x86/vp9_dct_ssse3_x86_64.asm
rename to vpx_dsp/x86/avg_ssse3_x86_64.asm
index 74c52df..26412e8 100644
--- a/vp9/encoder/x86/vp9_dct_ssse3_x86_64.asm
+++ b/vpx_dsp/x86/avg_ssse3_x86_64.asm
@@ -8,11 +8,11 @@
; be found in the AUTHORS file in the root of the source tree.
;
-%define private_prefix vp9
+%define private_prefix vpx
%include "third_party/x86inc/x86inc.asm"
-; This file provides SSSE3 version of the forward transformation. Part
+; This file provides SSSE3 version of the hadamard transformation. Part
; of the macro definitions are originally derived from the ffmpeg project.
; The current version applies to x86 64-bit only.
diff --git a/vpx_dsp/x86/highbd_intrapred_sse2.asm b/vpx_dsp/x86/highbd_intrapred_sse2.asm
index b12d29c..233958a 100644
--- a/vpx_dsp/x86/highbd_intrapred_sse2.asm
+++ b/vpx_dsp/x86/highbd_intrapred_sse2.asm
@@ -17,24 +17,20 @@
pw_32: times 4 dd 32
SECTION .text
-INIT_MMX sse
+INIT_XMM sse2
cglobal highbd_dc_predictor_4x4, 4, 5, 4, dst, stride, above, left, goffset
GET_GOT goffsetq
movq m0, [aboveq]
movq m2, [leftq]
- DEFINE_ARGS dst, stride, one
- mov oned, 0x0001
- pxor m1, m1
- movd m3, oned
- pshufw m3, m3, 0x0
paddw m0, m2
- pmaddwd m0, m3
- packssdw m0, m1
- pmaddwd m0, m3
+ pshuflw m1, m0, 0xe
+ paddw m0, m1
+ pshuflw m1, m0, 0x1
+ paddw m0, m1
paddw m0, [GLOBAL(pw_4)]
psraw m0, 3
- pshufw m0, m0, 0x0
+ pshuflw m0, m0, 0x0
movq [dstq ], m0
movq [dstq+strideq*2], m0
lea dstq, [dstq+strideq*4]
@@ -183,7 +179,7 @@
REP_RET
%endif
-INIT_MMX sse
+INIT_XMM sse2
cglobal highbd_v_predictor_4x4, 3, 3, 1, dst, stride, above
movq m0, [aboveq]
movq [dstq ], m0
@@ -261,43 +257,44 @@
jnz .loop
REP_RET
-INIT_MMX sse
-cglobal highbd_tm_predictor_4x4, 5, 6, 5, dst, stride, above, left, bps, one
+INIT_XMM sse2
+cglobal highbd_tm_predictor_4x4, 5, 5, 6, dst, stride, above, left, bps
movd m1, [aboveq-2]
movq m0, [aboveq]
- pshufw m1, m1, 0x0
+ pshuflw m1, m1, 0x0
+ movlhps m0, m0 ; t1 t2 t3 t4 t1 t2 t3 t4
+ movlhps m1, m1 ; tl tl tl tl tl tl tl tl
; Get the values to compute the maximum value at this bit depth
- mov oned, 1
- movd m3, oned
+ pcmpeqw m3, m3
movd m4, bpsd
- pshufw m3, m3, 0x0
- DEFINE_ARGS dst, stride, line, left
- mov lineq, -2
- mova m2, m3
+ psubw m0, m1 ; t1-tl t2-tl t3-tl t4-tl
psllw m3, m4
- add leftq, 8
- psubw m3, m2 ; max possible value
- pxor m4, m4 ; min possible value
- psubw m0, m1
-.loop:
- movq m1, [leftq+lineq*4]
- movq m2, [leftq+lineq*4+2]
- pshufw m1, m1, 0x0
- pshufw m2, m2, 0x0
- paddw m1, m0
+ pcmpeqw m2, m2
+ pxor m4, m4 ; min possible value
+ pxor m3, m2 ; max possible value
+ mova m1, [leftq]
+ pshuflw m2, m1, 0x0
+ pshuflw m5, m1, 0x55
+ movlhps m2, m5 ; l1 l1 l1 l1 l2 l2 l2 l2
paddw m2, m0
;Clamp to the bit-depth
- pminsw m1, m3
pminsw m2, m3
- pmaxsw m1, m4
pmaxsw m2, m4
;Store the values
