+++ /dev/null
-*-vs8/*.rules -crlf
-*-msvs/*.rules -crlf
+++ /dev/null
-x86*-win32-vs*
tgt_isa=x86_64
tgt_os=darwin16
;;
+ *darwin17*)
+ tgt_isa=x86_64
+ tgt_os=darwin17
+ ;;
x86_64*mingw32*)
tgt_os=win64
;;
add_cflags "-mmacosx-version-min=10.12"
add_ldflags "-mmacosx-version-min=10.12"
;;
+ *-darwin17-*)
+ add_cflags "-mmacosx-version-min=10.13"
+ add_ldflags "-mmacosx-version-min=10.13"
+ ;;
*-iphonesimulator-*)
add_cflags "-miphoneos-version-min=${IOS_VERSION_MIN}"
add_ldflags "-miphoneos-version-min=${IOS_VERSION_MIN}"
all_platforms="${all_platforms} x86-darwin14-gcc"
all_platforms="${all_platforms} x86-darwin15-gcc"
all_platforms="${all_platforms} x86-darwin16-gcc"
+all_platforms="${all_platforms} x86-darwin17-gcc"
all_platforms="${all_platforms} x86-iphonesimulator-gcc"
all_platforms="${all_platforms} x86-linux-gcc"
all_platforms="${all_platforms} x86-linux-icc"
all_platforms="${all_platforms} x86_64-darwin14-gcc"
all_platforms="${all_platforms} x86_64-darwin15-gcc"
all_platforms="${all_platforms} x86_64-darwin16-gcc"
+all_platforms="${all_platforms} x86_64-darwin17-gcc"
all_platforms="${all_platforms} x86_64-iphonesimulator-gcc"
all_platforms="${all_platforms} x86_64-linux-gcc"
all_platforms="${all_platforms} x86_64-linux-icc"
unistd_h
"
EXPERIMENT_LIST="
- spatial_svc
fp_mb_stats
emulate_hardware
"
endif
vpxenc.GUID = 548DEC74-7A15-4B2B-AFC3-AA102E7C25C1
vpxenc.DESCRIPTION = Full featured encoder
-ifeq ($(CONFIG_SPATIAL_SVC),yes)
- EXAMPLES-$(CONFIG_VP9_ENCODER) += vp9_spatial_svc_encoder.c
- vp9_spatial_svc_encoder.SRCS += args.c args.h
- vp9_spatial_svc_encoder.SRCS += ivfenc.c ivfenc.h
- vp9_spatial_svc_encoder.SRCS += tools_common.c tools_common.h
- vp9_spatial_svc_encoder.SRCS += video_common.h
- vp9_spatial_svc_encoder.SRCS += video_writer.h video_writer.c
- vp9_spatial_svc_encoder.SRCS += vpx_ports/msvc.h
- vp9_spatial_svc_encoder.SRCS += vpxstats.c vpxstats.h
- vp9_spatial_svc_encoder.GUID = 4A38598D-627D-4505-9C7B-D4020C84100D
- vp9_spatial_svc_encoder.DESCRIPTION = VP9 Spatial SVC Encoder
-endif
+
+EXAMPLES-$(CONFIG_VP9_ENCODER) += vp9_spatial_svc_encoder.c
+vp9_spatial_svc_encoder.SRCS += args.c args.h
+vp9_spatial_svc_encoder.SRCS += ivfenc.c ivfenc.h
+vp9_spatial_svc_encoder.SRCS += tools_common.c tools_common.h
+vp9_spatial_svc_encoder.SRCS += video_common.h
+vp9_spatial_svc_encoder.SRCS += video_writer.h video_writer.c
+vp9_spatial_svc_encoder.SRCS += vpx_ports/msvc.h
+vp9_spatial_svc_encoder.SRCS += vpxstats.c vpxstats.h
+vp9_spatial_svc_encoder.GUID = 4A38598D-627D-4505-9C7B-D4020C84100D
+vp9_spatial_svc_encoder.DESCRIPTION = VP9 Spatial SVC Encoder
ifneq ($(CONFIG_SHARED),yes)
EXAMPLES-$(CONFIG_VP9_ENCODER) += resize_util.c
// bypass/flexible mode. The pattern corresponds to the pattern
// VP9E_TEMPORAL_LAYERING_MODE_0101 (temporal_layering_mode == 2) used in
// non-flexible mode.
-void set_frame_flags_bypass_mode(int sl, int tl, int num_spatial_layers,
+void set_frame_flags_bypass_mode(int tl, int num_spatial_layers,
int is_key_frame,
vpx_svc_ref_frame_config_t *ref_frame_config) {
+ int sl;
for (sl = 0; sl < num_spatial_layers; ++sl) {
if (!tl) {
if (!sl) {
struct vpx_usec_timer timer;
int64_t cx_time = 0;
memset(&svc_ctx, 0, sizeof(svc_ctx));
- svc_ctx.log_print = 1;
exec_name = argv[0];
parse_command_line(argc, argv, &app_input, &svc_ctx, &enc_cfg);
die("Failed to initialize encoder\n");
#if OUTPUT_RC_STATS
+ rc.window_size = 15; // Silence a static analysis warning.
if (svc_ctx.output_rc_stat) {
set_rate_control_stats(&rc, &enc_cfg);
framerate = enc_cfg.g_timebase.den / enc_cfg.g_timebase.num;
vpx_codec_control(&codec, VP9E_SET_SVC_LAYER_ID, &layer_id);
// TODO(jianj): Fix the parameter passing for "is_key_frame" in
// set_frame_flags_bypass_model() for case of periodic key frames.
- set_frame_flags_bypass_mode(sl, layer_id.temporal_layer_id,
+ set_frame_flags_bypass_mode(layer_id.temporal_layer_id,
svc_ctx.spatial_layers, frame_cnt == 0,
&ref_frame_config);
vpx_codec_control(&codec, VP9E_SET_SVC_REF_FRAME_CONFIG,
vpx_usec_timer_mark(&timer);
cx_time += vpx_usec_timer_elapsed(&timer);
- printf("%s", vpx_svc_get_message(&svc_ctx));
fflush(stdout);
if (res != VPX_CODEC_OK) {
die_codec(&codec, "Failed to encode frame");
#if OUTPUT_RC_STATS
uint64_t sizes[8];
int count = 0;
+ vp9_zero(sizes);
#endif
vpx_video_writer_write_frame(writer, cx_pkt->data.frame.buf,
cx_pkt->data.frame.sz,
// window of size rc->window, shifted by rc->window / 2.
// Ignore first window segment, due to key frame.
if (frame_cnt > (unsigned int)rc.window_size) {
- tl = layer_id.temporal_layer_id;
for (sl = 0; sl < enc_cfg.ss_number_layers; ++sl) {
sum_bitrate += 0.001 * 8.0 * sizes[sl] * framerate;
}
// Second shifted window.
if (frame_cnt >
(unsigned int)(rc.window_size + rc.window_size / 2)) {
- tl = layer_id.temporal_layer_id;
for (sl = 0; sl < enc_cfg.ss_number_layers; ++sl) {
sum_bitrate2 += 0.001 * 8.0 * sizes[sl] * framerate;
}
1000000 * (double)frame_cnt / (double)cx_time);
vpx_img_free(&raw);
// display average size, psnr
- printf("%s", vpx_svc_dump_statistics(&svc_ctx));
+ vpx_svc_dump_statistics(&svc_ctx);
vpx_svc_release(&svc_ctx);
return EXIT_SUCCESS;
}
double sum_bitrate2 = 0.0;
double framerate = 30.0;
+ zero(rc.layer_target_bitrate);
+
exec_name = argv[0];
// Check usage and arguments.
if (argc < min_args) {
CODEC_EXPORTS-yes += $(addprefix $(VP9_PREFIX),$(VP9_CX_EXPORTS))
CODEC_SRCS-yes += $(VP9_PREFIX)vp9cx.mk vpx/vp8.h vpx/vp8cx.h
INSTALL-LIBS-yes += include/vpx/vp8.h include/vpx/vp8cx.h
- INSTALL-LIBS-$(CONFIG_SPATIAL_SVC) += include/vpx/svc_context.h
+ INSTALL-LIBS-yes += include/vpx/svc_context.h
INSTALL_MAPS += include/vpx/% $(SRC_PATH_BARE)/$(VP9_PREFIX)/%
CODEC_DOC_SRCS += vpx/vp8.h vpx/vp8cx.h
CODEC_DOC_SECTIONS += vp9 vp9_encoder
endif
CODEC_EXPORTS-yes += vpx/exports_com
CODEC_EXPORTS-$(CONFIG_ENCODERS) += vpx/exports_enc
-ifeq ($(CONFIG_SPATIAL_SVC),yes)
-CODEC_EXPORTS-$(CONFIG_ENCODERS) += vpx/exports_spatial_svc
-endif
+CODEC_EXPORTS-$(CONFIG_VP9_ENCODER) += vpx/exports_spatial_svc
CODEC_EXPORTS-$(CONFIG_DECODERS) += vpx/exports_dec
INSTALL-LIBS-yes += include/vpx/vpx_codec.h
--out=$@ $^
CLEAN-OBJS += vpx.def
+vpx.$(VCPROJ_SFX): VCPROJ_SRCS=$(filter-out $(addprefix %, $(ASM_INCLUDES)), $^)
+
vpx.$(VCPROJ_SFX): $(CODEC_SRCS) vpx.def
@echo " [CREATE] $@"
$(qexec)$(GEN_VCPROJ) \
--ver=$(CONFIG_VS_VERSION) \
--src-path-bare="$(SRC_PATH_BARE)" \
--out=$@ $(CFLAGS) \
- $(filter-out $(addprefix %, $(ASM_INCLUDES)), $^) \
+ $(filter $(SRC_PATH_BARE)/vp8/%.c, $(VCPROJ_SRCS)) \
+ $(filter $(SRC_PATH_BARE)/vp8/%.h, $(VCPROJ_SRCS)) \
+ $(filter $(SRC_PATH_BARE)/vp9/%.c, $(VCPROJ_SRCS)) \
+ $(filter $(SRC_PATH_BARE)/vp9/%.h, $(VCPROJ_SRCS)) \
+ $(filter $(SRC_PATH_BARE)/vpx/%, $(VCPROJ_SRCS)) \
+ $(filter $(SRC_PATH_BARE)/vpx_dsp/%, $(VCPROJ_SRCS)) \
+ $(filter-out $(addprefix $(SRC_PATH_BARE)/, \
+ vp8/%.c vp8/%.h vp9/%.c vp9/%.h vpx/% vpx_dsp/%), \
+ $(VCPROJ_SRCS)) \
--src-path-bare="$(SRC_PATH_BARE)" \
PROJECTS-yes += vpx.$(VCPROJ_SFX)
#define TEST_CLEAR_SYSTEM_STATE_H_
#include "./vpx_config.h"
-#if ARCH_X86 || ARCH_X86_64
-#include "vpx_ports/x86.h"
-#endif
+#include "vpx_ports/system_state.h"
namespace libvpx_test {
// Reset system to a known state. This function should be used for all non-API
// test cases.
-inline void ClearSystemState() {
-#if ARCH_X86 || ARCH_X86_64
- vpx_reset_mmx_state();
-#endif
-}
+inline void ClearSystemState() { vpx_clear_system_state(); }
} // namespace libvpx_test
#endif // TEST_CLEAR_SYSTEM_STATE_H_
dynamic_drop_layer_ = false;
change_bitrate_ = false;
last_pts_ref_ = 0;
+ middle_bitrate_ = 0;
+ top_bitrate_ = 0;
+ superframe_count_ = -1;
+ key_frame_spacing_ = 9999;
}
virtual void BeginPassHook(unsigned int /*pass*/) {}
encoder->Control(VP9E_SET_TUNE_CONTENT, tune_content_);
}
+ superframe_count_++;
+ temporal_layer_id_ = 0;
+ if (number_temporal_layers_ == 2)
+ temporal_layer_id_ = (superframe_count_ % 2 != 0);
+ else if (number_temporal_layers_ == 3) {
+ if (superframe_count_ % 2 != 0) temporal_layer_id_ = 2;
+ if (superframe_count_ > 1) {
+ if ((superframe_count_ - 2) % 4 == 0) temporal_layer_id_ = 1;
+ }
+ }
+
if (update_pattern_ && video->frame() >= 100) {
vpx_svc_layer_id_t layer_id;
if (video->frame() == 100) {
// Set layer id since the pattern changed.
layer_id.spatial_layer_id = 0;
layer_id.temporal_layer_id = (video->frame() % 2 != 0);
+ temporal_layer_id_ = layer_id.temporal_layer_id;
encoder->Control(VP9E_SET_SVC_LAYER_ID, &layer_id);
set_frame_flags_bypass_mode(layer_id.temporal_layer_id,
number_spatial_layers_, 0, &ref_frame_config);
}
if (dynamic_drop_layer_) {
- if (video->frame() == 100) {
- // Change layer bitrates to set top layer to 0. This will trigger skip
- // encoding/dropping of top spatial layer.
- cfg_.rc_target_bitrate -= cfg_.layer_target_bitrate[2];
+ if (video->frame() == 50) {
+ // Change layer bitrates to set top layers to 0. This will trigger skip
+ // encoding/dropping of top two spatial layers.
+ cfg_.rc_target_bitrate -=
+ (cfg_.layer_target_bitrate[1] + cfg_.layer_target_bitrate[2]);
+ middle_bitrate_ = cfg_.layer_target_bitrate[1];
+ top_bitrate_ = cfg_.layer_target_bitrate[2];
+ cfg_.layer_target_bitrate[1] = 0;
cfg_.layer_target_bitrate[2] = 0;
encoder->Config(&cfg_);
- } else if (video->frame() == 300) {
- // Change layer bitrate on top layer to non-zero to start encoding it
- // again.
- cfg_.layer_target_bitrate[2] = 500;
+ } else if (video->frame() == 100) {
+ // Change layer bitrate on second layer to non-zero to start
+ // encoding it again.
+ cfg_.layer_target_bitrate[1] = middle_bitrate_;
+ cfg_.rc_target_bitrate += cfg_.layer_target_bitrate[1];
+ encoder->Config(&cfg_);
+ } else if (video->frame() == 200) {
+ // Change layer bitrate on top layer to non-zero to start
+ // encoding it again.
+ cfg_.layer_target_bitrate[2] = top_bitrate_;
cfg_.rc_target_bitrate += cfg_.layer_target_bitrate[2];
encoder->Config(&cfg_);
}
duration_ = 0;
}
- virtual void PostEncodeFrameHook(::libvpx_test::Encoder *encoder) {
- vpx_svc_layer_id_t layer_id;
- encoder->Control(VP9E_GET_SVC_LAYER_ID, &layer_id);
- spatial_layer_id_ = layer_id.spatial_layer_id;
- temporal_layer_id_ = layer_id.temporal_layer_id;
- // Update buffer with per-layer target frame bandwidth, this is done
- // for every frame passed to the encoder (encoded or dropped).
- // For temporal layers, update the cumulative buffer level.
+ virtual void PostEncodeFrameHook() {
for (int sl = 0; sl < number_spatial_layers_; ++sl) {
for (int tl = temporal_layer_id_; tl < number_temporal_layers_; ++tl) {
const int layer = sl * number_temporal_layers_ + tl;
last_pts_ = pkt->data.frame.pts;
const bool key_frame =
(pkt->data.frame.flags & VPX_FRAME_IS_KEY) ? true : false;
+ if (key_frame) {
+ temporal_layer_id_ = 0;
+ superframe_count_ = 0;
+ }
parse_superframe_index(static_cast<const uint8_t *>(pkt->data.frame.buf),
pkt->data.frame.sz, sizes, &count);
- if (!dynamic_drop_layer_) ASSERT_EQ(count, number_spatial_layers_);
+ // Count may be less than number of spatial layers because of frame drops.
+ ASSERT_LE(count, number_spatial_layers_);
for (int sl = 0; sl < number_spatial_layers_; ++sl) {
sizes[sl] = sizes[sl] << 3;
// Update the total encoded bits per layer.
bits_in_buffer_model_[layer] -= static_cast<int64_t>(sizes[sl]);
// There should be no buffer underrun, except on the base
// temporal layer, since there may be key frames there.
- if (!key_frame && tl > 0) {
+ // Fo short key frame spacing, buffer can underrun on individual frames.
+ if (!key_frame && tl > 0 && key_frame_spacing_ < 100) {
ASSERT_GE(bits_in_buffer_model_[layer], 0)
<< "Buffer Underrun at frame " << pkt->data.frame.pts;
}
int update_pattern_;
bool change_bitrate_;
vpx_codec_pts_t last_pts_ref_;
+ int middle_bitrate_;
+ int top_bitrate_;
+ int superframe_count_;
+ int key_frame_spacing_;
};
// Check basic rate targeting for 1 pass CBR SVC: 2 spatial layers and 1
svc_params_.scaling_factor_den[0] = 288;
svc_params_.scaling_factor_num[1] = 288;
svc_params_.scaling_factor_den[1] = 288;
- cfg_.rc_dropframe_thresh = 0;
+ cfg_.rc_dropframe_thresh = 30;
cfg_.kf_max_dist = 9999;
number_spatial_layers_ = cfg_.ss_number_layers;
number_temporal_layers_ = cfg_.ts_number_layers;
svc_params_.scaling_factor_den[0] = 288;
svc_params_.scaling_factor_num[1] = 288;
svc_params_.scaling_factor_den[1] = 288;
- cfg_.rc_dropframe_thresh = 0;
+ cfg_.rc_dropframe_thresh = 30;
cfg_.kf_max_dist = 9999;
number_spatial_layers_ = cfg_.ss_number_layers;
number_temporal_layers_ = cfg_.ts_number_layers;
// 4 key neighboring key frame periods (so key frame will land on 0-2-1-2).
for (int j = 64; j <= 67; j++) {
cfg_.kf_max_dist = j;
+ key_frame_spacing_ = j;
ResetModel();
assign_layer_bitrates(&cfg_, &svc_params_, cfg_.ss_number_layers,
cfg_.ts_number_layers, cfg_.temporal_layering_mode,
svc_params_.scaling_factor_den[0] = 288;
svc_params_.scaling_factor_num[1] = 288;
svc_params_.scaling_factor_den[1] = 288;
- cfg_.rc_dropframe_thresh = 0;
+ cfg_.rc_dropframe_thresh = 30;
cfg_.kf_max_dist = 9999;
number_spatial_layers_ = cfg_.ss_number_layers;
number_temporal_layers_ = cfg_.ts_number_layers;
svc_params_.scaling_factor_den[1] = 288;
svc_params_.scaling_factor_num[2] = 288;
svc_params_.scaling_factor_den[2] = 288;
- cfg_.rc_dropframe_thresh = 0;
+ cfg_.rc_dropframe_thresh = 30;
cfg_.kf_max_dist = 9999;
number_spatial_layers_ = cfg_.ss_number_layers;
number_temporal_layers_ = cfg_.ts_number_layers;
svc_params_.scaling_factor_den[1] = 288;
svc_params_.scaling_factor_num[2] = 288;
svc_params_.scaling_factor_den[2] = 288;
- cfg_.rc_dropframe_thresh = 0;
+ cfg_.rc_dropframe_thresh = 30;
cfg_.kf_max_dist = 9999;
number_spatial_layers_ = cfg_.ss_number_layers;
number_temporal_layers_ = cfg_.ts_number_layers;
svc_params_.scaling_factor_den[1] = 288;
svc_params_.scaling_factor_num[2] = 288;
svc_params_.scaling_factor_den[2] = 288;
- cfg_.rc_dropframe_thresh = 0;
+ cfg_.rc_dropframe_thresh = 30;
cfg_.kf_max_dist = 9999;
number_spatial_layers_ = cfg_.ss_number_layers;
number_temporal_layers_ = cfg_.ts_number_layers;
}
// Check basic rate targeting for 1 pass CBR SVC with 3 spatial layers and on
-// the fly switching to 2 spatial layers and then back to 3. This switch is done
-// by setting top spatial layer bitrate to 0, and then back to non-zero, during
-// the sequence.
-TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SL_to_2SL_dynamic) {
+// the fly switching to 1 and then 2 and back to 3 spatial layers. This switch
+// is done by setting spatial layer bitrates to 0, and then back to non-zero,
+// during the sequence.
+TEST_P(DatarateOnePassCbrSvc, OnePassCbrSvc3SL_DisableEnableLayers) {
cfg_.rc_buf_initial_sz = 500;
cfg_.rc_buf_optimal_sz = 500;
cfg_.rc_buf_sz = 1000;
svc_params_.scaling_factor_den[1] = 288;
svc_params_.scaling_factor_num[2] = 288;
svc_params_.scaling_factor_den[2] = 288;
- cfg_.rc_dropframe_thresh = 0;
+ cfg_.rc_dropframe_thresh = 30;
cfg_.kf_max_dist = 9999;
number_spatial_layers_ = cfg_.ss_number_layers;
number_temporal_layers_ = cfg_.ts_number_layers;
// 4 key neighboring key frame periods (so key frame will land on 0-2-1-2).
for (int j = 32; j <= 35; j++) {
cfg_.kf_max_dist = j;
+ key_frame_spacing_ = j;
ResetModel();
assign_layer_bitrates(&cfg_, &svc_params_, cfg_.ss_number_layers,
cfg_.ts_number_layers, cfg_.temporal_layering_mode,
svc_params_.scaling_factor_den[1] = 288;
svc_params_.scaling_factor_num[2] = 288;
svc_params_.scaling_factor_den[2] = 288;
- cfg_.rc_dropframe_thresh = 0;
+ cfg_.rc_dropframe_thresh = 30;
cfg_.kf_max_dist = 9999;
number_spatial_layers_ = cfg_.ss_number_layers;
number_temporal_layers_ = cfg_.ts_number_layers;
EXPECT_GE(static_cast<uint32_t>(limit), error)
<< "Error: " << size_ << "x" << size_
<< " inverse transform has error " << error << " at " << w << ","
- << h << " org:" << (int)src_[h * stride_ + w]
- << " opt:" << (int)dst_[h * stride_ + w];
+ << h;
if (::testing::Test::HasFailure()) {
printf("Size: %d Transform type: %d\n", size_, tx_type_);
return;
static const FuncInfo ht_neon_func_info[] = {
#if CONFIG_VP9_HIGHBITDEPTH
-// TODO(linfengz): reenable these functions once test vector failures are
-// addressed.
-#if 0
{ &vp9_highbd_fht4x4_c, &highbd_iht_wrapper<vp9_highbd_iht4x4_16_add_neon>, 4,
2 },
{ &vp9_highbd_fht8x8_c, &highbd_iht_wrapper<vp9_highbd_iht8x8_64_add_neon>, 8,
2 },
-#endif
+ { &vp9_highbd_fht16x16_c,
+ &highbd_iht_wrapper<vp9_highbd_iht16x16_256_add_neon>, 16, 2 },
#endif
{ &vp9_fht4x4_c, &iht_wrapper<vp9_iht4x4_16_add_neon>, 4, 1 },
{ &vp9_fht8x8_c, &iht_wrapper<vp9_iht8x8_64_add_neon>, 8, 1 },
PreEncodeFrameHook(video, encoder.get());
encoder->EncodeFrame(video, frame_flags_);
- PostEncodeFrameHook(encoder.get());
+ PostEncodeFrameHook();
CxDataIterator iter = encoder->GetCxData();
virtual void PreEncodeFrameHook(VideoSource * /*video*/,
Encoder * /*encoder*/) {}
- virtual void PostEncodeFrameHook(Encoder * /*encoder*/) {}
+ virtual void PostEncodeFrameHook() {}
// Hook to be called on every compressed data packet.
virtual void FramePktHook(const vpx_codec_cx_pkt_t * /*pkt*/) {}
#endif // _WIN64
-#if ARCH_X86 || ARCH_X86_64
+#if ARCH_X86
#if defined(__GNUC__)
namespace libvpx_test {
} // namespace libvpx_test
#endif // __GNUC__
-#endif // ARCH_X86 || ARCH_X86_64
+#endif // ARCH_X86
#ifndef API_REGISTER_STATE_CHECK
#define API_REGISTER_STATE_CHECK ASM_REGISTER_STATE_CHECK
+++ /dev/null
-/*
- * Copyright (c) 2013 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 <string>
-#include "third_party/googletest/src/include/gtest/gtest.h"
-#include "test/codec_factory.h"
-#include "test/decode_test_driver.h"
-#include "test/i420_video_source.h"
-
-#include "vp9/decoder/vp9_decoder.h"
-
-#include "vpx/svc_context.h"
-#include "vpx/vp8cx.h"
-#include "vpx/vpx_encoder.h"
-
-namespace {
-
-using libvpx_test::CodecFactory;
-using libvpx_test::Decoder;
-using libvpx_test::DxDataIterator;
-using libvpx_test::VP9CodecFactory;
-
-class SvcTest : public ::testing::Test {
- protected:
- static const uint32_t kWidth = 352;
- static const uint32_t kHeight = 288;
-
- SvcTest()
- : codec_iface_(0), test_file_name_("hantro_collage_w352h288.yuv"),
- codec_initialized_(false), decoder_(0) {
- memset(&svc_, 0, sizeof(svc_));
- memset(&codec_, 0, sizeof(codec_));
- memset(&codec_enc_, 0, sizeof(codec_enc_));
- }
-
- virtual ~SvcTest() {}
-
- virtual void SetUp() {
- svc_.log_level = SVC_LOG_DEBUG;
- svc_.log_print = 0;
-
- codec_iface_ = vpx_codec_vp9_cx();
- const vpx_codec_err_t res =
- vpx_codec_enc_config_default(codec_iface_, &codec_enc_, 0);
- EXPECT_EQ(VPX_CODEC_OK, res);
-
- codec_enc_.g_w = kWidth;
- codec_enc_.g_h = kHeight;
- codec_enc_.g_timebase.num = 1;
- codec_enc_.g_timebase.den = 60;
- codec_enc_.kf_min_dist = 100;
- codec_enc_.kf_max_dist = 100;
-
- vpx_codec_dec_cfg_t dec_cfg = vpx_codec_dec_cfg_t();
- VP9CodecFactory codec_factory;
- decoder_ = codec_factory.CreateDecoder(dec_cfg, 0);
-
- tile_columns_ = 0;
- tile_rows_ = 0;
- }
-
- virtual void TearDown() {
- ReleaseEncoder();
- delete (decoder_);
- }
-
- void InitializeEncoder() {
- const vpx_codec_err_t res =
- vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
- EXPECT_EQ(VPX_CODEC_OK, res);
- vpx_codec_control(&codec_, VP8E_SET_CPUUSED, 4); // Make the test faster
- vpx_codec_control(&codec_, VP9E_SET_TILE_COLUMNS, tile_columns_);
- vpx_codec_control(&codec_, VP9E_SET_TILE_ROWS, tile_rows_);
- codec_initialized_ = true;
- }
-
- void ReleaseEncoder() {
- vpx_svc_release(&svc_);
- if (codec_initialized_) vpx_codec_destroy(&codec_);
- codec_initialized_ = false;
- }
-
- void GetStatsData(std::string *const stats_buf) {
- vpx_codec_iter_t iter = NULL;
- const vpx_codec_cx_pkt_t *cx_pkt;
-
- while ((cx_pkt = vpx_codec_get_cx_data(&codec_, &iter)) != NULL) {
- if (cx_pkt->kind == VPX_CODEC_STATS_PKT) {
- EXPECT_GT(cx_pkt->data.twopass_stats.sz, 0U);
- ASSERT_TRUE(cx_pkt->data.twopass_stats.buf != NULL);
- stats_buf->append(static_cast<char *>(cx_pkt->data.twopass_stats.buf),
- cx_pkt->data.twopass_stats.sz);
- }
- }
- }
-
- void Pass1EncodeNFrames(const int n, const int layers,
- std::string *const stats_buf) {
- vpx_codec_err_t res;
-
- ASSERT_GT(n, 0);
- ASSERT_GT(layers, 0);
- svc_.spatial_layers = layers;
- codec_enc_.g_pass = VPX_RC_FIRST_PASS;
- InitializeEncoder();
-
- libvpx_test::I420VideoSource video(
- test_file_name_, codec_enc_.g_w, codec_enc_.g_h,
- codec_enc_.g_timebase.den, codec_enc_.g_timebase.num, 0, 30);
- video.Begin();
-
- for (int i = 0; i < n; ++i) {
- res = vpx_svc_encode(&svc_, &codec_, video.img(), video.pts(),
- video.duration(), VPX_DL_GOOD_QUALITY);
- ASSERT_EQ(VPX_CODEC_OK, res);
- GetStatsData(stats_buf);
- video.Next();
- }
-
- // Flush encoder and test EOS packet.
- res = vpx_svc_encode(&svc_, &codec_, NULL, video.pts(), video.duration(),
- VPX_DL_GOOD_QUALITY);
- ASSERT_EQ(VPX_CODEC_OK, res);
- GetStatsData(stats_buf);
-
- ReleaseEncoder();
- }
-
- void StoreFrames(const size_t max_frame_received,
- struct vpx_fixed_buf *const outputs,
- size_t *const frame_received) {
- vpx_codec_iter_t iter = NULL;
- const vpx_codec_cx_pkt_t *cx_pkt;
-
- while ((cx_pkt = vpx_codec_get_cx_data(&codec_, &iter)) != NULL) {
- if (cx_pkt->kind == VPX_CODEC_CX_FRAME_PKT) {
- const size_t frame_size = cx_pkt->data.frame.sz;
-
- EXPECT_GT(frame_size, 0U);
- ASSERT_TRUE(cx_pkt->data.frame.buf != NULL);
- ASSERT_LT(*frame_received, max_frame_received);
-
- if (*frame_received == 0)
- EXPECT_EQ(1, !!(cx_pkt->data.frame.flags & VPX_FRAME_IS_KEY));
-
- outputs[*frame_received].buf = malloc(frame_size + 16);
- ASSERT_TRUE(outputs[*frame_received].buf != NULL);
- memcpy(outputs[*frame_received].buf, cx_pkt->data.frame.buf,
- frame_size);
- outputs[*frame_received].sz = frame_size;
- ++(*frame_received);
- }
- }
- }
-
- void Pass2EncodeNFrames(std::string *const stats_buf, const int n,
- const int layers,
- struct vpx_fixed_buf *const outputs) {
- vpx_codec_err_t res;
- size_t frame_received = 0;
-
- ASSERT_TRUE(outputs != NULL);
- ASSERT_GT(n, 0);
- ASSERT_GT(layers, 0);
- svc_.spatial_layers = layers;
- codec_enc_.rc_target_bitrate = 500;
- if (codec_enc_.g_pass == VPX_RC_LAST_PASS) {
- ASSERT_TRUE(stats_buf != NULL);
- ASSERT_GT(stats_buf->size(), 0U);
- codec_enc_.rc_twopass_stats_in.buf = &(*stats_buf)[0];
- codec_enc_.rc_twopass_stats_in.sz = stats_buf->size();
- }
- InitializeEncoder();
-
- libvpx_test::I420VideoSource video(
- test_file_name_, codec_enc_.g_w, codec_enc_.g_h,
- codec_enc_.g_timebase.den, codec_enc_.g_timebase.num, 0, 30);
- video.Begin();
-
- for (int i = 0; i < n; ++i) {
- res = vpx_svc_encode(&svc_, &codec_, video.img(), video.pts(),
- video.duration(), VPX_DL_GOOD_QUALITY);
- ASSERT_EQ(VPX_CODEC_OK, res);
- StoreFrames(n, outputs, &frame_received);
- video.Next();
- }
-
- // Flush encoder.
- res = vpx_svc_encode(&svc_, &codec_, NULL, 0, video.duration(),
- VPX_DL_GOOD_QUALITY);
- EXPECT_EQ(VPX_CODEC_OK, res);
- StoreFrames(n, outputs, &frame_received);
-
- EXPECT_EQ(frame_received, static_cast<size_t>(n));
-
- ReleaseEncoder();
- }
-
- void DecodeNFrames(const struct vpx_fixed_buf *const inputs, const int n) {
- int decoded_frames = 0;
- int received_frames = 0;
-
- ASSERT_TRUE(inputs != NULL);
- ASSERT_GT(n, 0);
-
- for (int i = 0; i < n; ++i) {
- ASSERT_TRUE(inputs[i].buf != NULL);
- ASSERT_GT(inputs[i].sz, 0U);
- const vpx_codec_err_t res_dec = decoder_->DecodeFrame(
- static_cast<const uint8_t *>(inputs[i].buf), inputs[i].sz);
- ASSERT_EQ(VPX_CODEC_OK, res_dec) << decoder_->DecodeError();
- ++decoded_frames;
-
- DxDataIterator dec_iter = decoder_->GetDxData();
- while (dec_iter.Next() != NULL) {
- ++received_frames;
- }
- }
- EXPECT_EQ(decoded_frames, n);
- EXPECT_EQ(received_frames, n);
- }
-
- void DropEnhancementLayers(struct vpx_fixed_buf *const inputs,
- const int num_super_frames,
- const int remained_spatial_layers) {
- ASSERT_TRUE(inputs != NULL);
- ASSERT_GT(num_super_frames, 0);
- ASSERT_GT(remained_spatial_layers, 0);
-
- for (int i = 0; i < num_super_frames; ++i) {
- uint32_t frame_sizes[8] = { 0 };
- int frame_count = 0;
- int frames_found = 0;
- int frame;
- ASSERT_TRUE(inputs[i].buf != NULL);
- ASSERT_GT(inputs[i].sz, 0U);
-
- vpx_codec_err_t res = vp9_parse_superframe_index(
- static_cast<const uint8_t *>(inputs[i].buf), inputs[i].sz,
- frame_sizes, &frame_count, NULL, NULL);
- ASSERT_EQ(VPX_CODEC_OK, res);
-
- if (frame_count == 0) {
- // There's no super frame but only a single frame.
- ASSERT_EQ(1, remained_spatial_layers);
- } else {
- // Found a super frame.
- uint8_t *frame_data = static_cast<uint8_t *>(inputs[i].buf);
- uint8_t *frame_start = frame_data;
- for (frame = 0; frame < frame_count; ++frame) {
- // Looking for a visible frame.
- if (frame_data[0] & 0x02) {
- ++frames_found;
- if (frames_found == remained_spatial_layers) break;
- }
- frame_data += frame_sizes[frame];
- }
- ASSERT_LT(frame, frame_count)
- << "Couldn't find a visible frame. "
- << "remained_spatial_layers: " << remained_spatial_layers
- << " super_frame: " << i;
- if (frame == frame_count - 1) continue;
-
- frame_data += frame_sizes[frame];
-
- // We need to add one more frame for multiple frame contexts.
- uint8_t marker =
- static_cast<const uint8_t *>(inputs[i].buf)[inputs[i].sz - 1];
- const uint32_t mag = ((marker >> 3) & 0x3) + 1;
- const size_t index_sz = 2 + mag * frame_count;
- const size_t new_index_sz = 2 + mag * (frame + 1);
- marker &= 0x0f8;
- marker |= frame;
-
- // Copy existing frame sizes.
- memmove(frame_data + 1, frame_start + inputs[i].sz - index_sz + 1,
- new_index_sz - 2);
- // New marker.
- frame_data[0] = marker;
- frame_data += (mag * (frame + 1) + 1);
-
- *frame_data++ = marker;
- inputs[i].sz = frame_data - frame_start;
- }
- }
- }
-
- void FreeBitstreamBuffers(struct vpx_fixed_buf *const inputs, const int n) {
- ASSERT_TRUE(inputs != NULL);
- ASSERT_GT(n, 0);
-
- for (int i = 0; i < n; ++i) {
- free(inputs[i].buf);
- inputs[i].buf = NULL;
- inputs[i].sz = 0;
- }
- }
-
- SvcContext svc_;
- vpx_codec_ctx_t codec_;
- struct vpx_codec_enc_cfg codec_enc_;
- vpx_codec_iface_t *codec_iface_;
- std::string test_file_name_;
- bool codec_initialized_;
- Decoder *decoder_;
- int tile_columns_;
- int tile_rows_;
-};
-
-TEST_F(SvcTest, SvcInit) {
- // test missing parameters
- vpx_codec_err_t res = vpx_svc_init(NULL, &codec_, codec_iface_, &codec_enc_);
- EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
- res = vpx_svc_init(&svc_, NULL, codec_iface_, &codec_enc_);
- EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
- res = vpx_svc_init(&svc_, &codec_, NULL, &codec_enc_);
- EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
- res = vpx_svc_init(&svc_, &codec_, codec_iface_, NULL);
- EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
- svc_.spatial_layers = 6; // too many layers
- res = vpx_svc_init(&svc_, &codec_, codec_iface_, &codec_enc_);
- EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
- svc_.spatial_layers = 0; // use default layers
- InitializeEncoder();
- EXPECT_EQ(VPX_SS_DEFAULT_LAYERS, svc_.spatial_layers);
-}
-
-TEST_F(SvcTest, InitTwoLayers) {
- svc_.spatial_layers = 2;
- InitializeEncoder();
-}
-
-TEST_F(SvcTest, InvalidOptions) {
- vpx_codec_err_t res = vpx_svc_set_options(&svc_, NULL);
- EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
- res = vpx_svc_set_options(&svc_, "not-an-option=1");
- EXPECT_EQ(VPX_CODEC_OK, res);
- res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
- EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-}
-
-TEST_F(SvcTest, SetLayersOption) {
- vpx_codec_err_t res = vpx_svc_set_options(&svc_, "spatial-layers=3");
- EXPECT_EQ(VPX_CODEC_OK, res);
- InitializeEncoder();
- EXPECT_EQ(3, svc_.spatial_layers);
-}
-
-TEST_F(SvcTest, SetMultipleOptions) {
- vpx_codec_err_t res =
- vpx_svc_set_options(&svc_, "spatial-layers=2 scale-factors=1/3,2/3");
- EXPECT_EQ(VPX_CODEC_OK, res);
- InitializeEncoder();
- EXPECT_EQ(2, svc_.spatial_layers);
-}
-
-TEST_F(SvcTest, SetScaleFactorsOption) {
- svc_.spatial_layers = 2;
- vpx_codec_err_t res =
- vpx_svc_set_options(&svc_, "scale-factors=not-scale-factors");
- EXPECT_EQ(VPX_CODEC_OK, res);
- res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
- EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
- res = vpx_svc_set_options(&svc_, "scale-factors=1/3, 3*3");
- EXPECT_EQ(VPX_CODEC_OK, res);
- res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
- EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
- res = vpx_svc_set_options(&svc_, "scale-factors=1/3");
- EXPECT_EQ(VPX_CODEC_OK, res);
- res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
- EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
- res = vpx_svc_set_options(&svc_, "scale-factors=1/3,2/3");
- EXPECT_EQ(VPX_CODEC_OK, res);
- InitializeEncoder();
-}
-
-TEST_F(SvcTest, SetQuantizersOption) {
- svc_.spatial_layers = 2;
- vpx_codec_err_t res = vpx_svc_set_options(&svc_, "max-quantizers=nothing");
- EXPECT_EQ(VPX_CODEC_OK, res);
- res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
- EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
- res = vpx_svc_set_options(&svc_, "min-quantizers=nothing");
- EXPECT_EQ(VPX_CODEC_OK, res);
- res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
- EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
- res = vpx_svc_set_options(&svc_, "max-quantizers=40");
- EXPECT_EQ(VPX_CODEC_OK, res);
- res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
- EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
- res = vpx_svc_set_options(&svc_, "min-quantizers=40");
- EXPECT_EQ(VPX_CODEC_OK, res);
- res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
- EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
- res = vpx_svc_set_options(&svc_, "max-quantizers=30,30 min-quantizers=40,40");
- EXPECT_EQ(VPX_CODEC_OK, res);
- res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
- EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
- res = vpx_svc_set_options(&svc_, "max-quantizers=40,40 min-quantizers=30,30");
- InitializeEncoder();
-}
-
-TEST_F(SvcTest, SetAutoAltRefOption) {
- svc_.spatial_layers = 5;
- vpx_codec_err_t res = vpx_svc_set_options(&svc_, "auto-alt-refs=none");
- EXPECT_EQ(VPX_CODEC_OK, res);
- res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
- EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
- res = vpx_svc_set_options(&svc_, "auto-alt-refs=1,1,1,1,0");
- EXPECT_EQ(VPX_CODEC_OK, res);
- res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
- EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
- vpx_svc_set_options(&svc_, "auto-alt-refs=0,1,1,1,0");
- InitializeEncoder();
-}
-
-// Test that decoder can handle an SVC frame as the first frame in a sequence.