- movq [dstq ], m1
- movq [dstq+strideq*2], m2
+ movq [dstq ], m2
+ movhpd [dstq+strideq*2], m2
lea dstq, [dstq+strideq*4]
- inc lineq
- jnz .loop
- REP_RET
+ pshuflw m2, m1, 0xaa
+ pshuflw m5, m1, 0xff
+ movlhps m2, m5
+ paddw m2, m0
+ ;Clamp to the bit-depth
+ pminsw m2, m3
+ pmaxsw m2, m4
+ ;Store the values
+ movq [dstq ], m2
+ movhpd [dstq+strideq*2], m2
+ RET
INIT_XMM sse2
cglobal highbd_tm_predictor_8x8, 5, 6, 5, dst, stride, above, left, bps, one
diff --git a/vpx_dsp/x86/highbd_subpel_variance_impl_sse2.asm b/vpx_dsp/x86/highbd_subpel_variance_impl_sse2.asm
index 5f9c963..30ee81b 100644
--- a/vpx_dsp/x86/highbd_subpel_variance_impl_sse2.asm
+++ b/vpx_dsp/x86/highbd_subpel_variance_impl_sse2.asm
@@ -79,20 +79,13 @@
%macro INC_SRC_BY_SRC_STRIDE 0
%if ARCH_X86=1 && CONFIG_PIC=1
- lea srcq, [srcq + src_stridemp*2]
+ add srcq, src_stridemp
+ add srcq, src_stridemp
%else
lea srcq, [srcq + src_strideq*2]
%endif
%endmacro
-%macro INC_SRC_BY_SRC_2STRIDE 0
-%if ARCH_X86=1 && CONFIG_PIC=1
- lea srcq, [srcq + src_stridemp*4]
-%else
- lea srcq, [srcq + src_strideq*4]
-%endif
-%endmacro
-
%macro SUBPEL_VARIANCE 1-2 0 ; W
%define bilin_filter_m bilin_filter_m_sse2
%define filter_idx_shift 5
@@ -123,7 +116,10 @@
%define sec_str sec_stridemp
; Store bilin_filter and pw_8 location in stack
- GET_GOT eax
+ %if GET_GOT_DEFINED == 1
+ GET_GOT eax
+ add esp, 4 ; restore esp
+ %endif
lea ecx, [GLOBAL(bilin_filter_m)]
mov g_bilin_filterm, ecx
@@ -131,7 +127,6 @@
lea ecx, [GLOBAL(pw_8)]
mov g_pw_8m, ecx
- RESTORE_GOT ; restore esp
LOAD_IF_USED 0, 1 ; load eax, ecx back
%else
cglobal highbd_sub_pixel_variance%1xh, 7, 7, 13, src, src_stride, \
@@ -140,7 +135,10 @@
%define block_height heightd
; Store bilin_filter and pw_8 location in stack
- GET_GOT eax
+ %if GET_GOT_DEFINED == 1
+ GET_GOT eax
+ add esp, 4 ; restore esp
+ %endif
lea ecx, [GLOBAL(bilin_filter_m)]
mov g_bilin_filterm, ecx
@@ -148,7 +146,6 @@
lea ecx, [GLOBAL(pw_8)]
mov g_pw_8m, ecx
- RESTORE_GOT ; restore esp
LOAD_IF_USED 0, 1 ; load eax, ecx back
%endif
%else
@@ -980,8 +977,9 @@
.x_other_y_other_loop:
movu m2, [srcq]
movu m4, [srcq+2]
- movu m3, [srcq+src_strideq*2]
- movu m5, [srcq+src_strideq*2+2]
+ INC_SRC_BY_SRC_STRIDE
+ movu m3, [srcq]
+ movu m5, [srcq+2]
pmullw m2, filter_x_a
pmullw m4, filter_x_b
paddw m2, filter_rnd
@@ -1014,7 +1012,7 @@
SUM_SSE m0, m2, m4, m3, m6, m7
mova m0, m5
- INC_SRC_BY_SRC_2STRIDE
+ INC_SRC_BY_SRC_STRIDE
lea dstq, [dstq + dst_strideq * 4]
%if %2 == 1 ; avg
add secq, sec_str
diff --git a/vpx_dsp/x86/highbd_variance_sse2.c b/vpx_dsp/x86/highbd_variance_sse2.c
index b45331c..81ec5db 100644
--- a/vpx_dsp/x86/highbd_variance_sse2.c
+++ b/vpx_dsp/x86/highbd_variance_sse2.c
@@ -243,13 +243,18 @@
}
#if CONFIG_USE_X86INC
+// The 2 unused parameters are place holders for PIC enabled build.
+// These definitions are for functions defined in
+// highbd_subpel_variance_impl_sse2.asm
#define DECL(w, opt) \
int vpx_highbd_sub_pixel_variance##w##xh_##opt(const uint16_t *src, \
ptrdiff_t src_stride, \
int x_offset, int y_offset, \
const uint16_t *dst, \
ptrdiff_t dst_stride, \
- int height, unsigned int *sse);
+ int height, \
+ unsigned int *sse, \
+ void *unused0, void *unused);
#define DECLS(opt1, opt2) \
DECL(8, opt1); \
DECL(16, opt1)
@@ -274,7 +279,7 @@
int se = vpx_highbd_sub_pixel_variance##wf##xh_##opt(src, src_stride, \
x_offset, y_offset, \
dst, dst_stride, h, \
- &sse); \
+ &sse, NULL, NULL); \
if (w > wf) { \
unsigned int sse2; \
int se2 = vpx_highbd_sub_pixel_variance##wf##xh_##opt(src + 16, \
@@ -282,19 +287,20 @@
x_offset, y_offset, \
dst + 16, \
dst_stride, \
- h, &sse2); \
+ h, &sse2, \
+ NULL, NULL); \
se += se2; \
sse += sse2; \
if (w > wf * 2) { \
se2 = vpx_highbd_sub_pixel_variance##wf##xh_##opt(src + 32, src_stride, \
x_offset, y_offset, \
dst + 32, dst_stride, \
- h, &sse2); \
+ h, &sse2, NULL, NULL); \
se += se2; \
sse += sse2; \
se2 = vpx_highbd_sub_pixel_variance##wf##xh_##opt( \
src + 48, src_stride, x_offset, y_offset, \
- dst + 48, dst_stride, h, &sse2); \
+ dst + 48, dst_stride, h, &sse2, NULL, NULL); \
se += se2; \
sse += sse2; \
} \
@@ -312,7 +318,7 @@
int se = vpx_highbd_sub_pixel_variance##wf##xh_##opt(src, src_stride, \
x_offset, y_offset, \
dst, dst_stride, \
- h, &sse); \
+ h, &sse, NULL, NULL); \
if (w > wf) { \
uint32_t sse2; \
int se2 = vpx_highbd_sub_pixel_variance##wf##xh_##opt(src + 16, \
@@ -320,20 +326,21 @@
x_offset, y_offset, \
dst + 16, \
dst_stride, \
- h, &sse2); \
+ h, &sse2, \
+ NULL, NULL); \
se += se2; \
sse += sse2; \
if (w > wf * 2) { \
se2 = vpx_highbd_sub_pixel_variance##wf##xh_##opt(src + 32, src_stride, \
x_offset, y_offset, \
dst + 32, dst_stride, \
- h, &sse2); \
+ h, &sse2, NULL, NULL); \
se += se2; \
sse += sse2; \
se2 = vpx_highbd_sub_pixel_variance##wf##xh_##opt(src + 48, src_stride, \
x_offset, y_offset, \
dst + 48, dst_stride, \
- h, &sse2); \
+ h, &sse2, NULL, NULL); \
se += se2; \
sse += sse2; \
} \
@@ -359,27 +366,27 @@
int se2 = vpx_highbd_sub_pixel_variance##wf##xh_##opt( \
src + (start_row * src_stride), src_stride, \
x_offset, y_offset, dst + (start_row * dst_stride), \
- dst_stride, height, &sse2); \
+ dst_stride, height, &sse2, NULL, NULL); \
se += se2; \
long_sse += sse2; \
if (w > wf) { \
se2 = vpx_highbd_sub_pixel_variance##wf##xh_##opt( \
src + 16 + (start_row * src_stride), src_stride, \
x_offset, y_offset, dst + 16 + (start_row * dst_stride), \
- dst_stride, height, &sse2); \
+ dst_stride, height, &sse2, NULL, NULL); \
se += se2; \
long_sse += sse2; \
if (w > wf * 2) { \
se2 = vpx_highbd_sub_pixel_variance##wf##xh_##opt( \
src + 32 + (start_row * src_stride), src_stride, \
x_offset, y_offset, dst + 32 + (start_row * dst_stride), \
- dst_stride, height, &sse2); \
+ dst_stride, height, &sse2, NULL, NULL); \
se += se2; \
long_sse += sse2; \
se2 = vpx_highbd_sub_pixel_variance##wf##xh_##opt( \
src + 48 + (start_row * src_stride), src_stride, \
x_offset, y_offset, dst + 48 + (start_row * dst_stride), \
- dst_stride, height, &sse2); \
+ dst_stride, height, &sse2, NULL, NULL); \
se += se2; \
long_sse += sse2; \
}\
@@ -410,6 +417,7 @@
#undef FNS
#undef FN
+// The 2 unused parameters are place holders for PIC enabled build.