-TEST_F(SvcTest, OnePassEncodeOneFrame) {
- codec_enc_.g_pass = VPX_RC_ONE_PASS;
- vpx_fixed_buf output = vpx_fixed_buf();
- Pass2EncodeNFrames(NULL, 1, 2, &output);
- DecodeNFrames(&output, 1);
- FreeBitstreamBuffers(&output, 1);
-}
-
-TEST_F(SvcTest, OnePassEncodeThreeFrames) {
- codec_enc_.g_pass = VPX_RC_ONE_PASS;
- codec_enc_.g_lag_in_frames = 0;
- vpx_fixed_buf outputs[3];
- memset(&outputs[0], 0, sizeof(outputs));
- Pass2EncodeNFrames(NULL, 3, 2, &outputs[0]);
- DecodeNFrames(&outputs[0], 3);
- FreeBitstreamBuffers(&outputs[0], 3);
-}
-
-TEST_F(SvcTest, TwoPassEncode10Frames) {
- // First pass encode
- std::string stats_buf;
- Pass1EncodeNFrames(10, 2, &stats_buf);
-
- // Second pass encode
- codec_enc_.g_pass = VPX_RC_LAST_PASS;
- vpx_fixed_buf outputs[10];
- memset(&outputs[0], 0, sizeof(outputs));
- Pass2EncodeNFrames(&stats_buf, 10, 2, &outputs[0]);
- DecodeNFrames(&outputs[0], 10);
- FreeBitstreamBuffers(&outputs[0], 10);
-}
-
-TEST_F(SvcTest, TwoPassEncode20FramesWithAltRef) {
- // First pass encode
- std::string stats_buf;
- Pass1EncodeNFrames(20, 2, &stats_buf);
-
- // Second pass encode
- codec_enc_.g_pass = VPX_RC_LAST_PASS;
- vpx_svc_set_options(&svc_, "auto-alt-refs=1,1");
- vpx_fixed_buf outputs[20];
- memset(&outputs[0], 0, sizeof(outputs));
- Pass2EncodeNFrames(&stats_buf, 20, 2, &outputs[0]);
- DecodeNFrames(&outputs[0], 20);
- FreeBitstreamBuffers(&outputs[0], 20);
-}
-
-TEST_F(SvcTest, TwoPassEncode2SpatialLayersDecodeBaseLayerOnly) {
- // First pass encode
- std::string stats_buf;
- Pass1EncodeNFrames(10, 2, &stats_buf);
-
- // Second pass encode
- codec_enc_.g_pass = VPX_RC_LAST_PASS;
- vpx_svc_set_options(&svc_, "auto-alt-refs=1,1");
- vpx_fixed_buf outputs[10];
- memset(&outputs[0], 0, sizeof(outputs));
- Pass2EncodeNFrames(&stats_buf, 10, 2, &outputs[0]);
- DropEnhancementLayers(&outputs[0], 10, 1);
- DecodeNFrames(&outputs[0], 10);
- FreeBitstreamBuffers(&outputs[0], 10);
-}
-
-TEST_F(SvcTest, TwoPassEncode5SpatialLayersDecode54321Layers) {
- // First pass encode
- std::string stats_buf;
- Pass1EncodeNFrames(10, 5, &stats_buf);
-
- // Second pass encode
- codec_enc_.g_pass = VPX_RC_LAST_PASS;
- vpx_svc_set_options(&svc_, "auto-alt-refs=0,1,1,1,0");
- vpx_fixed_buf outputs[10];
- memset(&outputs[0], 0, sizeof(outputs));
- Pass2EncodeNFrames(&stats_buf, 10, 5, &outputs[0]);
-
- DecodeNFrames(&outputs[0], 10);
- DropEnhancementLayers(&outputs[0], 10, 4);
- DecodeNFrames(&outputs[0], 10);
- DropEnhancementLayers(&outputs[0], 10, 3);
- DecodeNFrames(&outputs[0], 10);
- DropEnhancementLayers(&outputs[0], 10, 2);
- DecodeNFrames(&outputs[0], 10);
- DropEnhancementLayers(&outputs[0], 10, 1);
- DecodeNFrames(&outputs[0], 10);
-
- FreeBitstreamBuffers(&outputs[0], 10);
-}
-
-TEST_F(SvcTest, TwoPassEncode2SNRLayers) {
- // First pass encode
- std::string stats_buf;
- vpx_svc_set_options(&svc_, "scale-factors=1/1,1/1");
- Pass1EncodeNFrames(20, 2, &stats_buf);
-
- // Second pass encode
- codec_enc_.g_pass = VPX_RC_LAST_PASS;
- vpx_svc_set_options(&svc_, "auto-alt-refs=1,1 scale-factors=1/1,1/1");
- vpx_fixed_buf outputs[20];
- memset(&outputs[0], 0, sizeof(outputs));
- Pass2EncodeNFrames(&stats_buf, 20, 2, &outputs[0]);
- DecodeNFrames(&outputs[0], 20);
- FreeBitstreamBuffers(&outputs[0], 20);
-}
-
-TEST_F(SvcTest, TwoPassEncode3SNRLayersDecode321Layers) {
- // First pass encode
- std::string stats_buf;
- vpx_svc_set_options(&svc_, "scale-factors=1/1,1/1,1/1");
- Pass1EncodeNFrames(20, 3, &stats_buf);
-
- // Second pass encode
- codec_enc_.g_pass = VPX_RC_LAST_PASS;
- vpx_svc_set_options(&svc_, "auto-alt-refs=1,1,1 scale-factors=1/1,1/1,1/1");
- vpx_fixed_buf outputs[20];
- memset(&outputs[0], 0, sizeof(outputs));
- Pass2EncodeNFrames(&stats_buf, 20, 3, &outputs[0]);
- DecodeNFrames(&outputs[0], 20);
- DropEnhancementLayers(&outputs[0], 20, 2);
- DecodeNFrames(&outputs[0], 20);
- DropEnhancementLayers(&outputs[0], 20, 1);
- DecodeNFrames(&outputs[0], 20);
-
- FreeBitstreamBuffers(&outputs[0], 20);
-}
-
-TEST_F(SvcTest, SetMultipleFrameContextsOption) {
- svc_.spatial_layers = 5;
- vpx_codec_err_t res = vpx_svc_set_options(&svc_, "multi-frame-contexts=1");
- EXPECT_EQ(VPX_CODEC_OK, res);
- res = vpx_svc_init(&svc_, &codec_, vpx_codec_vp9_cx(), &codec_enc_);
- EXPECT_EQ(VPX_CODEC_INVALID_PARAM, res);
-
- svc_.spatial_layers = 2;
- res = vpx_svc_set_options(&svc_, "multi-frame-contexts=1");
- InitializeEncoder();
-}
-
-TEST_F(SvcTest, TwoPassEncode2SpatialLayersWithMultipleFrameContexts) {
- // First pass encode
- std::string stats_buf;
- Pass1EncodeNFrames(10, 2, &stats_buf);
-
- // Second pass encode
- codec_enc_.g_pass = VPX_RC_LAST_PASS;
- codec_enc_.g_error_resilient = 0;
- vpx_svc_set_options(&svc_, "auto-alt-refs=1,1 multi-frame-contexts=1");
- vpx_fixed_buf outputs[10];
- memset(&outputs[0], 0, sizeof(outputs));
- Pass2EncodeNFrames(&stats_buf, 10, 2, &outputs[0]);
- DecodeNFrames(&outputs[0], 10);
- FreeBitstreamBuffers(&outputs[0], 10);
-}
-
-TEST_F(SvcTest,
- TwoPassEncode2SpatialLayersWithMultipleFrameContextsDecodeBaselayer) {
- // First pass encode
- std::string stats_buf;
- Pass1EncodeNFrames(10, 2, &stats_buf);
-
- // Second pass encode
- codec_enc_.g_pass = VPX_RC_LAST_PASS;
- codec_enc_.g_error_resilient = 0;
- vpx_svc_set_options(&svc_, "auto-alt-refs=1,1 multi-frame-contexts=1");
- vpx_fixed_buf outputs[10];
- memset(&outputs[0], 0, sizeof(outputs));
- Pass2EncodeNFrames(&stats_buf, 10, 2, &outputs[0]);
- DropEnhancementLayers(&outputs[0], 10, 1);
- DecodeNFrames(&outputs[0], 10);
- FreeBitstreamBuffers(&outputs[0], 10);
-}
-
-TEST_F(SvcTest, TwoPassEncode2SNRLayersWithMultipleFrameContexts) {
- // First pass encode
- std::string stats_buf;
- vpx_svc_set_options(&svc_, "scale-factors=1/1,1/1");
- Pass1EncodeNFrames(10, 2, &stats_buf);
-
- // Second pass encode
- codec_enc_.g_pass = VPX_RC_LAST_PASS;
- codec_enc_.g_error_resilient = 0;
- vpx_svc_set_options(&svc_,
- "auto-alt-refs=1,1 scale-factors=1/1,1/1 "
- "multi-frame-contexts=1");
- vpx_fixed_buf outputs[10];
- memset(&outputs[0], 0, sizeof(outputs));
- Pass2EncodeNFrames(&stats_buf, 10, 2, &outputs[0]);
- DecodeNFrames(&outputs[0], 10);
- FreeBitstreamBuffers(&outputs[0], 10);
-}
-
-TEST_F(SvcTest,
- TwoPassEncode3SNRLayersWithMultipleFrameContextsDecode321Layer) {
- // First pass encode
- std::string stats_buf;
- vpx_svc_set_options(&svc_, "scale-factors=1/1,1/1,1/1");
- Pass1EncodeNFrames(10, 3, &stats_buf);
-
- // Second pass encode
- codec_enc_.g_pass = VPX_RC_LAST_PASS;
- codec_enc_.g_error_resilient = 0;
- vpx_svc_set_options(&svc_,
- "auto-alt-refs=1,1,1 scale-factors=1/1,1/1,1/1 "
- "multi-frame-contexts=1");
- vpx_fixed_buf outputs[10];
- memset(&outputs[0], 0, sizeof(outputs));
- Pass2EncodeNFrames(&stats_buf, 10, 3, &outputs[0]);
-
- DecodeNFrames(&outputs[0], 10);
- DropEnhancementLayers(&outputs[0], 10, 2);
- DecodeNFrames(&outputs[0], 10);
- DropEnhancementLayers(&outputs[0], 10, 1);
- DecodeNFrames(&outputs[0], 10);
-
- FreeBitstreamBuffers(&outputs[0], 10);
-}
-
-TEST_F(SvcTest, TwoPassEncode2TemporalLayers) {
- // First pass encode
- std::string stats_buf;
- vpx_svc_set_options(&svc_, "scale-factors=1/1");
- svc_.temporal_layers = 2;
- Pass1EncodeNFrames(10, 1, &stats_buf);
-
- // Second pass encode
- codec_enc_.g_pass = VPX_RC_LAST_PASS;
- svc_.temporal_layers = 2;
- vpx_svc_set_options(&svc_, "auto-alt-refs=1 scale-factors=1/1");
- vpx_fixed_buf outputs[10];
- memset(&outputs[0], 0, sizeof(outputs));
- Pass2EncodeNFrames(&stats_buf, 10, 1, &outputs[0]);
- DecodeNFrames(&outputs[0], 10);
- FreeBitstreamBuffers(&outputs[0], 10);
-}
-
-TEST_F(SvcTest, TwoPassEncode2TemporalLayersWithMultipleFrameContexts) {
- // First pass encode
- std::string stats_buf;
- vpx_svc_set_options(&svc_, "scale-factors=1/1");
- svc_.temporal_layers = 2;
- Pass1EncodeNFrames(10, 1, &stats_buf);
-
- // Second pass encode
- codec_enc_.g_pass = VPX_RC_LAST_PASS;
- svc_.temporal_layers = 2;
- codec_enc_.g_error_resilient = 0;
- vpx_svc_set_options(&svc_,
- "auto-alt-refs=1 scale-factors=1/1 "
- "multi-frame-contexts=1");
- vpx_fixed_buf outputs[10];
- memset(&outputs[0], 0, sizeof(outputs));
- Pass2EncodeNFrames(&stats_buf, 10, 1, &outputs[0]);
- DecodeNFrames(&outputs[0], 10);
- FreeBitstreamBuffers(&outputs[0], 10);
-}
-
-TEST_F(SvcTest, TwoPassEncode2TemporalLayersDecodeBaseLayer) {
- // First pass encode
- std::string stats_buf;
- vpx_svc_set_options(&svc_, "scale-factors=1/1");
- svc_.temporal_layers = 2;
- Pass1EncodeNFrames(10, 1, &stats_buf);
-
- // Second pass encode
- codec_enc_.g_pass = VPX_RC_LAST_PASS;
- svc_.temporal_layers = 2;
- vpx_svc_set_options(&svc_, "auto-alt-refs=1 scale-factors=1/1");
- vpx_fixed_buf outputs[10];
- memset(&outputs[0], 0, sizeof(outputs));
- Pass2EncodeNFrames(&stats_buf, 10, 1, &outputs[0]);
-
- vpx_fixed_buf base_layer[5];
- for (int i = 0; i < 5; ++i) base_layer[i] = outputs[i * 2];
-
- DecodeNFrames(&base_layer[0], 5);
- FreeBitstreamBuffers(&outputs[0], 10);
-}
-
-TEST_F(SvcTest,
- TwoPassEncode2TemporalLayersWithMultipleFrameContextsDecodeBaseLayer) {
- // First pass encode
- std::string stats_buf;
- vpx_svc_set_options(&svc_, "scale-factors=1/1");
- svc_.temporal_layers = 2;
- Pass1EncodeNFrames(10, 1, &stats_buf);
-
- // Second pass encode
- codec_enc_.g_pass = VPX_RC_LAST_PASS;
- svc_.temporal_layers = 2;
- codec_enc_.g_error_resilient = 0;
- vpx_svc_set_options(&svc_,
- "auto-alt-refs=1 scale-factors=1/1 "
- "multi-frame-contexts=1");
- vpx_fixed_buf outputs[10];
- memset(&outputs[0], 0, sizeof(outputs));
- Pass2EncodeNFrames(&stats_buf, 10, 1, &outputs[0]);
-
- vpx_fixed_buf base_layer[5];
- for (int i = 0; i < 5; ++i) base_layer[i] = outputs[i * 2];
-
- DecodeNFrames(&base_layer[0], 5);
- FreeBitstreamBuffers(&outputs[0], 10);
-}
-
-TEST_F(SvcTest, TwoPassEncode2TemporalLayersWithTiles) {
- // First pass encode
- std::string stats_buf;
- vpx_svc_set_options(&svc_, "scale-factors=1/1");
- svc_.temporal_layers = 2;
- Pass1EncodeNFrames(10, 1, &stats_buf);
-
- // Second pass encode
- codec_enc_.g_pass = VPX_RC_LAST_PASS;
- svc_.temporal_layers = 2;
- vpx_svc_set_options(&svc_, "auto-alt-refs=1 scale-factors=1/1");
- codec_enc_.g_w = 704;
- codec_enc_.g_h = 144;
- tile_columns_ = 1;
- tile_rows_ = 1;
- vpx_fixed_buf outputs[10];
- memset(&outputs[0], 0, sizeof(outputs));
- Pass2EncodeNFrames(&stats_buf, 10, 1, &outputs[0]);
- DecodeNFrames(&outputs[0], 10);
- FreeBitstreamBuffers(&outputs[0], 10);
-}
-
-TEST_F(SvcTest, TwoPassEncode2TemporalLayersWithMultipleFrameContextsAndTiles) {
- // First pass encode
- std::string stats_buf;
- vpx_svc_set_options(&svc_, "scale-factors=1/1");
- svc_.temporal_layers = 2;
- Pass1EncodeNFrames(10, 1, &stats_buf);
-
- // Second pass encode
- codec_enc_.g_pass = VPX_RC_LAST_PASS;
- svc_.temporal_layers = 2;
- codec_enc_.g_error_resilient = 0;
- codec_enc_.g_w = 704;
- codec_enc_.g_h = 144;
- tile_columns_ = 1;
- tile_rows_ = 1;
- vpx_svc_set_options(&svc_,
- "auto-alt-refs=1 scale-factors=1/1 "
- "multi-frame-contexts=1");
- vpx_fixed_buf outputs[10];
- memset(&outputs[0], 0, sizeof(outputs));
- Pass2EncodeNFrames(&stats_buf, 10, 1, &outputs[0]);
- DecodeNFrames(&outputs[0], 10);
- FreeBitstreamBuffers(&outputs[0], 10);
-}
-
-} // namespace
LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_subtract_test.cc
ifeq ($(CONFIG_VP9_ENCODER),yes)
-LIBVPX_TEST_SRCS-$(CONFIG_SPATIAL_SVC) += svc_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_INTERNAL_STATS) += blockiness_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_INTERNAL_STATS) += consistency_test.cc
endif
+++ /dev/null
-#!/bin/sh
-##
-## 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.
-##
-## This file tests the libvpx vp9_spatial_svc_encoder example. To add new
-## tests to to this file, do the following:
-## 1. Write a shell function (this is your test).
-## 2. Add the function to vp9_spatial_svc_tests (on a new line).
-##
-. $(dirname $0)/tools_common.sh
-
-# Environment check: $YUV_RAW_INPUT is required.
-vp9_spatial_svc_encoder_verify_environment() {
- if [ ! -e "${YUV_RAW_INPUT}" ]; then
- echo "Libvpx test data must exist in LIBVPX_TEST_DATA_PATH."
- return 1
- fi
-}
-
-# Runs vp9_spatial_svc_encoder. $1 is the test name.
-vp9_spatial_svc_encoder() {
- local readonly \
- encoder="${LIBVPX_BIN_PATH}/vp9_spatial_svc_encoder${VPX_TEST_EXE_SUFFIX}"
- local readonly test_name="$1"
- local readonly \
- output_file="${VPX_TEST_OUTPUT_DIR}/vp9_ssvc_encoder${test_name}.ivf"
- local readonly frames_to_encode=10
- local readonly max_kf=9999
-
- shift
-
- if [ ! -x "${encoder}" ]; then
- elog "${encoder} does not exist or is not executable."
- return 1
- fi
-
- eval "${VPX_TEST_PREFIX}" "${encoder}" -w "${YUV_RAW_INPUT_WIDTH}" \
- -h "${YUV_RAW_INPUT_HEIGHT}" -k "${max_kf}" -f "${frames_to_encode}" \
- "$@" "${YUV_RAW_INPUT}" "${output_file}" ${devnull}
-
- [ -e "${output_file}" ] || return 1
-}
-
-# Each test is run with layer count 1-$vp9_ssvc_test_layers.
-vp9_ssvc_test_layers=5
-
-vp9_spatial_svc() {
- if [ "$(vp9_encode_available)" = "yes" ]; then
- local readonly test_name="vp9_spatial_svc"
- for layers in $(seq 1 ${vp9_ssvc_test_layers}); do
- vp9_spatial_svc_encoder "${test_name}" -sl ${layers}
- done
- fi
-}
-
-readonly vp9_spatial_svc_tests="DISABLED_vp9_spatial_svc_mode_i
- DISABLED_vp9_spatial_svc_mode_altip
- DISABLED_vp9_spatial_svc_mode_ip
- DISABLED_vp9_spatial_svc_mode_gf
- vp9_spatial_svc"
-
-if [ "$(vpx_config_option_enabled CONFIG_SPATIAL_SVC)" = "yes" ]; then
- run_tests \
- vp9_spatial_svc_encoder_verify_environment \
- "${vp9_spatial_svc_tests}"
-fi
#include "vpx_scale/yv12config.h"
namespace libvpx_test {
+namespace {
+
+#if ARCH_ARM || (ARCH_MIPS && !HAVE_MIPS64) || ARCH_X86
+// Avoid OOM failures on 32-bit platforms.
+const int kNumSizesToTest = 7;
+#else
+const int kNumSizesToTest = 8;
+#endif
+const int kSizesToTest[] = { 1, 15, 33, 145, 512, 1025, 3840, 16383 };
typedef void (*ExtendFrameBorderFunc)(YV12_BUFFER_CONFIG *ybf);
typedef void (*CopyFrameFunc)(const YV12_BUFFER_CONFIG *src_ybf,
void ExtendBorder() { ASM_REGISTER_STATE_CHECK(extend_fn_(&img_)); }
void RunTest() {
-#if ARCH_ARM
- // Some arm devices OOM when trying to allocate the largest buffers.
- static const int kNumSizesToTest = 6;
-#else
- static const int kNumSizesToTest = 7;
-#endif
- static const int kSizesToTest[] = { 1, 15, 33, 145, 512, 1025, 16383 };
for (int h = 0; h < kNumSizesToTest; ++h) {
for (int w = 0; w < kNumSizesToTest; ++w) {
ASSERT_NO_FATAL_FAILURE(ResetImages(kSizesToTest[w], kSizesToTest[h]));
}
void RunTest() {
-#if ARCH_ARM
- // Some arm devices OOM when trying to allocate the largest buffers.
- static const int kNumSizesToTest = 6;
-#else
- static const int kNumSizesToTest = 7;
-#endif
- static const int kSizesToTest[] = { 1, 15, 33, 145, 512, 1025, 16383 };
for (int h = 0; h < kNumSizesToTest; ++h) {
for (int w = 0; w < kNumSizesToTest; ++w) {
ASSERT_NO_FATAL_FAILURE(ResetImages(kSizesToTest[w], kSizesToTest[h]));
INSTANTIATE_TEST_CASE_P(C, CopyFrameTest,
::testing::Values(vp8_yv12_copy_frame_c));
+} // namespace
} // namespace libvpx_test
#include "./vp8_rtcd.h"
#include "./vpx_dsp_rtcd.h"
+#include "vp8/common/common.h"
#include "vp8/common/postproc.h"
#include "vpx_dsp/variance.h"
#include "vpx_mem/vpx_mem.h"
{ 0, 1, 4, 5 }, { 2, 3, 6, 7 }, { 8, 9, 12, 13 }, { 10, 11, 14, 15 }
};
int i, j;
+ vp8_zero(*map);
for (i = 0; i < 4; ++i) {
map[i] = 1;
for (j = 0; j < 4 && map[j]; ++j) {
* NEAREST for subsequent blocks. The "Weight" parameter allows, to a
* limited extent, for some account to be taken of these factors.
*/
- return ((mvcost[0][(mv->as_mv.row - ref->as_mv.row) >> 1] +
- mvcost[1][(mv->as_mv.col - ref->as_mv.col) >> 1]) *
- Weight) >>
- 7;
+ const int mv_idx_row =
+ clamp((mv->as_mv.row - ref->as_mv.row) >> 1, 0, MVvals);
+ const int mv_idx_col =
+ clamp((mv->as_mv.col - ref->as_mv.col) >> 1, 0, MVvals);
+ return ((mvcost[0][mv_idx_row] + mvcost[1][mv_idx_col]) * Weight) >> 7;
}
static int mv_err_cost(int_mv *mv, int_mv *ref, int *mvcost[2],
int error_per_bit) {
/* Ignore mv costing if mvcost is NULL */
if (mvcost) {
- return ((mvcost[0][(mv->as_mv.row - ref->as_mv.row) >> 1] +
- mvcost[1][(mv->as_mv.col - ref->as_mv.col) >> 1]) *
- error_per_bit +
+ const int mv_idx_row =
+ clamp((mv->as_mv.row - ref->as_mv.row) >> 1, 0, MVvals);
+ const int mv_idx_col =
+ clamp((mv->as_mv.col - ref->as_mv.col) >> 1, 0, MVvals);
+ return ((mvcost[0][mv_idx_row] + mvcost[1][mv_idx_col]) * error_per_bit +
128) >>
8;
}
best_rd_sse = UINT_MAX;
#endif
+ // _uv variables are not set consistantly before calling update_best_mode.
+ rd.rate_uv = 0;
+ rd.distortion_uv = 0;
+
mode_mv = mode_mv_sb[sign_bias];
best_ref_mv.as_int = 0;
best_mode.rd = INT_MAX;
pkt.data.frame.flags = lib_flags << 16;
pkt.data.frame.width[0] = cpi->common.Width;
pkt.data.frame.height[0] = cpi->common.Height;
+ pkt.data.frame.last_spatial_layer_encoded = 0;
if (lib_flags & FRAMEFLAGS_KEY) {
pkt.data.frame.flags |= VPX_FRAME_IS_KEY;
--- /dev/null
+/*
+ * Copyright (c) 2018 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 "./vpx_dsp_rtcd.h"
+#include "vp9/common/vp9_enums.h"
+#include "vp9/common/arm/neon/vp9_iht_neon.h"
+#include "vpx_dsp/arm/highbd_idct_neon.h"
+#include "vpx_dsp/arm/idct_neon.h"
+#include "vpx_dsp/arm/transpose_neon.h"
+#include "vpx_dsp/inv_txfm.h"
+
+// Use macros to make sure argument lane is passed in as an constant integer.
+
+#define vmull_lane_s32_dual(in, c, lane, out) \
+ do { \
+ out[0].val[0] = vmull_lane_s32(vget_low_s32(in.val[0]), c, lane); \
+ out[0].val[1] = vmull_lane_s32(vget_low_s32(in.val[1]), c, lane); \
+ out[1].val[0] = vmull_lane_s32(vget_high_s32(in.val[0]), c, lane); \
+ out[1].val[1] = vmull_lane_s32(vget_high_s32(in.val[1]), c, lane); \
+ } while (0)
+
+#define vmlal_lane_s32_dual(in, c, lane, out) \
+ do { \
+ out[0].val[0] = \
+ vmlal_lane_s32(out[0].val[0], vget_low_s32(in.val[0]), c, lane); \
+ out[0].val[1] = \
+ vmlal_lane_s32(out[0].val[1], vget_low_s32(in.val[1]), c, lane); \
+ out[1].val[0] = \
+ vmlal_lane_s32(out[1].val[0], vget_high_s32(in.val[0]), c, lane); \
+ out[1].val[1] = \
+ vmlal_lane_s32(out[1].val[1], vget_high_s32(in.val[1]), c, lane); \
+ } while (0)
+
+#define vmlsl_lane_s32_dual(in, c, lane, out) \
+ do { \
+ out[0].val[0] = \
+ vmlsl_lane_s32(out[0].val[0], vget_low_s32(in.val[0]), c, lane); \
+ out[0].val[1] = \
+ vmlsl_lane_s32(out[0].val[1], vget_low_s32(in.val[1]), c, lane); \
+ out[1].val[0] = \
+ vmlsl_lane_s32(out[1].val[0], vget_high_s32(in.val[0]), c, lane); \
+ out[1].val[1] = \
+ vmlsl_lane_s32(out[1].val[1], vget_high_s32(in.val[1]), c, lane); \
+ } while (0)
+
+static INLINE int32x4x2_t
+highbd_dct_const_round_shift_low_8(const int64x2x2_t *const in) {
+ int32x4x2_t out;
+ out.val[0] = vcombine_s32(vrshrn_n_s64(in[0].val[0], DCT_CONST_BITS),
+ vrshrn_n_s64(in[1].val[0], DCT_CONST_BITS));
+ out.val[1] = vcombine_s32(vrshrn_n_s64(in[0].val[1], DCT_CONST_BITS),
+ vrshrn_n_s64(in[1].val[1], DCT_CONST_BITS));
+ return out;
+}
+
+#define highbd_iadst_half_butterfly(in, c, lane, out) \
+ do { \
+ int64x2x2_t t[2]; \
+ vmull_lane_s32_dual(in, c, lane, t); \
+ out = highbd_dct_const_round_shift_low_8(t); \
+ } while (0)
+
+#define highbd_iadst_butterfly(in0, in1, c, lane0, lane1, s0, s1) \
+ do { \
+ vmull_lane_s32_dual(in0, c, lane0, s0); \
+ vmull_lane_s32_dual(in0, c, lane1, s1); \
+ vmlal_lane_s32_dual(in1, c, lane1, s0); \
+ vmlsl_lane_s32_dual(in1, c, lane0, s1); \
+ } while (0)
+
+static INLINE int32x4x2_t vaddq_s32_dual(const int32x4x2_t in0,
+ const int32x4x2_t in1) {
+ int32x4x2_t out;
+ out.val[0] = vaddq_s32(in0.val[0], in1.val[0]);
+ out.val[1] = vaddq_s32(in0.val[1], in1.val[1]);
+ return out;
+}
+
+static INLINE int64x2x2_t vaddq_s64_dual(const int64x2x2_t in0,
+ const int64x2x2_t in1) {
+ int64x2x2_t out;
+ out.val[0] = vaddq_s64(in0.val[0], in1.val[0]);
+ out.val[1] = vaddq_s64(in0.val[1], in1.val[1]);
+ return out;
+}
+
+static INLINE int32x4x2_t vsubq_s32_dual(const int32x4x2_t in0,
+ const int32x4x2_t in1) {
+ int32x4x2_t out;
+ out.val[0] = vsubq_s32(in0.val[0], in1.val[0]);
+ out.val[1] = vsubq_s32(in0.val[1], in1.val[1]);
+ return out;
+}
+
+static INLINE int64x2x2_t vsubq_s64_dual(const int64x2x2_t in0,
+ const int64x2x2_t in1) {
+ int64x2x2_t out;
+ out.val[0] = vsubq_s64(in0.val[0], in1.val[0]);
+ out.val[1] = vsubq_s64(in0.val[1], in1.val[1]);
+ return out;
+}
+
+static INLINE int32x4x2_t vcombine_s32_dual(const int32x2x2_t in0,
+ const int32x2x2_t in1) {
+ int32x4x2_t out;
+ out.val[0] = vcombine_s32(in0.val[0], in1.val[0]);
+ out.val[1] = vcombine_s32(in0.val[1], in1.val[1]);
+ return out;
+}
+
+static INLINE int32x4x2_t highbd_add_dct_const_round_shift_low_8(
+ const int64x2x2_t *const in0, const int64x2x2_t *const in1) {
+ const int64x2x2_t sum_lo = vaddq_s64_dual(in0[0], in1[0]);
+ const int64x2x2_t sum_hi = vaddq_s64_dual(in0[1], in1[1]);
+ int32x2x2_t out_lo, out_hi;
+
+ out_lo.val[0] = vrshrn_n_s64(sum_lo.val[0], DCT_CONST_BITS);
+ out_lo.val[1] = vrshrn_n_s64(sum_lo.val[1], DCT_CONST_BITS);
+ out_hi.val[0] = vrshrn_n_s64(sum_hi.val[0], DCT_CONST_BITS);
+ out_hi.val[1] = vrshrn_n_s64(sum_hi.val[1], DCT_CONST_BITS);
+ return vcombine_s32_dual(out_lo, out_hi);
+}
+
+static INLINE int32x4x2_t highbd_sub_dct_const_round_shift_low_8(
+ const int64x2x2_t *const in0, const int64x2x2_t *const in1) {
+ const int64x2x2_t sub_lo = vsubq_s64_dual(in0[0], in1[0]);
+ const int64x2x2_t sub_hi = vsubq_s64_dual(in0[1], in1[1]);
+ int32x2x2_t out_lo, out_hi;
+
+ out_lo.val[0] = vrshrn_n_s64(sub_lo.val[0], DCT_CONST_BITS);
+ out_lo.val[1] = vrshrn_n_s64(sub_lo.val[1], DCT_CONST_BITS);
+ out_hi.val[0] = vrshrn_n_s64(sub_hi.val[0], DCT_CONST_BITS);
+ out_hi.val[1] = vrshrn_n_s64(sub_hi.val[1], DCT_CONST_BITS);
+ return vcombine_s32_dual(out_lo, out_hi);
+}
+
+static INLINE int32x4x2_t vnegq_s32_dual(const int32x4x2_t in) {
+ int32x4x2_t out;
+ out.val[0] = vnegq_s32(in.val[0]);
+ out.val[1] = vnegq_s32(in.val[1]);
+ return out;
+}
+
+void vpx_highbd_iadst16_neon(const int32_t *input, int32_t *output,
+ uint16_t *dest, const int stride, const int bd) {
+ const int32x4_t c_1_31_5_27 =
+ create_s32x4_neon(cospi_1_64, cospi_31_64, cospi_5_64, cospi_27_64);
+ const int32x4_t c_9_23_13_19 =
+ create_s32x4_neon(cospi_9_64, cospi_23_64, cospi_13_64, cospi_19_64);
+ const int32x4_t c_17_15_21_11 =
+ create_s32x4_neon(cospi_17_64, cospi_15_64, cospi_21_64, cospi_11_64);
+ const int32x4_t c_25_7_29_3 =
+ create_s32x4_neon(cospi_25_64, cospi_7_64, cospi_29_64, cospi_3_64);
+ const int32x4_t c_4_28_20_12 =
+ create_s32x4_neon(cospi_4_64, cospi_28_64, cospi_20_64, cospi_12_64);
+ const int32x4_t c_16_n16_8_24 =
+ create_s32x4_neon(cospi_16_64, -cospi_16_64, cospi_8_64, cospi_24_64);
+ int32x4x2_t in[16], out[16];
+ int32x4x2_t x[16], t[12];
+ int64x2x2_t s0[2], s1[2], s2[2], s3[2], s4[2], s5[2], s6[2], s7[2];
+ int64x2x2_t s8[2], s9[2], s10[2], s11[2], s12[2], s13[2], s14[2], s15[2];
+
+ // Load input (16x8)
+ in[0].val[0] = vld1q_s32(input);
+ in[0].val[1] = vld1q_s32(input + 4);
+ input += 8;
+ in[8].val[0] = vld1q_s32(input);
+ in[8].val[1] = vld1q_s32(input + 4);
+ input += 8;
+ in[1].val[0] = vld1q_s32(input);
+ in[1].val[1] = vld1q_s32(input + 4);
+ input += 8;
+ in[9].val[0] = vld1q_s32(input);
+ in[9].val[1] = vld1q_s32(input + 4);
+ input += 8;
+ in[2].val[0] = vld1q_s32(input);
+ in[2].val[1] = vld1q_s32(input + 4);
+ input += 8;
+ in[10].val[0] = vld1q_s32(input);
+ in[10].val[1] = vld1q_s32(input + 4);
+ input += 8;
+ in[3].val[0] = vld1q_s32(input);
+ in[3].val[1] = vld1q_s32(input + 4);
+ input += 8;
+ in[11].val[0] = vld1q_s32(input);
+ in[11].val[1] = vld1q_s32(input + 4);
+ input += 8;
+ in[4].val[0] = vld1q_s32(input);
+ in[4].val[1] = vld1q_s32(input + 4);
+ input += 8;
+ in[12].val[0] = vld1q_s32(input);
+ in[12].val[1] = vld1q_s32(input + 4);
+ input += 8;
+ in[5].val[0] = vld1q_s32(input);
+ in[5].val[1] = vld1q_s32(input + 4);
+ input += 8;
+ in[13].val[0] = vld1q_s32(input);
+ in[13].val[1] = vld1q_s32(input + 4);
+ input += 8;
+ in[6].val[0] = vld1q_s32(input);
+ in[6].val[1] = vld1q_s32(input + 4);
+ input += 8;
+ in[14].val[0] = vld1q_s32(input);