#define DECL(w, opt) \
int vpx_highbd_sub_pixel_avg_variance##w##xh_##opt(const uint16_t *src, \
ptrdiff_t src_stride, \
@@ -419,7 +427,8 @@
const uint16_t *sec, \
ptrdiff_t sec_stride, \
int height, \
- unsigned int *sse);
+ unsigned int *sse, \
+ void *unused0, void *unused);
#define DECLS(opt1) \
DECL(16, opt1) \
DECL(8, opt1)
@@ -439,23 +448,23 @@
uint16_t *sec = CONVERT_TO_SHORTPTR(sec8); \
int se = vpx_highbd_sub_pixel_avg_variance##wf##xh_##opt( \
src, src_stride, x_offset, \
- y_offset, dst, dst_stride, sec, w, h, &sse); \
+ y_offset, dst, dst_stride, sec, w, h, &sse, NULL, NULL); \
if (w > wf) { \
uint32_t sse2; \
int se2 = vpx_highbd_sub_pixel_avg_variance##wf##xh_##opt( \
src + 16, src_stride, x_offset, y_offset, \
- dst + 16, dst_stride, sec + 16, w, h, &sse2); \
+ dst + 16, dst_stride, sec + 16, w, h, &sse2, NULL, NULL); \
se += se2; \
sse += sse2; \
if (w > wf * 2) { \
se2 = vpx_highbd_sub_pixel_avg_variance##wf##xh_##opt( \
src + 32, src_stride, x_offset, y_offset, \
- dst + 32, dst_stride, sec + 32, w, h, &sse2); \
+ dst + 32, dst_stride, sec + 32, w, h, &sse2, NULL, NULL); \
se += se2; \
sse += sse2; \
se2 = vpx_highbd_sub_pixel_avg_variance##wf##xh_##opt( \
src + 48, src_stride, x_offset, y_offset, \
- dst + 48, dst_stride, sec + 48, w, h, &sse2); \
+ dst + 48, dst_stride, sec + 48, w, h, &sse2, NULL, NULL); \
se += se2; \
sse += sse2; \
} \
@@ -475,14 +484,15 @@
int se = vpx_highbd_sub_pixel_avg_variance##wf##xh_##opt( \
src, src_stride, x_offset, \
y_offset, dst, dst_stride, \
- sec, w, h, &sse); \
+ sec, w, h, &sse, NULL, NULL); \
if (w > wf) { \
uint32_t sse2; \
int se2 = vpx_highbd_sub_pixel_avg_variance##wf##xh_##opt( \
src + 16, src_stride, \
x_offset, y_offset, \
dst + 16, dst_stride, \
- sec + 16, w, h, &sse2); \
+ sec + 16, w, h, &sse2, \
+ NULL, NULL); \
se += se2; \
sse += sse2; \
if (w > wf * 2) { \
@@ -490,14 +500,16 @@
src + 32, src_stride, \
x_offset, y_offset, \
dst + 32, dst_stride, \
- sec + 32, w, h, &sse2); \
+ sec + 32, w, h, &sse2, \
+ NULL, NULL); \
se += se2; \
sse += sse2; \
se2 = vpx_highbd_sub_pixel_avg_variance##wf##xh_##opt( \
src + 48, src_stride, \
x_offset, y_offset, \
dst + 48, dst_stride, \
- sec + 48, w, h, &sse2); \
+ sec + 48, w, h, &sse2, \
+ NULL, NULL); \
se += se2; \
sse += sse2; \
} \
@@ -525,7 +537,7 @@
int se2 = vpx_highbd_sub_pixel_avg_variance##wf##xh_##opt( \
src + (start_row * src_stride), src_stride, x_offset, \
y_offset, dst + (start_row * dst_stride), dst_stride, \
- sec + (start_row * w), w, height, &sse2); \
+ sec + (start_row * w), w, height, &sse2, NULL, NULL); \
se += se2; \
long_sse += sse2; \
if (w > wf) { \