+ in[14].val[1] = vld1q_s32(input + 4);
+ input += 8;
+ in[7].val[0] = vld1q_s32(input);
+ in[7].val[1] = vld1q_s32(input + 4);
+ input += 8;
+ in[15].val[0] = vld1q_s32(input);
+ in[15].val[1] = vld1q_s32(input + 4);
+
+ // Transpose
+ transpose_s32_8x8(&in[0], &in[1], &in[2], &in[3], &in[4], &in[5], &in[6],
+ &in[7]);
+ transpose_s32_8x8(&in[8], &in[9], &in[10], &in[11], &in[12], &in[13], &in[14],
+ &in[15]);
+
+ x[0] = in[15];
+ x[1] = in[0];
+ x[2] = in[13];
+ x[3] = in[2];
+ x[4] = in[11];
+ x[5] = in[4];
+ x[6] = in[9];
+ x[7] = in[6];
+ x[8] = in[7];
+ x[9] = in[8];
+ x[10] = in[5];
+ x[11] = in[10];
+ x[12] = in[3];
+ x[13] = in[12];
+ x[14] = in[1];
+ x[15] = in[14];
+
+ // stage 1
+ highbd_iadst_butterfly(x[0], x[1], vget_low_s32(c_1_31_5_27), 0, 1, s0, s1);
+ highbd_iadst_butterfly(x[2], x[3], vget_high_s32(c_1_31_5_27), 0, 1, s2, s3);
+ highbd_iadst_butterfly(x[4], x[5], vget_low_s32(c_9_23_13_19), 0, 1, s4, s5);
+ highbd_iadst_butterfly(x[6], x[7], vget_high_s32(c_9_23_13_19), 0, 1, s6, s7);
+ highbd_iadst_butterfly(x[8], x[9], vget_low_s32(c_17_15_21_11), 0, 1, s8, s9);
+ highbd_iadst_butterfly(x[10], x[11], vget_high_s32(c_17_15_21_11), 0, 1, s10,
+ s11);
+ highbd_iadst_butterfly(x[12], x[13], vget_low_s32(c_25_7_29_3), 0, 1, s12,
+ s13);
+ highbd_iadst_butterfly(x[14], x[15], vget_high_s32(c_25_7_29_3), 0, 1, s14,
+ s15);
+
+ x[0] = highbd_add_dct_const_round_shift_low_8(s0, s8);
+ x[1] = highbd_add_dct_const_round_shift_low_8(s1, s9);
+ x[2] = highbd_add_dct_const_round_shift_low_8(s2, s10);
+ x[3] = highbd_add_dct_const_round_shift_low_8(s3, s11);
+ x[4] = highbd_add_dct_const_round_shift_low_8(s4, s12);
+ x[5] = highbd_add_dct_const_round_shift_low_8(s5, s13);
+ x[6] = highbd_add_dct_const_round_shift_low_8(s6, s14);
+ x[7] = highbd_add_dct_const_round_shift_low_8(s7, s15);
+ x[8] = highbd_sub_dct_const_round_shift_low_8(s0, s8);
+ x[9] = highbd_sub_dct_const_round_shift_low_8(s1, s9);
+ x[10] = highbd_sub_dct_const_round_shift_low_8(s2, s10);
+ x[11] = highbd_sub_dct_const_round_shift_low_8(s3, s11);
+ x[12] = highbd_sub_dct_const_round_shift_low_8(s4, s12);
+ x[13] = highbd_sub_dct_const_round_shift_low_8(s5, s13);
+ x[14] = highbd_sub_dct_const_round_shift_low_8(s6, s14);
+ x[15] = highbd_sub_dct_const_round_shift_low_8(s7, s15);
+
+ // stage 2
+ t[0] = x[0];
+ t[1] = x[1];
+ t[2] = x[2];
+ t[3] = x[3];
+ t[4] = x[4];
+ t[5] = x[5];
+ t[6] = x[6];
+ t[7] = x[7];
+ highbd_iadst_butterfly(x[8], x[9], vget_low_s32(c_4_28_20_12), 0, 1, s8, s9);
+ highbd_iadst_butterfly(x[10], x[11], vget_high_s32(c_4_28_20_12), 0, 1, s10,
+ s11);
+ highbd_iadst_butterfly(x[13], x[12], vget_low_s32(c_4_28_20_12), 1, 0, s13,
+ s12);
+ highbd_iadst_butterfly(x[15], x[14], vget_high_s32(c_4_28_20_12), 1, 0, s15,
+ s14);
+
+ x[0] = vaddq_s32_dual(t[0], t[4]);
+ x[1] = vaddq_s32_dual(t[1], t[5]);
+ x[2] = vaddq_s32_dual(t[2], t[6]);
+ x[3] = vaddq_s32_dual(t[3], t[7]);
+ x[4] = vsubq_s32_dual(t[0], t[4]);
+ x[5] = vsubq_s32_dual(t[1], t[5]);
+ x[6] = vsubq_s32_dual(t[2], t[6]);
+ x[7] = vsubq_s32_dual(t[3], t[7]);
+ x[8] = highbd_add_dct_const_round_shift_low_8(s8, s12);
+ x[9] = highbd_add_dct_const_round_shift_low_8(s9, s13);
+ x[10] = highbd_add_dct_const_round_shift_low_8(s10, s14);
+ x[11] = highbd_add_dct_const_round_shift_low_8(s11, s15);
+ x[12] = highbd_sub_dct_const_round_shift_low_8(s8, s12);
+ x[13] = highbd_sub_dct_const_round_shift_low_8(s9, s13);
+ x[14] = highbd_sub_dct_const_round_shift_low_8(s10, s14);
+ x[15] = highbd_sub_dct_const_round_shift_low_8(s11, s15);
+
+ // stage 3
+ t[0] = x[0];
+ t[1] = x[1];
+ t[2] = x[2];
+ t[3] = x[3];
+ highbd_iadst_butterfly(x[4], x[5], vget_high_s32(c_16_n16_8_24), 0, 1, s4,
+ s5);
+ highbd_iadst_butterfly(x[7], x[6], vget_high_s32(c_16_n16_8_24), 1, 0, s7,
+ s6);
+ t[8] = x[8];
+ t[9] = x[9];
+ t[10] = x[10];
+ t[11] = x[11];
+ highbd_iadst_butterfly(x[12], x[13], vget_high_s32(c_16_n16_8_24), 0, 1, s12,
+ s13);
+ highbd_iadst_butterfly(x[15], x[14], vget_high_s32(c_16_n16_8_24), 1, 0, s15,
+ s14);
+
+ x[0] = vaddq_s32_dual(t[0], t[2]);
+ x[1] = vaddq_s32_dual(t[1], t[3]);
+ x[2] = vsubq_s32_dual(t[0], t[2]);
+ x[3] = vsubq_s32_dual(t[1], t[3]);
+ x[4] = highbd_add_dct_const_round_shift_low_8(s4, s6);
+ x[5] = highbd_add_dct_const_round_shift_low_8(s5, s7);
+ x[6] = highbd_sub_dct_const_round_shift_low_8(s4, s6);
+ x[7] = highbd_sub_dct_const_round_shift_low_8(s5, s7);
+ x[8] = vaddq_s32_dual(t[8], t[10]);
+ x[9] = vaddq_s32_dual(t[9], t[11]);
+ x[10] = vsubq_s32_dual(t[8], t[10]);
+ x[11] = vsubq_s32_dual(t[9], t[11]);
+ x[12] = highbd_add_dct_const_round_shift_low_8(s12, s14);
+ x[13] = highbd_add_dct_const_round_shift_low_8(s13, s15);
+ x[14] = highbd_sub_dct_const_round_shift_low_8(s12, s14);
+ x[15] = highbd_sub_dct_const_round_shift_low_8(s13, s15);
+
+ // stage 4
+ {
+ const int32x4x2_t sum = vaddq_s32_dual(x[2], x[3]);
+ const int32x4x2_t sub = vsubq_s32_dual(x[2], x[3]);
+ highbd_iadst_half_butterfly(sum, vget_low_s32(c_16_n16_8_24), 1, x[2]);
+ highbd_iadst_half_butterfly(sub, vget_low_s32(c_16_n16_8_24), 0, x[3]);
+ }
+ {
+ const int32x4x2_t sum = vaddq_s32_dual(x[7], x[6]);
+ const int32x4x2_t sub = vsubq_s32_dual(x[7], x[6]);
+ highbd_iadst_half_butterfly(sum, vget_low_s32(c_16_n16_8_24), 0, x[6]);
+ highbd_iadst_half_butterfly(sub, vget_low_s32(c_16_n16_8_24), 0, x[7]);
+ }
+ {
+ const int32x4x2_t sum = vaddq_s32_dual(x[11], x[10]);
+ const int32x4x2_t sub = vsubq_s32_dual(x[11], x[10]);
+ highbd_iadst_half_butterfly(sum, vget_low_s32(c_16_n16_8_24), 0, x[10]);
+ highbd_iadst_half_butterfly(sub, vget_low_s32(c_16_n16_8_24), 0, x[11]);
+ }
+ {
+ const int32x4x2_t sum = vaddq_s32_dual(x[14], x[15]);
+ const int32x4x2_t sub = vsubq_s32_dual(x[14], x[15]);
+ highbd_iadst_half_butterfly(sum, vget_low_s32(c_16_n16_8_24), 1, x[14]);
+ highbd_iadst_half_butterfly(sub, vget_low_s32(c_16_n16_8_24), 0, x[15]);
+ }
+
+ out[0] = x[0];
+ out[1] = vnegq_s32_dual(x[8]);
+ out[2] = x[12];
+ out[3] = vnegq_s32_dual(x[4]);
+ out[4] = x[6];
+ out[5] = x[14];
+ out[6] = x[10];
+ out[7] = x[2];
+ out[8] = x[3];
+ out[9] = x[11];
+ out[10] = x[15];
+ out[11] = x[7];
+ out[12] = x[5];
+ out[13] = vnegq_s32_dual(x[13]);
+ out[14] = x[9];
+ out[15] = vnegq_s32_dual(x[1]);
+
+ if (output) {
+ highbd_idct16x16_store_pass1(out, output);
+ } else {
+ highbd_idct16x16_add_store(out, dest, stride, bd);
+ }
+}
+
+typedef void (*highbd_iht_1d)(const int32_t *input, int32_t *output,
+ uint16_t *dest, const int stride, const int bd);
+
+typedef struct {
+ highbd_iht_1d cols, rows; // vertical and horizontal
+} highbd_iht_2d;
+
+void vp9_highbd_iht16x16_256_add_neon(const tran_low_t *input, uint16_t *dest,
+ int stride, int tx_type, int bd) {
+ if (bd == 8) {
+ static const iht_2d IHT_16[] = {
+ { vpx_idct16x16_256_add_half1d,
+ vpx_idct16x16_256_add_half1d }, // DCT_DCT = 0
+ { vpx_iadst16x16_256_add_half1d,
+ vpx_idct16x16_256_add_half1d }, // ADST_DCT = 1
+ { vpx_idct16x16_256_add_half1d,
+ vpx_iadst16x16_256_add_half1d }, // DCT_ADST = 2
+ { vpx_iadst16x16_256_add_half1d,
+ vpx_iadst16x16_256_add_half1d } // ADST_ADST = 3
+ };
+ const iht_2d ht = IHT_16[tx_type];
+ int16_t row_output[16 * 16];
+
+ // pass 1
+ ht.rows(input, row_output, dest, stride, 1); // upper 8 rows
+ ht.rows(input + 8 * 16, row_output + 8, dest, stride, 1); // lower 8 rows
+
+ // pass 2
+ ht.cols(row_output, NULL, dest, stride, 1); // left 8 columns
+ ht.cols(row_output + 16 * 8, NULL, dest + 8, stride, 1); // right 8 columns
+ } else {
+ static const highbd_iht_2d IHT_16[] = {
+ { vpx_highbd_idct16x16_256_add_half1d,
+ vpx_highbd_idct16x16_256_add_half1d }, // DCT_DCT = 0
+ { vpx_highbd_iadst16_neon,
+ vpx_highbd_idct16x16_256_add_half1d }, // ADST_DCT = 1
+ { vpx_highbd_idct16x16_256_add_half1d,
+ vpx_highbd_iadst16_neon }, // DCT_ADST = 2
+ { vpx_highbd_iadst16_neon, vpx_highbd_iadst16_neon } // ADST_ADST = 3
+ };
+ const highbd_iht_2d ht = IHT_16[tx_type];
+ int32_t row_output[16 * 16];
+
+ // pass 1
+ ht.rows(input, row_output, dest, stride, bd); // upper 8 rows
+ ht.rows(input + 8 * 16, row_output + 8, dest, stride, bd); // lower 8 rows
+
+ // pass 2
+ ht.cols(row_output, NULL, dest, stride, bd); // left 8 columns
+ ht.cols(row_output + 8 * 16, NULL, dest + 8, stride,
+ bd); // right 8 columns
+ }
+}
static INLINE void highbd_iadst4(int32x4_t *const io) {
const int32_t sinpis[4] = { sinpi_1_9, sinpi_2_9, sinpi_3_9, sinpi_4_9 };
const int32x4_t sinpi = vld1q_s32(sinpis);
- int32x4_t s[8];
-
- s[0] = vmulq_lane_s32(io[0], vget_low_s32(sinpi), 0);
- s[1] = vmulq_lane_s32(io[0], vget_low_s32(sinpi), 1);
- s[2] = vmulq_lane_s32(io[1], vget_high_s32(sinpi), 0);
- s[3] = vmulq_lane_s32(io[2], vget_high_s32(sinpi), 1);
- s[4] = vmulq_lane_s32(io[2], vget_low_s32(sinpi), 0);
- s[5] = vmulq_lane_s32(io[3], vget_low_s32(sinpi), 1);
- s[6] = vmulq_lane_s32(io[3], vget_high_s32(sinpi), 1);
- s[7] = vsubq_s32(io[0], io[2]);
- s[7] = vaddq_s32(s[7], io[3]);
-
- s[0] = vaddq_s32(s[0], s[3]);
- s[0] = vaddq_s32(s[0], s[5]);
- s[1] = vsubq_s32(s[1], s[4]);
- s[1] = vsubq_s32(s[1], s[6]);
+ int64x2x2_t s[7], t[4];
+ int32x4_t s7;
+
+ s[0].val[0] = vmull_lane_s32(vget_low_s32(io[0]), vget_low_s32(sinpi), 0);
+ s[0].val[1] = vmull_lane_s32(vget_high_s32(io[0]), vget_low_s32(sinpi), 0);
+ s[1].val[0] = vmull_lane_s32(vget_low_s32(io[0]), vget_low_s32(sinpi), 1);
+ s[1].val[1] = vmull_lane_s32(vget_high_s32(io[0]), vget_low_s32(sinpi), 1);
+ s[2].val[0] = vmull_lane_s32(vget_low_s32(io[1]), vget_high_s32(sinpi), 0);
+ s[2].val[1] = vmull_lane_s32(vget_high_s32(io[1]), vget_high_s32(sinpi), 0);
+ s[3].val[0] = vmull_lane_s32(vget_low_s32(io[2]), vget_high_s32(sinpi), 1);
+ s[3].val[1] = vmull_lane_s32(vget_high_s32(io[2]), vget_high_s32(sinpi), 1);
+ s[4].val[0] = vmull_lane_s32(vget_low_s32(io[2]), vget_low_s32(sinpi), 0);
+ s[4].val[1] = vmull_lane_s32(vget_high_s32(io[2]), vget_low_s32(sinpi), 0);
+ s[5].val[0] = vmull_lane_s32(vget_low_s32(io[3]), vget_low_s32(sinpi), 1);
+ s[5].val[1] = vmull_lane_s32(vget_high_s32(io[3]), vget_low_s32(sinpi), 1);
+ s[6].val[0] = vmull_lane_s32(vget_low_s32(io[3]), vget_high_s32(sinpi), 1);
+ s[6].val[1] = vmull_lane_s32(vget_high_s32(io[3]), vget_high_s32(sinpi), 1);
+ s7 = vsubq_s32(io[0], io[2]);
+ s7 = vaddq_s32(s7, io[3]);
+
+ s[0].val[0] = vaddq_s64(s[0].val[0], s[3].val[0]);
+ s[0].val[1] = vaddq_s64(s[0].val[1], s[3].val[1]);
+ s[0].val[0] = vaddq_s64(s[0].val[0], s[5].val[0]);
+ s[0].val[1] = vaddq_s64(s[0].val[1], s[5].val[1]);
+ s[1].val[0] = vsubq_s64(s[1].val[0], s[4].val[0]);
+ s[1].val[1] = vsubq_s64(s[1].val[1], s[4].val[1]);
+ s[1].val[0] = vsubq_s64(s[1].val[0], s[6].val[0]);
+ s[1].val[1] = vsubq_s64(s[1].val[1], s[6].val[1]);
s[3] = s[2];
- s[2] = vmulq_lane_s32(s[7], vget_high_s32(sinpi), 0);
-
- io[0] = vaddq_s32(s[0], s[3]);
- io[1] = vaddq_s32(s[1], s[3]);
- io[2] = s[2];
- io[3] = vaddq_s32(s[0], s[1]);
- io[3] = vsubq_s32(io[3], s[3]);
- io[0] = vrshrq_n_s32(io[0], DCT_CONST_BITS);
- io[1] = vrshrq_n_s32(io[1], DCT_CONST_BITS);
- io[2] = vrshrq_n_s32(io[2], DCT_CONST_BITS);
- io[3] = vrshrq_n_s32(io[3], DCT_CONST_BITS);
+ s[2].val[0] = vmull_lane_s32(vget_low_s32(s7), vget_high_s32(sinpi), 0);
+ s[2].val[1] = vmull_lane_s32(vget_high_s32(s7), vget_high_s32(sinpi), 0);
+
+ t[0].val[0] = vaddq_s64(s[0].val[0], s[3].val[0]);
+ t[0].val[1] = vaddq_s64(s[0].val[1], s[3].val[1]);
+ t[1].val[0] = vaddq_s64(s[1].val[0], s[3].val[0]);
+ t[1].val[1] = vaddq_s64(s[1].val[1], s[3].val[1]);
+ t[2] = s[2];
+ t[3].val[0] = vaddq_s64(s[0].val[0], s[1].val[0]);
+ t[3].val[1] = vaddq_s64(s[0].val[1], s[1].val[1]);
+ t[3].val[0] = vsubq_s64(t[3].val[0], s[3].val[0]);
+ t[3].val[1] = vsubq_s64(t[3].val[1], s[3].val[1]);
+ io[0] = vcombine_s32(vrshrn_n_s64(t[0].val[0], DCT_CONST_BITS),
+ vrshrn_n_s64(t[0].val[1], DCT_CONST_BITS));
+ io[1] = vcombine_s32(vrshrn_n_s64(t[1].val[0], DCT_CONST_BITS),
+ vrshrn_n_s64(t[1].val[1], DCT_CONST_BITS));
+ io[2] = vcombine_s32(vrshrn_n_s64(t[2].val[0], DCT_CONST_BITS),
+ vrshrn_n_s64(t[2].val[1], DCT_CONST_BITS));
+ io[3] = vcombine_s32(vrshrn_n_s64(t[3].val[0], DCT_CONST_BITS),
+ vrshrn_n_s64(t[3].val[1], DCT_CONST_BITS));
}
void vp9_highbd_iht4x4_16_add_neon(const tran_low_t *input, uint16_t *dest,
return vcombine_s32(out_lo, out_hi);
}
-static INLINE void iadst8_bd10(int32x4_t *const io0, int32x4_t *const io1,
- int32x4_t *const io2, int32x4_t *const io3,
- int32x4_t *const io4, int32x4_t *const io5,
- int32x4_t *const io6, int32x4_t *const io7) {
- const int32x4_t c0 =
- create_s32x4_neon(cospi_2_64, cospi_30_64, cospi_10_64, cospi_22_64);
- const int32x4_t c1 =
- create_s32x4_neon(cospi_18_64, cospi_14_64, cospi_26_64, cospi_6_64);
- const int32x4_t c2 =
- create_s32x4_neon(cospi_16_64, 0, cospi_8_64, cospi_24_64);
- int32x4_t x[8], t[4];
- int32x4_t s[8];
-
- x[0] = *io7;
- x[1] = *io0;
- x[2] = *io5;
- x[3] = *io2;
- x[4] = *io3;
- x[5] = *io4;
- x[6] = *io1;
- x[7] = *io6;
-
- // stage 1
- iadst_butterfly_lane_0_1_bd10_neon(x[0], x[1], vget_low_s32(c0), &s[0],
- &s[1]);
- iadst_butterfly_lane_0_1_bd10_neon(x[2], x[3], vget_high_s32(c0), &s[2],
- &s[3]);
- iadst_butterfly_lane_0_1_bd10_neon(x[4], x[5], vget_low_s32(c1), &s[4],
- &s[5]);
- iadst_butterfly_lane_0_1_bd10_neon(x[6], x[7], vget_high_s32(c1), &s[6],
- &s[7]);
-
- x[0] = add_dct_const_round_shift_low_8_bd10(s[0], s[4]);
- x[1] = add_dct_const_round_shift_low_8_bd10(s[1], s[5]);
- x[2] = add_dct_const_round_shift_low_8_bd10(s[2], s[6]);
- x[3] = add_dct_const_round_shift_low_8_bd10(s[3], s[7]);
- x[4] = sub_dct_const_round_shift_low_8_bd10(s[0], s[4]);
- x[5] = sub_dct_const_round_shift_low_8_bd10(s[1], s[5]);
- x[6] = sub_dct_const_round_shift_low_8_bd10(s[2], s[6]);
- x[7] = sub_dct_const_round_shift_low_8_bd10(s[3], s[7]);
-
- // stage 2
- t[0] = x[0];
- t[1] = x[1];
- t[2] = x[2];
- t[3] = x[3];
- iadst_butterfly_lane_0_1_bd10_neon(x[4], x[5], vget_high_s32(c2), &s[4],
- &s[5]);
- iadst_butterfly_lane_1_0_bd10_neon(x[7], x[6], vget_high_s32(c2), &s[7],
- &s[6]);
-
- x[0] = vaddq_s32(t[0], t[2]);
- x[1] = vaddq_s32(t[1], t[3]);
- x[2] = vsubq_s32(t[0], t[2]);
- x[3] = vsubq_s32(t[1], t[3]);
- x[4] = add_dct_const_round_shift_low_8_bd10(s[4], s[6]);
- x[5] = add_dct_const_round_shift_low_8_bd10(s[5], s[7]);
- x[6] = sub_dct_const_round_shift_low_8_bd10(s[4], s[6]);
- x[7] = sub_dct_const_round_shift_low_8_bd10(s[5], s[7]);
-
- // stage 3
- iadst_half_butterfly_bd10_neon(x + 2, vget_low_s32(c2));
- iadst_half_butterfly_bd10_neon(x + 6, vget_low_s32(c2));
-
- *io0 = x[0];
- *io1 = vnegq_s32(x[4]);
- *io2 = x[6];
- *io3 = vnegq_s32(x[2]);
- *io4 = x[3];
- *io5 = vnegq_s32(x[7]);
- *io6 = x[5];
- *io7 = vnegq_s32(x[1]);
-}
-
-static INLINE void iadst8_bd12(int32x4_t *const io0, int32x4_t *const io1,
- int32x4_t *const io2, int32x4_t *const io3,
- int32x4_t *const io4, int32x4_t *const io5,
- int32x4_t *const io6, int32x4_t *const io7) {
+static INLINE void highbd_iadst8(int32x4_t *const io0, int32x4_t *const io1,
+ int32x4_t *const io2, int32x4_t *const io3,
+ int32x4_t *const io4, int32x4_t *const io5,
+ int32x4_t *const io6, int32x4_t *const io7) {
const int32x4_t c0 =
create_s32x4_neon(cospi_2_64, cospi_30_64, cospi_10_64, cospi_22_64);
const int32x4_t c1 =
const int32x4_t cospis1 =
vld1q_s32(kCospi32 + 4); // cospi 4, 12, 20, 28
- if (bd == 10) {
- idct8x8_64_half1d_bd10(cospis0, cospis1, &a[0], &a[1], &a[2], &a[3],
- &a[4], &a[5], &a[6], &a[7]);
- idct8x8_64_half1d_bd10(cospis0, cospis1, &a[8], &a[9], &a[10], &a[11],
- &a[12], &a[13], &a[14], &a[15]);
- transpose_s32_8x4(&a[0], &a[8], &a[1], &a[9], &a[2], &a[10], &a[3],
- &a[11]);
- iadst8_bd10(&a[0], &a[8], &a[1], &a[9], &a[2], &a[10], &a[3], &a[11]);
- transpose_s32_8x4(&a[4], &a[12], &a[5], &a[13], &a[6], &a[14], &a[7],
- &a[15]);
- iadst8_bd10(&a[4], &a[12], &a[5], &a[13], &a[6], &a[14], &a[7],
+ idct8x8_64_half1d_bd12(cospis0, cospis1, &a[0], &a[1], &a[2], &a[3],
+ &a[4], &a[5], &a[6], &a[7]);
+ idct8x8_64_half1d_bd12(cospis0, cospis1, &a[8], &a[9], &a[10], &a[11],
+ &a[12], &a[13], &a[14], &a[15]);
+ transpose_s32_8x4(&a[0], &a[8], &a[1], &a[9], &a[2], &a[10], &a[3],
+ &a[11]);
+ highbd_iadst8(&a[0], &a[8], &a[1], &a[9], &a[2], &a[10], &a[3], &a[11]);
+ transpose_s32_8x4(&a[4], &a[12], &a[5], &a[13], &a[6], &a[14], &a[7],
+ &a[15]);
+ highbd_iadst8(&a[4], &a[12], &a[5], &a[13], &a[6], &a[14], &a[7],
&a[15]);
- } else {
- idct8x8_64_half1d_bd12(cospis0, cospis1, &a[0], &a[1], &a[2], &a[3],
- &a[4], &a[5], &a[6], &a[7]);
- idct8x8_64_half1d_bd12(cospis0, cospis1, &a[8], &a[9], &a[10], &a[11],
- &a[12], &a[13], &a[14], &a[15]);
- transpose_s32_8x4(&a[0], &a[8], &a[1], &a[9], &a[2], &a[10], &a[3],
- &a[11]);
- iadst8_bd12(&a[0], &a[8], &a[1], &a[9], &a[2], &a[10], &a[3], &a[11]);
- transpose_s32_8x4(&a[4], &a[12], &a[5], &a[13], &a[6], &a[14], &a[7],
- &a[15]);
- iadst8_bd12(&a[4], &a[12], &a[5], &a[13], &a[6], &a[14], &a[7],
- &a[15]);
- }
break;
}
const int32x4_t cospis1 =
vld1q_s32(kCospi32 + 4); // cospi 4, 12, 20, 28
- if (bd == 10) {
- transpose_s32_8x4(&a[0], &a[1], &a[2], &a[3], &a[4], &a[5], &a[6],
- &a[7]);
- iadst8_bd10(&a[0], &a[1], &a[2], &a[3], &a[4], &a[5], &a[6], &a[7]);
- transpose_s32_8x4(&a[8], &a[9], &a[10], &a[11], &a[12], &a[13],
- &a[14], &a[15]);
- iadst8_bd10(&a[8], &a[9], &a[10], &a[11], &a[12], &a[13], &a[14],
+ transpose_s32_8x4(&a[0], &a[1], &a[2], &a[3], &a[4], &a[5], &a[6],
+ &a[7]);
+ highbd_iadst8(&a[0], &a[1], &a[2], &a[3], &a[4], &a[5], &a[6], &a[7]);
+ transpose_s32_8x4(&a[8], &a[9], &a[10], &a[11], &a[12], &a[13], &a[14],
+ &a[15]);
+ highbd_iadst8(&a[8], &a[9], &a[10], &a[11], &a[12], &a[13], &a[14],
&a[15]);
- idct8x8_64_half1d_bd10(cospis0, cospis1, &a[0], &a[8], &a[1], &a[9],
- &a[2], &a[10], &a[3], &a[11]);
- idct8x8_64_half1d_bd10(cospis0, cospis1, &a[4], &a[12], &a[5], &a[13],
- &a[6], &a[14], &a[7], &a[15]);
- } else {
- transpose_s32_8x4(&a[0], &a[1], &a[2], &a[3], &a[4], &a[5], &a[6],
- &a[7]);
- iadst8_bd12(&a[0], &a[1], &a[2], &a[3], &a[4], &a[5], &a[6], &a[7]);
- transpose_s32_8x4(&a[8], &a[9], &a[10], &a[11], &a[12], &a[13],
- &a[14], &a[15]);
- iadst8_bd12(&a[8], &a[9], &a[10], &a[11], &a[12], &a[13], &a[14],
- &a[15]);
- idct8x8_64_half1d_bd12(cospis0, cospis1, &a[0], &a[8], &a[1], &a[9],
- &a[2], &a[10], &a[3], &a[11]);
- idct8x8_64_half1d_bd12(cospis0, cospis1, &a[4], &a[12], &a[5], &a[13],
- &a[6], &a[14], &a[7], &a[15]);
- }
+ idct8x8_64_half1d_bd12(cospis0, cospis1, &a[0], &a[8], &a[1], &a[9],
+ &a[2], &a[10], &a[3], &a[11]);
+ idct8x8_64_half1d_bd12(cospis0, cospis1, &a[4], &a[12], &a[5], &a[13],
+ &a[6], &a[14], &a[7], &a[15]);
break;
}
default: {
assert(tx_type == ADST_ADST);
- if (bd == 10) {
- transpose_s32_8x4(&a[0], &a[1], &a[2], &a[3], &a[4], &a[5], &a[6],
- &a[7]);
- iadst8_bd10(&a[0], &a[1], &a[2], &a[3], &a[4], &a[5], &a[6], &a[7]);
- transpose_s32_8x4(&a[8], &a[9], &a[10], &a[11], &a[12], &a[13],
- &a[14], &a[15]);
- iadst8_bd10(&a[8], &a[9], &a[10], &a[11], &a[12], &a[13], &a[14],
+ transpose_s32_8x4(&a[0], &a[1], &a[2], &a[3], &a[4], &a[5], &a[6],
+ &a[7]);
+ highbd_iadst8(&a[0], &a[1], &a[2], &a[3], &a[4], &a[5], &a[6], &a[7]);
+ transpose_s32_8x4(&a[8], &a[9], &a[10], &a[11], &a[12], &a[13], &a[14],
+ &a[15]);
+ highbd_iadst8(&a[8], &a[9], &a[10], &a[11], &a[12], &a[13], &a[14],
&a[15]);
- transpose_s32_8x4(&a[0], &a[8], &a[1], &a[9], &a[2], &a[10], &a[3],
- &a[11]);
- iadst8_bd10(&a[0], &a[8], &a[1], &a[9], &a[2], &a[10], &a[3], &a[11]);
- transpose_s32_8x4(&a[4], &a[12], &a[5], &a[13], &a[6], &a[14], &a[7],
- &a[15]);
- iadst8_bd10(&a[4], &a[12], &a[5], &a[13], &a[6], &a[14], &a[7],
+ transpose_s32_8x4(&a[0], &a[8], &a[1], &a[9], &a[2], &a[10], &a[3],
+ &a[11]);
+ highbd_iadst8(&a[0], &a[8], &a[1], &a[9], &a[2], &a[10], &a[3], &a[11]);
+ transpose_s32_8x4(&a[4], &a[12], &a[5], &a[13], &a[6], &a[14], &a[7],
+ &a[15]);
+ highbd_iadst8(&a[4], &a[12], &a[5], &a[13], &a[6], &a[14], &a[7],
&a[15]);
- } else {
- transpose_s32_8x4(&a[0], &a[1], &a[2], &a[3], &a[4], &a[5], &a[6],
- &a[7]);
- iadst8_bd12(&a[0], &a[1], &a[2], &a[3], &a[4], &a[5], &a[6], &a[7]);
- transpose_s32_8x4(&a[8], &a[9], &a[10], &a[11], &a[12], &a[13],
- &a[14], &a[15]);
- iadst8_bd12(&a[8], &a[9], &a[10], &a[11], &a[12], &a[13], &a[14],
- &a[15]);
- transpose_s32_8x4(&a[0], &a[8], &a[1], &a[9], &a[2], &a[10], &a[3],
- &a[11]);
- iadst8_bd12(&a[0], &a[8], &a[1], &a[9], &a[2], &a[10], &a[3], &a[11]);
- transpose_s32_8x4(&a[4], &a[12], &a[5], &a[13], &a[6], &a[14], &a[7],
- &a[15]);
- iadst8_bd12(&a[4], &a[12], &a[5], &a[13], &a[6], &a[14], &a[7],
- &a[15]);
- }
break;
}
}
#include "vpx_dsp/arm/mem_neon.h"
#include "vpx_dsp/arm/transpose_neon.h"
-static void iadst16x16_256_add_half1d(const void *const input, int16_t *output,
- void *const dest, const int stride,
- const int highbd_flag) {
+void vpx_iadst16x16_256_add_half1d(const void *const input, int16_t *output,
+ void *const dest, const int stride,
+ const int highbd_flag) {
int16x8_t in[16], out[16];
const int16x4_t c_1_31_5_27 =
create_s16x4_neon(cospi_1_64, cospi_31_64, cospi_5_64, cospi_27_64);
}
}
-typedef void (*iht_1d)(const void *const input, int16_t *output,
- void *const dest, const int stride,
- const int highbd_flag);
-
-typedef struct {
- iht_1d cols, rows; // vertical and horizontal
-} iht_2d;
-
void vp9_iht16x16_256_add_neon(const tran_low_t *input, uint8_t *dest,
int stride, int tx_type) {
static const iht_2d IHT_16[] = {
{ vpx_idct16x16_256_add_half1d,
vpx_idct16x16_256_add_half1d }, // DCT_DCT = 0
- { iadst16x16_256_add_half1d,
+ { vpx_iadst16x16_256_add_half1d,
vpx_idct16x16_256_add_half1d }, // ADST_DCT = 1
{ vpx_idct16x16_256_add_half1d,
- iadst16x16_256_add_half1d }, // DCT_ADST = 2
- { iadst16x16_256_add_half1d, iadst16x16_256_add_half1d } // ADST_ADST = 3
+ vpx_iadst16x16_256_add_half1d }, // DCT_ADST = 2
+ { vpx_iadst16x16_256_add_half1d,
+ vpx_iadst16x16_256_add_half1d } // ADST_ADST = 3
};
const iht_2d ht = IHT_16[tx_type];
int16_t row_output[16 * 16];
io[7] = vnegq_s16(x[1]);
}
+void vpx_iadst16x16_256_add_half1d(const void *const input, int16_t *output,
+ void *const dest, const int stride,
+ const int highbd_flag);
+
+typedef void (*iht_1d)(const void *const input, int16_t *output,
+ void *const dest, const int stride,
+ const int highbd_flag);
+
+typedef struct {
+ iht_1d cols, rows; // vertical and horizontal
+} iht_2d;
+
#endif // VP9_COMMON_ARM_NEON_VP9_IHT_NEON_H_
add_proto qw/void vp9_highbd_iht16x16_256_add/, "const tran_low_t *input, uint16_t *output, int pitch, int tx_type, int bd";
if (vpx_config("CONFIG_EMULATE_HARDWARE") ne "yes") {
- specialize qw/vp9_highbd_iht4x4_16_add sse4_1/;
- specialize qw/vp9_highbd_iht8x8_64_add sse4_1/;
- specialize qw/vp9_highbd_iht16x16_256_add sse4_1/;
+ specialize qw/vp9_highbd_iht4x4_16_add neon sse4_1/;
+ specialize qw/vp9_highbd_iht8x8_64_add neon sse4_1/;
+ specialize qw/vp9_highbd_iht16x16_256_add neon sse4_1/;
}
}
// SVC: skip encoding of enhancement layer if the layer target bandwidth = 0.
if (cpi->use_svc && cpi->svc.spatial_layer_id > 0 &&
- !cpi->svc.rc_drop_superframe && cpi->oxcf.target_bandwidth == 0) {
+ cpi->oxcf.target_bandwidth == 0) {
cpi->svc.skip_enhancement_layer = 1;
vp9_rc_postencode_update_drop_frame(cpi);
- if (cpi->svc.spatial_layer_id == cpi->svc.number_spatial_layers - 1)
- vp9_inc_frame_in_layer(cpi);
+ vp9_inc_frame_in_layer(cpi);
cpi->ext_refresh_frame_flags_pending = 0;
return;
}
}
// For 1 pass CBR, check if we are dropping this frame.
- // For spatial layers, for now only check for frame-dropping on first spatial
- // layer, and if decision is to drop, we drop whole super-frame.
+ // For spatial layers, for now if we decide to drop current spatial
+ // layer then we will also drop all upper spatial layers.
+ // TODO(marpan): Allow for the case of dropping single layer only without
+ // dropping all upper layers.
if (oxcf->pass == 0 && oxcf->rc_mode == VPX_CBR &&
cm->frame_type != KEY_FRAME) {
if (vp9_rc_drop_frame(cpi) ||
- (is_one_pass_cbr_svc(cpi) && cpi->svc.rc_drop_superframe == 1)) {
+ (is_one_pass_cbr_svc(cpi) &&
+ cpi->svc.rc_drop_spatial_layer[cpi->svc.spatial_layer_id] == 1)) {
vp9_rc_postencode_update_drop_frame(cpi);
cpi->ext_refresh_frame_flags_pending = 0;
- cpi->svc.rc_drop_superframe = 1;
cpi->last_frame_dropped = 1;
- // TODO(marpan): Advancing the svc counters on dropped frames can break
- // the referencing scheme for the fixed svc patterns defined in
- // vp9_one_pass_cbr_svc_start_layer(). Look into fixing this issue, but
- // for now, don't advance the svc frame counters on dropped frame.
- // if (cpi->use_svc)
- // vp9_inc_frame_in_layer(cpi);
-
+ if (cpi->use_svc) {
+ int i;
+ // If we are dropping this spatial layer, then we will drop all
+ // upper spatial layers.