@@ -533,7 +545,7 @@
src + 16 + (start_row * src_stride), src_stride, \
x_offset, y_offset, \
dst + 16 + (start_row * dst_stride), dst_stride, \
- sec + 16 + (start_row * w), w, height, &sse2); \
+ sec + 16 + (start_row * w), w, height, &sse2, NULL, NULL); \
se += se2; \
long_sse += sse2; \
if (w > wf * 2) { \
@@ -541,14 +553,14 @@
src + 32 + (start_row * src_stride), src_stride, \
x_offset, y_offset, \
dst + 32 + (start_row * dst_stride), dst_stride, \
- sec + 32 + (start_row * w), w, height, &sse2); \
+ sec + 32 + (start_row * w), w, height, &sse2, NULL, NULL); \
se += se2; \
long_sse += sse2; \
se2 = vpx_highbd_sub_pixel_avg_variance##wf##xh_##opt( \
src + 48 + (start_row * src_stride), src_stride, \
x_offset, y_offset, \
dst + 48 + (start_row * dst_stride), dst_stride, \
- sec + 48 + (start_row * w), w, height, &sse2); \
+ sec + 48 + (start_row * w), w, height, &sse2, NULL, NULL); \
se += se2; \
long_sse += sse2; \
} \
diff --git a/vpx_dsp/x86/intrapred_sse2.asm b/vpx_dsp/x86/intrapred_sse2.asm
index edbf05e..c24d536 100644
--- a/vpx_dsp/x86/intrapred_sse2.asm
+++ b/vpx_dsp/x86/intrapred_sse2.asm
@@ -47,9 +47,9 @@
INIT_XMM sse2
cglobal dc_left_predictor_4x4, 2, 5, 2, dst, stride, above, left, goffset
+ movifnidn leftq, leftmp
GET_GOT goffsetq
- movifnidn leftq, leftmp
pxor m1, m1
movd m0, [leftq]
psadbw m0, m1
@@ -143,9 +143,9 @@
INIT_XMM sse2
cglobal dc_left_predictor_8x8, 2, 5, 2, dst, stride, above, left, goffset
+ movifnidn leftq, leftmp
GET_GOT goffsetq
- movifnidn leftq, leftmp
pxor m1, m1
movq m0, [leftq]
DEFINE_ARGS dst, stride, stride3
@@ -239,14 +239,11 @@
GET_GOT goffsetq
pxor m1, m1
- pxor m2, m2
mova m0, [aboveq]
DEFINE_ARGS dst, stride, stride3, lines4
lea stride3q, [strideq*3]
mov lines4d, 4
psadbw m0, m1
- psadbw m2, m1
- paddw m0, m2
movhlps m2, m0
paddw m0, m2
paddw m0, [GLOBAL(pw2_16)]
@@ -271,14 +268,11 @@
GET_GOT goffsetq
pxor m1, m1
- pxor m2, m2
mova m0, [leftq]
DEFINE_ARGS dst, stride, stride3, lines4
lea stride3q, [strideq*3]
mov lines4d, 4
psadbw m0, m1
- psadbw m2, m1
- paddw m0, m2
movhlps m2, m0
paddw m0, m2
paddw m0, [GLOBAL(pw2_16)]
@@ -582,6 +576,34 @@
REP_RET
INIT_XMM sse2
+cglobal h_predictor_32x32, 2, 5, 3, dst, stride, line, left
+ movifnidn leftq, leftmp
+ mov lineq, -8
+ DEFINE_ARGS dst, stride, line, left, stride3
+ lea stride3q, [strideq*3]
+.loop:
+ movd m0, [leftq]
+ punpcklbw m0, m0
+ punpcklbw m0, m0 ; l1 to l4 each repeated 4 times
+ pshufd m1, m0, 0x0 ; l1 repeated 16 times
+ pshufd m2, m0, 0x55 ; l2 repeated 16 times
+ mova [dstq ], m1
+ mova [dstq+16 ], m1
+ mova [dstq+strideq ], m2
+ mova [dstq+strideq+16 ], m2
+ pshufd m1, m0, 0xaa
+ pshufd m2, m0, 0xff
+ mova [dstq+strideq*2 ], m1
+ mova [dstq+strideq*2+16], m1
+ mova [dstq+stride3q ], m2
+ mova [dstq+stride3q+16 ], m2
+ inc lineq
+ lea leftq, [leftq+4 ]
+ lea dstq, [dstq+strideq*4]
+ jnz .loop
+ REP_RET
+
+INIT_XMM sse2
cglobal tm_predictor_4x4, 4, 4, 5, dst, stride, above, left
pxor m1, m1
movq m0, [aboveq-1]; [63:0] tl t1 t2 t3 t4 x x x
@@ -641,45 +663,45 @@
REP_RET
INIT_XMM sse2
-cglobal tm_predictor_16x16, 4, 4, 7, dst, stride, above, left
+cglobal tm_predictor_16x16, 4, 5, 8, dst, stride, above, left
pxor m1, m1
- movd m2, [aboveq-1]
- mova m0, [aboveq]
- punpcklbw m2, m1
+ mova m2, [aboveq-16];
+ mova m0, [aboveq] ; t1 t2 ... t16 [byte]
+ punpckhbw m2, m1 ; [127:112] tl [word]
punpckhbw m4, m0, m1
- punpcklbw m0, m1
- pshuflw m2, m2, 0x0
- DEFINE_ARGS dst, stride, line, left
+ punpcklbw m0, m1 ; m0:m4 t1 t2 ... t16 [word]
+ DEFINE_ARGS dst, stride, line, left, stride8
mov lineq, -8
- punpcklqdq m2, m2
- add leftq, 16
+ pshufhw m2, m2, 0xff
+ mova m3, [leftq] ; l1 l2 ... l16 [byte]
+ punpckhqdq m2, m2 ; tl repeated 8 times [word]
psubw m0, m2
- psubw m4, m2
+ psubw m4, m2 ; m0:m4 t1-tl t2-tl ... t16-tl [word]
+ punpckhbw m5, m3, m1
+ punpcklbw m3, m1 ; m3:m5 l1 l2 ... l16 [word]
+ lea stride8q, [strideq*8]
.loop:
- movd m2, [leftq+lineq*2]
- movd m3, [leftq+lineq*2+1]
- punpcklbw m2, m1
- punpcklbw m3, m1
- pshuflw m2, m2, 0x0
- pshuflw m3, m3, 0x0
- punpcklqdq m2, m2
- punpcklqdq m3, m3
- paddw m5, m2, m0
- paddw m6, m3, m0
- paddw m2, m4
- paddw m3, m4
- packuswb m5, m2
- packuswb m6, m3
- mova [dstq ], m5
- mova [dstq+strideq], m6
- lea dstq, [dstq+strideq*2]
+ pshuflw m6, m3, 0x0
+ pshuflw m7, m5, 0x0
+ punpcklqdq m6, m6 ; l1 repeated 8 times [word]
+ punpcklqdq m7, m7 ; l8 repeated 8 times [word]
+ paddw m1, m6, m0
+ paddw m6, m4 ; m1:m6 ti-tl+l1 [i=1,15] [word]
+ psrldq m5, 2
+ packuswb m1, m6
+ mova [dstq ], m1
+ paddw m1, m7, m0
+ paddw m7, m4 ; m1:m7 ti-tl+l8 [i=1,15] [word]
+ psrldq m3, 2
+ packuswb m1, m7
+ mova [dstq+stride8q], m1
inc lineq
+ lea dstq, [dstq+strideq]
jnz .loop
REP_RET
-%if ARCH_X86_64
INIT_XMM sse2
-cglobal tm_predictor_32x32, 4, 4, 10, dst, stride, above, left
+cglobal tm_predictor_32x32, 4, 4, 8, dst, stride, above, left
pxor m1, m1
movd m2, [aboveq-1]
mova m0, [aboveq]
@@ -700,31 +722,29 @@
psubw m5, m2
.loop:
movd m2, [leftq+lineq*2]
- movd m6, [leftq+lineq*2+1]
+ pxor m1, m1
punpcklbw m2, m1
- punpcklbw m6, m1
+ pshuflw m7, m2, 0x55
pshuflw m2, m2, 0x0
- pshuflw m6, m6, 0x0
punpcklqdq m2, m2
- punpcklqdq m6, m6
- paddw m7, m2, m0
- paddw m8, m2, m3
- paddw m9, m2, m4
- paddw m2, m5
- packuswb m7, m8
- packuswb m9, m2