+ for (i = cpi->svc.spatial_layer_id; i < cpi->svc.number_spatial_layers;
+ i++)
+ cpi->svc.rc_drop_spatial_layer[i] = 1;
+ vp9_inc_frame_in_layer(cpi);
+ if (cpi->svc.rc_drop_spatial_layer[0] == 0)
+ cpi->svc.skip_enhancement_layer = 1;
+ }
return;
}
}
}
cpi->last_frame_dropped = 0;
+ cpi->svc.last_layer_encoded = cpi->svc.spatial_layer_id;
// Disable segmentation if it decrease rate/distortion ratio
if (cpi->oxcf.aq_mode == LOOKAHEAD_AQ)
int i;
if (is_two_pass_svc(cpi)) {
-#if CONFIG_SPATIAL_SVC
- vp9_svc_start_frame(cpi);
- // Use a small empty frame instead of a real frame
- if (cpi->svc.encode_empty_frame_state == ENCODING)
- source = &cpi->svc.empty_frame;
-#endif
if (oxcf->pass == 2) vp9_restore_layer_context(cpi);
} else if (is_one_pass_cbr_svc(cpi)) {
vp9_one_pass_cbr_svc_start_layer(cpi);
if ((source = vp9_lookahead_peek(cpi->lookahead, arf_src_index)) != NULL) {
cpi->alt_ref_source = source;
-#if CONFIG_SPATIAL_SVC
- if (is_two_pass_svc(cpi) && cpi->svc.spatial_layer_id > 0) {
- int i;
- // Reference a hidden frame from a lower layer
- for (i = cpi->svc.spatial_layer_id - 1; i >= 0; --i) {
- if (oxcf->ss_enable_auto_arf[i]) {
- cpi->gld_fb_idx = cpi->svc.layer_context[i].alt_ref_idx;
- break;
- }
- }
- }
- cpi->svc.layer_context[cpi->svc.spatial_layer_id].has_alt_frame = 1;
-#endif
#if !CONFIG_REALTIME_ONLY
if ((oxcf->mode != REALTIME) && (oxcf->arnr_max_frames > 0) &&
(oxcf->arnr_strength > 0)) {
#include "vpx/vpx_encoder.h"
#include "vpx/vpx_integer.h"
-#if CONFIG_SPATIAL_SVC
-#include "vpx/vp8cx.h"
-#endif
-
#ifdef __cplusplus
extern "C" {
#endif
RATE_CONTROL *const rc = &cpi->rc;
if (!oxcf->drop_frames_water_mark ||
(is_one_pass_cbr_svc(cpi) &&
- cpi->svc.spatial_layer_id > cpi->svc.first_spatial_layer_to_encode)) {
+ cpi->svc.rc_drop_spatial_layer[cpi->svc.spatial_layer_id] == 1)) {
return 0;
} else {
if (rc->buffer_level < 0) {
uint16_t *arrbuf2 = (uint16_t *)malloc(sizeof(uint16_t) * height2);
if (intbuf == NULL || tmpbuf == NULL || arrbuf == NULL || arrbuf2 == NULL)
goto Error;
+ assert(width > 0);
+ assert(height > 0);
+ assert(width2 > 0);
+ assert(height2 > 0);
for (i = 0; i < height; ++i) {
highbd_resize_multistep(CONVERT_TO_SHORTPTR(input + in_stride * i), width,
intbuf + width2 * i, width2, tmpbuf, bd);
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->use_base_mv = 0;
svc->use_partition_reuse = 0;
svc->current_superframe = 0;
svc->non_reference_frame = 0;
svc->skip_enhancement_layer = 0;
+ svc->last_layer_encoded = 0;
for (i = 0; i < REF_FRAMES; ++i) svc->ref_frame_index[i] = -1;
for (sl = 0; sl < oxcf->ss_number_layers; ++sl) {
+ svc->rc_drop_spatial_layer[sl] = 0;
svc->ext_frame_flags[sl] = 0;
svc->ext_lst_fb_idx[sl] = 0;
svc->ext_gld_fb_idx[sl] = 1;
}
}
- if (cpi->svc.spatial_layer_id == cpi->svc.first_spatial_layer_to_encode)
- cpi->svc.rc_drop_superframe = 0;
+ // Reset the drop flags for all spatial lauyers, on the base layer.
+ if (cpi->svc.spatial_layer_id == 0) {
+ int i;
+ for (i = 0; i < cpi->svc.number_spatial_layers; i++)
+ cpi->svc.rc_drop_spatial_layer[i] = 0;
+ }
lc = &cpi->svc.layer_context[cpi->svc.spatial_layer_id *
cpi->svc.number_temporal_layers +
return 0;
}
-#if CONFIG_SPATIAL_SVC
-#define SMALL_FRAME_FB_IDX 7
-
-int vp9_svc_start_frame(VP9_COMP *const cpi) {
- int width = 0, height = 0;
- LAYER_CONTEXT *lc;
- struct lookahead_entry *buf;
- int count = 1 << (cpi->svc.number_temporal_layers - 1);
-
- cpi->svc.spatial_layer_id = cpi->svc.spatial_layer_to_encode;
- lc = &cpi->svc.layer_context[cpi->svc.spatial_layer_id];
-
- cpi->svc.temporal_layer_id = 0;
- while ((lc->current_video_frame_in_layer % count) != 0) {
- ++cpi->svc.temporal_layer_id;
- count >>= 1;
- }
-
- cpi->ref_frame_flags = VP9_ALT_FLAG | VP9_GOLD_FLAG | VP9_LAST_FLAG;
-
- cpi->lst_fb_idx = cpi->svc.spatial_layer_id;
-
- if (cpi->svc.spatial_layer_id == 0)
- cpi->gld_fb_idx =
- (lc->gold_ref_idx >= 0) ? lc->gold_ref_idx : cpi->lst_fb_idx;
- else
- cpi->gld_fb_idx = cpi->svc.spatial_layer_id - 1;
-
- if (lc->current_video_frame_in_layer == 0) {
- if (cpi->svc.spatial_layer_id >= 2) {
- cpi->alt_fb_idx = cpi->svc.spatial_layer_id - 2;
- } else {
- cpi->alt_fb_idx = cpi->lst_fb_idx;
- cpi->ref_frame_flags &= (~VP9_LAST_FLAG & ~VP9_ALT_FLAG);
- }
- } else {
- if (cpi->oxcf.ss_enable_auto_arf[cpi->svc.spatial_layer_id]) {
- cpi->alt_fb_idx = lc->alt_ref_idx;
- if (!lc->has_alt_frame) cpi->ref_frame_flags &= (~VP9_ALT_FLAG);
- } else {
- // Find a proper alt_fb_idx for layers that don't have alt ref frame
- if (cpi->svc.spatial_layer_id == 0) {
- cpi->alt_fb_idx = cpi->lst_fb_idx;
- } else {
- LAYER_CONTEXT *lc_lower =
- &cpi->svc.layer_context[cpi->svc.spatial_layer_id - 1];
-
- if (cpi->oxcf.ss_enable_auto_arf[cpi->svc.spatial_layer_id - 1] &&
- lc_lower->alt_ref_source != NULL)
- cpi->alt_fb_idx = lc_lower->alt_ref_idx;
- else if (cpi->svc.spatial_layer_id >= 2)
- cpi->alt_fb_idx = cpi->svc.spatial_layer_id - 2;
- else
- cpi->alt_fb_idx = cpi->lst_fb_idx;
- }
- }
- }
-
- get_layer_resolution(cpi->oxcf.width, cpi->oxcf.height,
- lc->scaling_factor_num, lc->scaling_factor_den, &width,
- &height);
-
- // Workaround for multiple frame contexts. In some frames we can't use prev_mi
- // since its previous frame could be changed during decoding time. The idea is
- // we put a empty invisible frame in front of them, then we will not use
- // prev_mi when encoding these frames.
-
- buf = vp9_lookahead_peek(cpi->lookahead, 0);
- if (cpi->oxcf.error_resilient_mode == 0 && cpi->oxcf.pass == 2 &&
- cpi->svc.encode_empty_frame_state == NEED_TO_ENCODE &&
- lc->rc.frames_to_key != 0 &&
- !(buf != NULL && (buf->flags & VPX_EFLAG_FORCE_KF))) {
- if ((cpi->svc.number_temporal_layers > 1 &&
- cpi->svc.temporal_layer_id < cpi->svc.number_temporal_layers - 1) ||
- (cpi->svc.number_spatial_layers > 1 &&
- cpi->svc.spatial_layer_id == 0)) {
- struct lookahead_entry *buf = vp9_lookahead_peek(cpi->lookahead, 0);
-
- if (buf != NULL) {
- cpi->svc.empty_frame.ts_start = buf->ts_start;
- cpi->svc.empty_frame.ts_end = buf->ts_end;
- cpi->svc.encode_empty_frame_state = ENCODING;
- cpi->common.show_frame = 0;
- cpi->ref_frame_flags = 0;
- cpi->common.frame_type = INTER_FRAME;
- cpi->lst_fb_idx = cpi->gld_fb_idx = cpi->alt_fb_idx =
- SMALL_FRAME_FB_IDX;
-
- if (cpi->svc.encode_intra_empty_frame != 0) cpi->common.intra_only = 1;
-
- width = SMALL_FRAME_WIDTH;
- height = SMALL_FRAME_HEIGHT;
- }
- }
- }
-
- cpi->oxcf.worst_allowed_q = vp9_quantizer_to_qindex(lc->max_q);
- cpi->oxcf.best_allowed_q = vp9_quantizer_to_qindex(lc->min_q);
-
- vp9_change_config(cpi, &cpi->oxcf);
-
- if (vp9_set_size_literal(cpi, width, height) != 0)
- return VPX_CODEC_INVALID_PARAM;
-
- vp9_set_high_precision_mv(cpi, 1);
-
- cpi->alt_ref_source = get_layer_context(cpi)->alt_ref_source;
-
- return 0;
-}
-
-#undef SMALL_FRAME_FB_IDX
-#endif // CONFIG_SPATIAL_SVC
-
struct lookahead_entry *vp9_svc_lookahead_pop(VP9_COMP *const cpi,
struct lookahead_ctx *ctx,
int drain) {
int spatial_layer_to_encode;
int first_spatial_layer_to_encode;
- int rc_drop_superframe;
+ int rc_drop_spatial_layer[VPX_MAX_LAYERS];
// Workaround for multiple frame contexts
enum { ENCODED = 0, ENCODING, NEED_TO_ENCODE } encode_empty_frame_state;
int skip_enhancement_layer;
int lower_layer_qindex;
+
+ int last_layer_encoded;
} SVC;
struct VP9_COMP;
else
VP9_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/vp9_highbd_iht4x4_add_neon.c
VP9_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/vp9_highbd_iht8x8_add_neon.c
+VP9_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/vp9_highbd_iht16x16_add_neon.c
VP9_COMMON_SRCS-$(HAVE_SSE4_1) += common/x86/vp9_highbd_iht4x4_add_sse4.c
VP9_COMMON_SRCS-$(HAVE_SSE4_1) += common/x86/vp9_highbd_iht8x8_add_sse4.c
VP9_COMMON_SRCS-$(HAVE_SSE4_1) += common/x86/vp9_highbd_iht16x16_add_sse4.c
ERROR("ts_rate_decimator factors are not powers of 2");
}
-#if CONFIG_SPATIAL_SVC
-
- if ((cfg->ss_number_layers > 1 || cfg->ts_number_layers > 1) &&
- cfg->g_pass == VPX_RC_LAST_PASS) {
- unsigned int i, alt_ref_sum = 0;
- for (i = 0; i < cfg->ss_number_layers; ++i) {
- if (cfg->ss_enable_auto_alt_ref[i]) ++alt_ref_sum;
- }
- if (alt_ref_sum > REF_FRAMES - cfg->ss_number_layers)
- ERROR("Not enough ref buffers for svc alt ref frames");
- if (cfg->ss_number_layers * cfg->ts_number_layers > 3 &&
- cfg->g_error_resilient == 0)
- ERROR("Multiple frame context are not supported for more than 3 layers");
- }
-#endif
-
// VP9 does not support a lower bound on the keyframe interval in
// automatic keyframe placement mode.
if (cfg->kf_mode != VPX_KF_DISABLED && cfg->kf_min_dist != cfg->kf_max_dist &&
oxcf->motion_vector_unit_test = extra_cfg->motion_vector_unit_test;
for (sl = 0; sl < oxcf->ss_number_layers; ++sl) {
-#if CONFIG_SPATIAL_SVC
- oxcf->ss_enable_auto_arf[sl] = cfg->ss_enable_auto_alt_ref[sl];
-#endif
for (tl = 0; tl < oxcf->ts_number_layers; ++tl) {
oxcf->layer_target_bitrate[sl * oxcf->ts_number_layers + tl] =
1000 * cfg->layer_target_bitrate[sl * oxcf->ts_number_layers + tl];
}
if (oxcf->ss_number_layers == 1 && oxcf->pass != 0) {
oxcf->ss_target_bitrate[0] = (int)oxcf->target_bandwidth;
-#if CONFIG_SPATIAL_SVC
- oxcf->ss_enable_auto_arf[0] = extra_cfg->enable_auto_alt_ref;
-#endif
}
if (oxcf->ts_number_layers > 1) {
for (tl = 0; tl < VPX_TS_MAX_LAYERS; ++tl) {
if (size || (cpi->use_svc && cpi->svc.skip_enhancement_layer)) {
vpx_codec_cx_pkt_t pkt;
-#if CONFIG_SPATIAL_SVC
- if (cpi->use_svc)
- cpi->svc
- .layer_context[cpi->svc.spatial_layer_id *
- cpi->svc.number_temporal_layers]
- .layer_size += size;
-#endif
-
// Pack invisible frames with the next visible frame
if (!cpi->common.show_frame ||
(cpi->use_svc &&
cpi->svc.spatial_layer_id < cpi->svc.number_spatial_layers - 1)) {
if (ctx->pending_cx_data == 0) ctx->pending_cx_data = cx_data;
ctx->pending_cx_data_sz += size;
- ctx->pending_frame_sizes[ctx->pending_frame_count++] = size;
+ if (size) ctx->pending_frame_sizes[ctx->pending_frame_count++] = size;
ctx->pending_frame_magnitude |= size;
cx_data += size;
cx_data_sz -= size;
pkt.data.frame.width[cpi->svc.spatial_layer_id] = cpi->common.width;
pkt.data.frame.height[cpi->svc.spatial_layer_id] = cpi->common.height;
+ pkt.data.frame.last_spatial_layer_encoded =
+ cpi->svc.last_layer_encoded;
if (ctx->output_cx_pkt_cb.output_cx_pkt) {
pkt.kind = VPX_CODEC_CX_FRAME_PKT;
pkt.data.frame.flags = get_frame_pkt_flags(cpi, lib_flags);
pkt.data.frame.width[cpi->svc.spatial_layer_id] = cpi->common.width;
pkt.data.frame.height[cpi->svc.spatial_layer_id] = cpi->common.height;
+ pkt.data.frame.last_spatial_layer_encoded = cpi->svc.last_layer_encoded;
if (ctx->pending_cx_data) {
if (size) ctx->pending_frame_sizes[ctx->pending_frame_count++] = size;
cx_data += size;
cx_data_sz -= size;
-#if CONFIG_SPATIAL_SVC && defined(VPX_TEST_SPATIAL_SVC)
- if (cpi->use_svc && !ctx->output_cx_pkt_cb.output_cx_pkt) {
- vpx_codec_cx_pkt_t pkt_sizes, pkt_psnr;
- int sl;
- vp9_zero(pkt_sizes);
- vp9_zero(pkt_psnr);
- pkt_sizes.kind = VPX_CODEC_SPATIAL_SVC_LAYER_SIZES;
- pkt_psnr.kind = VPX_CODEC_SPATIAL_SVC_LAYER_PSNR;
- for (sl = 0; sl < cpi->svc.number_spatial_layers; ++sl) {
- LAYER_CONTEXT *lc =
- &cpi->svc.layer_context[sl * cpi->svc.number_temporal_layers];
- pkt_sizes.data.layer_sizes[sl] = lc->layer_size;
- pkt_psnr.data.layer_psnr[sl] = lc->psnr_pkt;
- lc->layer_size = 0;
- }
-
- vpx_codec_pkt_list_add(&ctx->pkt_list.head, &pkt_sizes);
-
- vpx_codec_pkt_list_add(&ctx->pkt_list.head, &pkt_psnr);
- }
-#endif
if (is_one_pass_cbr_svc(cpi) &&
(cpi->svc.spatial_layer_id == cpi->svc.number_spatial_layers - 1)) {
// Encoded all spatial layers; exit loop.
text vpx_svc_dump_statistics
text vpx_svc_encode
-text vpx_svc_get_message
text vpx_svc_init
text vpx_svc_release
text vpx_svc_set_options
return (const SvcInternal_t *)svc_ctx->internal;
}
-static void svc_log_reset(SvcContext *svc_ctx) {
- SvcInternal_t *const si = (SvcInternal_t *)svc_ctx->internal;
- si->message_buffer[0] = '\0';
-}
-
static int svc_log(SvcContext *svc_ctx, SVC_LOG_LEVEL level, const char *fmt,
...) {
char buf[512];
int retval = 0;
va_list ap;
- SvcInternal_t *const si = get_svc_internal(svc_ctx);
if (level > svc_ctx->log_level) {
return retval;
retval = vsnprintf(buf, sizeof(buf), fmt, ap);
va_end(ap);
- if (svc_ctx->log_print) {
- printf("%s", buf);
- } else {
- strncat(si->message_buffer, buf,
- sizeof(si->message_buffer) - strlen(si->message_buffer) - 1);
- }
+ printf("%s", buf);
- if (level == SVC_LOG_ERROR) {
- si->codec_ctx->err_detail = si->message_buffer;
- }
return retval;
}
return VPX_CODEC_INVALID_PARAM;
}
-#if CONFIG_SPATIAL_SVC
- for (i = 0; i < svc_ctx->spatial_layers; ++i)
- enc_cfg->ss_enable_auto_alt_ref[i] = si->enable_auto_alt_ref[i];
-#endif
-
if (svc_ctx->temporal_layers > 1) {
int i;
for (i = 0; i < svc_ctx->temporal_layers; ++i) {
return VPX_CODEC_INVALID_PARAM;
}
- svc_log_reset(svc_ctx);
-
res =
vpx_codec_encode(codec_ctx, rawimg, pts, (uint32_t)duration, 0, deadline);
if (res != VPX_CODEC_OK) {
iter = NULL;
while ((cx_pkt = vpx_codec_get_cx_data(codec_ctx, &iter))) {
switch (cx_pkt->kind) {
-#if CONFIG_SPATIAL_SVC && defined(VPX_TEST_SPATIAL_SVC)
- case VPX_CODEC_SPATIAL_SVC_LAYER_PSNR: {
- int i;
- for (i = 0; i < svc_ctx->spatial_layers; ++i) {
- int j;
- svc_log(svc_ctx, SVC_LOG_DEBUG,
- "SVC frame: %d, layer: %d, PSNR(Total/Y/U/V): "
- "%2.3f %2.3f %2.3f %2.3f \n",
- si->psnr_pkt_received, i, cx_pkt->data.layer_psnr[i].psnr[0],
- cx_pkt->data.layer_psnr[i].psnr[1],
- cx_pkt->data.layer_psnr[i].psnr[2],
- cx_pkt->data.layer_psnr[i].psnr[3]);
- svc_log(svc_ctx, SVC_LOG_DEBUG,
- "SVC frame: %d, layer: %d, SSE(Total/Y/U/V): "
- "%2.3f %2.3f %2.3f %2.3f \n",
- si->psnr_pkt_received, i, cx_pkt->data.layer_psnr[i].sse[0],
- cx_pkt->data.layer_psnr[i].sse[1],
- cx_pkt->data.layer_psnr[i].sse[2],
- cx_pkt->data.layer_psnr[i].sse[3]);
-
- for (j = 0; j < COMPONENTS; ++j) {
- si->psnr_sum[i][j] += cx_pkt->data.layer_psnr[i].psnr[j];
- si->sse_sum[i][j] += cx_pkt->data.layer_psnr[i].sse[j];
- }
- }
- ++si->psnr_pkt_received;
- break;
- }
- case VPX_CODEC_SPATIAL_SVC_LAYER_SIZES: {
- int i;
- for (i = 0; i < svc_ctx->spatial_layers; ++i)
- si->bytes_sum[i] += cx_pkt->data.layer_sizes[i];
- break;
- }
-#endif
case VPX_CODEC_PSNR_PKT: {
-#if CONFIG_SPATIAL_SVC && defined(VPX_TEST_SPATIAL_SVC)
- int j;
- svc_log(svc_ctx, SVC_LOG_DEBUG,
- "frame: %d, layer: %d, PSNR(Total/Y/U/V): "
- "%2.3f %2.3f %2.3f %2.3f \n",
- si->psnr_pkt_received, 0, cx_pkt->data.layer_psnr[0].psnr[0],
- cx_pkt->data.layer_psnr[0].psnr[1],
- cx_pkt->data.layer_psnr[0].psnr[2],
- cx_pkt->data.layer_psnr[0].psnr[3]);
- for (j = 0; j < COMPONENTS; ++j) {
- si->psnr_sum[0][j] += cx_pkt->data.layer_psnr[0].psnr[j];
- si->sse_sum[0][j] += cx_pkt->data.layer_psnr[0].sse[j];
- }
-#endif
}
++si->psnr_pkt_received;
break;
return VPX_CODEC_OK;
}
-const char *vpx_svc_get_message(const SvcContext *svc_ctx) {
- const SvcInternal_t *const si = get_const_svc_internal(svc_ctx);
- if (svc_ctx == NULL || si == NULL) return NULL;
- return si->message_buffer;
-}
-
static double calc_psnr(double d) {
if (d == 0) return 100;
return -10.0 * log(d) / log(10.0);
}
// dump accumulated statistics and reset accumulated values
-const char *vpx_svc_dump_statistics(SvcContext *svc_ctx) {
+void vpx_svc_dump_statistics(SvcContext *svc_ctx) {
int number_of_frames;
int i, j;
uint32_t bytes_total = 0;
double y_scale;
SvcInternal_t *const si = get_svc_internal(svc_ctx);
- if (svc_ctx == NULL || si == NULL) return NULL;
-
- svc_log_reset(svc_ctx);
+ if (svc_ctx == NULL || si == NULL) return;
number_of_frames = si->psnr_pkt_received;
- if (number_of_frames <= 0) return vpx_svc_get_message(svc_ctx);
+ if (number_of_frames <= 0) return;
svc_log(svc_ctx, SVC_LOG_INFO, "\n");
for (i = 0; i < svc_ctx->spatial_layers; ++i) {
si->psnr_pkt_received = 0;
svc_log(svc_ctx, SVC_LOG_INFO, "Total Bytes=[%u]\n", bytes_total);
- return vpx_svc_get_message(svc_ctx);
}
void vpx_svc_release(SvcContext *svc_ctx) {
int temporal_layers; // number of temporal layers
int temporal_layering_mode;
SVC_LOG_LEVEL log_level; // amount of information to display
- int log_print; // when set, printf log messages instead of returning the
- // message with svc_get_message
- int output_rc_stat; // for outputting rc stats
- int speed; // speed setting for codec
+ int output_rc_stat; // for outputting rc stats
+ int speed; // speed setting for codec
int threads;
int aqmode; // turns on aq-mode=3 (cyclic_refresh): 0=off, 1=on.