- paddw m2, m6, m0
- paddw m8, m6, m3
- mova [dstq ], m7
- paddw m7, m6, m4
- paddw m6, m5
- mova [dstq +16], m9
- packuswb m2, m8
- packuswb m7, m6
- mova [dstq+strideq ], m2
- mova [dstq+strideq+16], m7
+ punpcklqdq m7, m7
+ paddw m6, m2, m3
+ paddw m1, m2, m0
+ packuswb m1, m6
+ mova [dstq ], m1
+ paddw m6, m2, m5
+ paddw m1, m2, m4
+ packuswb m1, m6
+ mova [dstq+16 ], m1
+ paddw m6, m7, m3
+ paddw m1, m7, m0
+ packuswb m1, m6
+ mova [dstq+strideq ], m1
+ paddw m6, m7, m5
+ paddw m1, m7, m4
+ packuswb m1, m6
+ mova [dstq+strideq+16], m1
lea dstq, [dstq+strideq*2]
inc lineq
jnz .loop
REP_RET
-%endif
diff --git a/vpx_dsp/x86/intrapred_ssse3.asm b/vpx_dsp/x86/intrapred_ssse3.asm
index f1a193b..d061278 100644
--- a/vpx_dsp/x86/intrapred_ssse3.asm
+++ b/vpx_dsp/x86/intrapred_ssse3.asm
@@ -33,26 +33,6 @@
SECTION .text
-INIT_XMM ssse3
-cglobal h_predictor_32x32, 2, 4, 3, dst, stride, line, left
- movifnidn leftq, leftmp
- add leftq, 32
- mov lineq, -16
- pxor m0, m0
-.loop:
- movd m1, [leftq+lineq*2 ]
- movd m2, [leftq+lineq*2+1]
- pshufb m1, m0
- pshufb m2, m0
- mova [dstq ], m1
- mova [dstq +16], m1
- mova [dstq+strideq ], m2
- mova [dstq+strideq+16], m2
- lea dstq, [dstq+strideq*2]
- inc lineq
- jnz .loop
- REP_RET
-
INIT_MMX ssse3
cglobal d45_predictor_4x4, 3, 4, 4, dst, stride, above, goffset
GET_GOT goffsetq
diff --git a/vpx_dsp/x86/sad_sse2.asm b/vpx_dsp/x86/sad_sse2.asm
index 0defe1b..1ec906c 100644
--- a/vpx_dsp/x86/sad_sse2.asm
+++ b/vpx_dsp/x86/sad_sse2.asm
@@ -17,7 +17,7 @@
%if %3 == 5
cglobal sad%1x%2, 4, %3, 5, src, src_stride, ref, ref_stride, n_rows
%else ; %3 == 7
-cglobal sad%1x%2, 4, %3, 5, src, src_stride, ref, ref_stride, \
+cglobal sad%1x%2, 4, %3, 6, src, src_stride, ref, ref_stride, \
src_stride3, ref_stride3, n_rows
%endif ; %3 == 5/7
%else ; avg
@@ -25,7 +25,7 @@
cglobal sad%1x%2_avg, 5, 1 + %3, 5, src, src_stride, ref, ref_stride, \
second_pred, n_rows
%else ; %3 == 7
-cglobal sad%1x%2_avg, 5, ARCH_X86_64 + %3, 5, src, src_stride, \
+cglobal sad%1x%2_avg, 5, ARCH_X86_64 + %3, 6, src, src_stride, \
ref, ref_stride, \
second_pred, \
src_stride3, ref_stride3
@@ -222,8 +222,8 @@
SAD8XN 8, 1 ; sad8x8_avg_sse2
SAD8XN 4, 1 ; sad8x4_avg_sse2
-; unsigned int vpx_sad4x{4, 8}_sse(uint8_t *src, int src_stride,
-; uint8_t *ref, int ref_stride);
+; unsigned int vpx_sad4x{4, 8}_sse2(uint8_t *src, int src_stride,
+; uint8_t *ref, int ref_stride);
%macro SAD4XN 1-2 0
SAD_FN 4, %1, 7, %2
mov n_rowsd, %1/4
@@ -236,31 +236,32 @@
movd m4, [refq+ref_stride3q]
punpckldq m1, m2
punpckldq m3, m4
+ movlhps m1, m3
%if %2 == 1
pavgb m1, [second_predq+mmsize*0]
- pavgb m3, [second_predq+mmsize*1]
- lea second_predq, [second_predq+mmsize*2]
+ lea second_predq, [second_predq+mmsize*1]
%endif
movd m2, [srcq]
movd m5, [srcq+src_strideq]
movd m4, [srcq+src_strideq*2]
- movd m6, [srcq+src_stride3q]
+ movd m3, [srcq+src_stride3q]
punpckldq m2, m5
- punpckldq m4, m6
+ punpckldq m4, m3
+ movlhps m2, m4
psadbw m1, m2
- psadbw m3, m4
lea refq, [refq+ref_strideq*4]
paddd m0, m1
lea srcq, [srcq+src_strideq*4]
- paddd m0, m3
dec n_rowsd
jg .loop
+ movhlps m1, m0
+ paddd m0, m1
movd eax, m0
RET
%endmacro
-INIT_MMX sse
+INIT_XMM sse2
SAD4XN 8 ; sad4x8_sse
SAD4XN 4 ; sad4x4_sse
SAD4XN 8, 1 ; sad4x8_avg_sse
diff --git a/vpx_dsp/x86/subpel_variance_sse2.asm b/vpx_dsp/x86/subpel_variance_sse2.asm
index 1176a2f..c655e4b 100644
--- a/vpx_dsp/x86/subpel_variance_sse2.asm
+++ b/vpx_dsp/x86/subpel_variance_sse2.asm
@@ -139,7 +139,10 @@
%define sec_str sec_stridemp
;Store bilin_filter and pw_8 location in stack
- GET_GOT eax
+ %if GET_GOT_DEFINED == 1
+ GET_GOT eax
+ add esp, 4 ; restore esp
+ %endif
lea ecx, [GLOBAL(bilin_filter_m)]
mov g_bilin_filterm, ecx
@@ -147,7 +150,6 @@
lea ecx, [GLOBAL(pw_8)]
mov g_pw_8m, ecx
- RESTORE_GOT ; restore esp
LOAD_IF_USED 0, 1 ; load eax, ecx back
%else
cglobal sub_pixel_variance%1xh, 7, 7, 13, src, src_stride, x_offset, \
@@ -156,7 +158,10 @@
%define block_height heightd
;Store bilin_filter and pw_8 location in stack
- GET_GOT eax
+ %if GET_GOT_DEFINED == 1
+ GET_GOT eax
+ add esp, 4 ; restore esp
+ %endif
lea ecx, [GLOBAL(bilin_filter_m)]
mov g_bilin_filterm, ecx
@@ -164,7 +169,6 @@
lea ecx, [GLOBAL(pw_8)]
mov g_pw_8m, ecx
- RESTORE_GOT ; restore esp
LOAD_IF_USED 0, 1 ; load eax, ecx back
%endif
%else
diff --git a/vpx_ports/x86_abi_support.asm b/vpx_ports/x86_abi_support.asm
index c94b76a..708fa10 100644
--- a/vpx_ports/x86_abi_support.asm
+++ b/vpx_ports/x86_abi_support.asm
@@ -189,7 +189,6 @@
%if ABI_IS_32BIT
%if CONFIG_PIC=1
%ifidn __OUTPUT_FORMAT__,elf32
- %define GET_GOT_SAVE_ARG 1
%define WRT_PLT wrt ..plt
%macro GET_GOT 1
extern _GLOBAL_OFFSET_TABLE_
@@ -208,7 +207,6 @@
%define RESTORE_GOT pop %1
%endmacro
%elifidn __OUTPUT_FORMAT__,macho32
- %define GET_GOT_SAVE_ARG 1
%macro GET_GOT 1
push %1
call %%get_got
diff --git a/vpxenc.c b/vpxenc.c
index fd2acf6..afbaeac 100644
--- a/vpxenc.c
+++ b/vpxenc.c
@@ -2060,9 +2060,11 @@
#if !CONFIG_WEBM_IO
FOREACH_STREAM({
- stream->config.write_webm = 0;
- warn("vpxenc was compiled without WebM container support."
- "Producing IVF output");
+ if (stream->config.write_webm) {
+ stream->config.write_webm = 0;
+ warn("vpxenc was compiled without WebM container support."
+ "Producing IVF output");
+ }
});
#endif