// private storage for vpx_svc_encode
int layer;
int use_multiple_frame_contexts;
- char message_buffer[2048];
vpx_codec_ctx_t *codec_ctx;
} SvcInternal_t;
/**
* dump accumulated statistics and reset accumulated values
*/
-const char *vpx_svc_dump_statistics(SvcContext *svc_ctx);
-
-/**
- * get status message from previous encode
- */
-const char *vpx_svc_get_message(const SvcContext *svc_ctx);
+void vpx_svc_dump_statistics(SvcContext *svc_ctx);
#ifdef __cplusplus
} // extern "C"
API_SRCS-$(CONFIG_VP8_ENCODER) += vp8cx.h
API_DOC_SRCS-$(CONFIG_VP8_ENCODER) += vp8.h
API_DOC_SRCS-$(CONFIG_VP8_ENCODER) += vp8cx.h
-ifeq ($(CONFIG_VP9_ENCODER),yes)
- API_SRCS-$(CONFIG_SPATIAL_SVC) += src/svc_encodeframe.c
- API_SRCS-$(CONFIG_SPATIAL_SVC) += svc_context.h
-endif
+API_SRCS-$(CONFIG_VP9_ENCODER) += src/svc_encodeframe.c
+API_SRCS-$(CONFIG_VP9_ENCODER) += svc_context.h
API_SRCS-$(CONFIG_VP8_DECODER) += vp8.h
API_SRCS-$(CONFIG_VP8_DECODER) += vp8dx.h
* extend this list to provide additional functionality.
*/
enum vpx_codec_cx_pkt_kind {
- VPX_CODEC_CX_FRAME_PKT, /**< Compressed video frame */
- VPX_CODEC_STATS_PKT, /**< Two-pass statistics for this frame */
- VPX_CODEC_FPMB_STATS_PKT, /**< first pass mb statistics for this frame */
- VPX_CODEC_PSNR_PKT, /**< PSNR statistics for this frame */
-// Spatial SVC is still experimental and may be removed.
-#if defined(VPX_TEST_SPATIAL_SVC)
- VPX_CODEC_SPATIAL_SVC_LAYER_SIZES, /**< Sizes for each layer in this frame*/
- VPX_CODEC_SPATIAL_SVC_LAYER_PSNR, /**< PSNR for each layer in this frame*/
-#endif
+ VPX_CODEC_CX_FRAME_PKT, /**< Compressed video frame */
+ VPX_CODEC_STATS_PKT, /**< Two-pass statistics for this frame */
+ VPX_CODEC_FPMB_STATS_PKT, /**< first pass mb statistics for this frame */
+ VPX_CODEC_PSNR_PKT, /**< PSNR statistics for this frame */
VPX_CODEC_CUSTOM_PKT = 256 /**< Algorithm extensions */
};
* first one.*/
unsigned int width[VPX_SS_MAX_LAYERS]; /**< frame width */
unsigned int height[VPX_SS_MAX_LAYERS]; /**< frame height */
+ /*!\brief Last spatial layer frame in this packet. VP8 will always be set
+ * to 0.*/
+ unsigned int last_spatial_layer_encoded;
} frame; /**< data for compressed frame packet */
vpx_fixed_buf_t twopass_stats; /**< data for two-pass packet */
vpx_fixed_buf_t firstpass_mb_stats; /**< first pass mb packet */
double psnr[4]; /**< PSNR, total/y/u/v */
} psnr; /**< data for PSNR packet */
vpx_fixed_buf_t raw; /**< data for arbitrary packets */
-// Spatial SVC is still experimental and may be removed.
-#if defined(VPX_TEST_SPATIAL_SVC)
- size_t layer_sizes[VPX_SS_MAX_LAYERS];
- struct vpx_psnr_pkt layer_psnr[VPX_SS_MAX_LAYERS];
-#endif
/* This packet size is fixed to allow codecs to extend this
* interface without having to manage storage for raw packets,
#include <arm_neon.h>
#include "./vpx_dsp_rtcd.h"
+#include "vpx_dsp/arm/highbd_idct_neon.h"
#include "vpx_dsp/arm/idct_neon.h"
#include "vpx_dsp/inv_txfm.h"
out[15] = vsubq_s32(step2[0], step2[15]);
}
-static INLINE void highbd_idct16x16_store_pass1(const int32x4x2_t *const out,
- int32_t *output) {
- // Save the result into output
- vst1q_s32(output + 0, out[0].val[0]);
- vst1q_s32(output + 4, out[0].val[1]);
- output += 16;
- vst1q_s32(output + 0, out[1].val[0]);
- vst1q_s32(output + 4, out[1].val[1]);
- output += 16;
- vst1q_s32(output + 0, out[2].val[0]);
- vst1q_s32(output + 4, out[2].val[1]);
- output += 16;
- vst1q_s32(output + 0, out[3].val[0]);
- vst1q_s32(output + 4, out[3].val[1]);
- output += 16;
- vst1q_s32(output + 0, out[4].val[0]);
- vst1q_s32(output + 4, out[4].val[1]);
- output += 16;
- vst1q_s32(output + 0, out[5].val[0]);
- vst1q_s32(output + 4, out[5].val[1]);
- output += 16;
- vst1q_s32(output + 0, out[6].val[0]);
- vst1q_s32(output + 4, out[6].val[1]);
- output += 16;
- vst1q_s32(output + 0, out[7].val[0]);
- vst1q_s32(output + 4, out[7].val[1]);
- output += 16;
- vst1q_s32(output + 0, out[8].val[0]);
- vst1q_s32(output + 4, out[8].val[1]);
- output += 16;
- vst1q_s32(output + 0, out[9].val[0]);
- vst1q_s32(output + 4, out[9].val[1]);
- output += 16;
- vst1q_s32(output + 0, out[10].val[0]);
- vst1q_s32(output + 4, out[10].val[1]);
- output += 16;
- vst1q_s32(output + 0, out[11].val[0]);
- vst1q_s32(output + 4, out[11].val[1]);
- output += 16;
- vst1q_s32(output + 0, out[12].val[0]);
- vst1q_s32(output + 4, out[12].val[1]);
- output += 16;
- vst1q_s32(output + 0, out[13].val[0]);
- vst1q_s32(output + 4, out[13].val[1]);
- output += 16;
- vst1q_s32(output + 0, out[14].val[0]);
- vst1q_s32(output + 4, out[14].val[1]);
- output += 16;
- vst1q_s32(output + 0, out[15].val[0]);
- vst1q_s32(output + 4, out[15].val[1]);
-}
-
-static void vpx_highbd_idct16x16_256_add_half1d(const int32_t *input,
- int32_t *output, uint16_t *dest,
- const int stride,
- const int bd) {
+void vpx_highbd_idct16x16_256_add_half1d(const int32_t *input, int32_t *output,
+ uint16_t *dest, const int stride,
+ const int bd) {
const int32x4_t cospi_0_8_16_24 = vld1q_s32(kCospi32 + 0);
const int32x4_t cospi_4_12_20N_28 = vld1q_s32(kCospi32 + 4);
const int32x4_t cospi_2_30_10_22 = vld1q_s32(kCospi32 + 8);
#include "./vpx_config.h"
#include "./vpx_dsp_rtcd.h"
+#include "vpx_dsp/arm/highbd_idct_neon.h"
#include "vpx_dsp/arm/idct_neon.h"
#include "vpx_dsp/arm/transpose_neon.h"
#include "vpx_dsp/txfm_common.h"
#include "./vpx_config.h"
#include "./vpx_dsp_rtcd.h"
+#include "vpx_dsp/arm/highbd_idct_neon.h"
#include "vpx_dsp/arm/idct_neon.h"
#include "vpx_dsp/arm/transpose_neon.h"
#include "vpx_dsp/txfm_common.h"
*io7 = vsubq_s32(step1[0], step2[7]);
}
+static INLINE void highbd_idct16x16_store_pass1(const int32x4x2_t *const out,
+ int32_t *output) {
+ // Save the result into output
+ vst1q_s32(output + 0, out[0].val[0]);
+ vst1q_s32(output + 4, out[0].val[1]);
+ output += 16;
+ vst1q_s32(output + 0, out[1].val[0]);
+ vst1q_s32(output + 4, out[1].val[1]);
+ output += 16;
+ vst1q_s32(output + 0, out[2].val[0]);
+ vst1q_s32(output + 4, out[2].val[1]);
+ output += 16;
+ vst1q_s32(output + 0, out[3].val[0]);
+ vst1q_s32(output + 4, out[3].val[1]);
+ output += 16;
+ vst1q_s32(output + 0, out[4].val[0]);
+ vst1q_s32(output + 4, out[4].val[1]);
+ output += 16;
+ vst1q_s32(output + 0, out[5].val[0]);
+ vst1q_s32(output + 4, out[5].val[1]);
+ output += 16;
+ vst1q_s32(output + 0, out[6].val[0]);
+ vst1q_s32(output + 4, out[6].val[1]);
+ output += 16;
+ vst1q_s32(output + 0, out[7].val[0]);
+ vst1q_s32(output + 4, out[7].val[1]);
+ output += 16;
+ vst1q_s32(output + 0, out[8].val[0]);
+ vst1q_s32(output + 4, out[8].val[1]);
+ output += 16;
+ vst1q_s32(output + 0, out[9].val[0]);
+ vst1q_s32(output + 4, out[9].val[1]);
+ output += 16;
+ vst1q_s32(output + 0, out[10].val[0]);
+ vst1q_s32(output + 4, out[10].val[1]);
+ output += 16;
+ vst1q_s32(output + 0, out[11].val[0]);
+ vst1q_s32(output + 4, out[11].val[1]);
+ output += 16;
+ vst1q_s32(output + 0, out[12].val[0]);
+ vst1q_s32(output + 4, out[12].val[1]);
+ output += 16;
+ vst1q_s32(output + 0, out[13].val[0]);
+ vst1q_s32(output + 4, out[13].val[1]);
+ output += 16;
+ vst1q_s32(output + 0, out[14].val[0]);
+ vst1q_s32(output + 4, out[14].val[1]);
+ output += 16;
+ vst1q_s32(output + 0, out[15].val[0]);
+ vst1q_s32(output + 4, out[15].val[1]);
+}
+
+static INLINE void highbd_idct16x16_add_store(const int32x4x2_t *const out,
+ uint16_t *dest, const int stride,
+ const int bd) {
+ // Add the result to dest
+ const int16x8_t max = vdupq_n_s16((1 << bd) - 1);
+ int16x8_t o[16];
+ o[0] = vcombine_s16(vrshrn_n_s32(out[0].val[0], 6),
+ vrshrn_n_s32(out[0].val[1], 6));
+ o[1] = vcombine_s16(vrshrn_n_s32(out[1].val[0], 6),
+ vrshrn_n_s32(out[1].val[1], 6));
+ o[2] = vcombine_s16(vrshrn_n_s32(out[2].val[0], 6),
+ vrshrn_n_s32(out[2].val[1], 6));
+ o[3] = vcombine_s16(vrshrn_n_s32(out[3].val[0], 6),
+ vrshrn_n_s32(out[3].val[1], 6));
+ o[4] = vcombine_s16(vrshrn_n_s32(out[4].val[0], 6),
+ vrshrn_n_s32(out[4].val[1], 6));
+ o[5] = vcombine_s16(vrshrn_n_s32(out[5].val[0], 6),
+ vrshrn_n_s32(out[5].val[1], 6));
+ o[6] = vcombine_s16(vrshrn_n_s32(out[6].val[0], 6),
+ vrshrn_n_s32(out[6].val[1], 6));
+ o[7] = vcombine_s16(vrshrn_n_s32(out[7].val[0], 6),
+ vrshrn_n_s32(out[7].val[1], 6));
+ o[8] = vcombine_s16(vrshrn_n_s32(out[8].val[0], 6),
+ vrshrn_n_s32(out[8].val[1], 6));
+ o[9] = vcombine_s16(vrshrn_n_s32(out[9].val[0], 6),
+ vrshrn_n_s32(out[9].val[1], 6));
+ o[10] = vcombine_s16(vrshrn_n_s32(out[10].val[0], 6),
+ vrshrn_n_s32(out[10].val[1], 6));
+ o[11] = vcombine_s16(vrshrn_n_s32(out[11].val[0], 6),
+ vrshrn_n_s32(out[11].val[1], 6));
+ o[12] = vcombine_s16(vrshrn_n_s32(out[12].val[0], 6),
+ vrshrn_n_s32(out[12].val[1], 6));
+ o[13] = vcombine_s16(vrshrn_n_s32(out[13].val[0], 6),
+ vrshrn_n_s32(out[13].val[1], 6));
+ o[14] = vcombine_s16(vrshrn_n_s32(out[14].val[0], 6),
+ vrshrn_n_s32(out[14].val[1], 6));
+ o[15] = vcombine_s16(vrshrn_n_s32(out[15].val[0], 6),
+ vrshrn_n_s32(out[15].val[1], 6));
+ highbd_idct16x16_add8x1(o[0], max, &dest, stride);
+ highbd_idct16x16_add8x1(o[1], max, &dest, stride);
+ highbd_idct16x16_add8x1(o[2], max, &dest, stride);
+ highbd_idct16x16_add8x1(o[3], max, &dest, stride);
+ highbd_idct16x16_add8x1(o[4], max, &dest, stride);
+ highbd_idct16x16_add8x1(o[5], max, &dest, stride);
+ highbd_idct16x16_add8x1(o[6], max, &dest, stride);
+ highbd_idct16x16_add8x1(o[7], max, &dest, stride);
+ highbd_idct16x16_add8x1(o[8], max, &dest, stride);
+ highbd_idct16x16_add8x1(o[9], max, &dest, stride);
+ highbd_idct16x16_add8x1(o[10], max, &dest, stride);
+ highbd_idct16x16_add8x1(o[11], max, &dest, stride);
+ highbd_idct16x16_add8x1(o[12], max, &dest, stride);
+ highbd_idct16x16_add8x1(o[13], max, &dest, stride);
+ highbd_idct16x16_add8x1(o[14], max, &dest, stride);
+ highbd_idct16x16_add8x1(o[15], max, &dest, stride);
+}
+
+void vpx_highbd_idct16x16_256_add_half1d(const int32_t *input, int32_t *output,
+ uint16_t *dest, const int stride,
+ const int bd);
+
#endif // VPX_DSP_ARM_HIGHBD_IDCT_NEON_H_
highbd_idct16x16_add8x1_bd8(a[31], &out, stride);
}
-static INLINE void highbd_idct16x16_add_store(const int32x4x2_t *const out,
- uint16_t *dest, const int stride,
- const int bd) {
- // Add the result to dest
- const int16x8_t max = vdupq_n_s16((1 << bd) - 1);
- int16x8_t o[16];
- o[0] = vcombine_s16(vrshrn_n_s32(out[0].val[0], 6),
- vrshrn_n_s32(out[0].val[1], 6));
- o[1] = vcombine_s16(vrshrn_n_s32(out[1].val[0], 6),
- vrshrn_n_s32(out[1].val[1], 6));
- o[2] = vcombine_s16(vrshrn_n_s32(out[2].val[0], 6),
- vrshrn_n_s32(out[2].val[1], 6));
- o[3] = vcombine_s16(vrshrn_n_s32(out[3].val[0], 6),
- vrshrn_n_s32(out[3].val[1], 6));
- o[4] = vcombine_s16(vrshrn_n_s32(out[4].val[0], 6),
- vrshrn_n_s32(out[4].val[1], 6));
- o[5] = vcombine_s16(vrshrn_n_s32(out[5].val[0], 6),
- vrshrn_n_s32(out[5].val[1], 6));
- o[6] = vcombine_s16(vrshrn_n_s32(out[6].val[0], 6),
- vrshrn_n_s32(out[6].val[1], 6));
- o[7] = vcombine_s16(vrshrn_n_s32(out[7].val[0], 6),
- vrshrn_n_s32(out[7].val[1], 6));
- o[8] = vcombine_s16(vrshrn_n_s32(out[8].val[0], 6),
- vrshrn_n_s32(out[8].val[1], 6));
- o[9] = vcombine_s16(vrshrn_n_s32(out[9].val[0], 6),
- vrshrn_n_s32(out[9].val[1], 6));
- o[10] = vcombine_s16(vrshrn_n_s32(out[10].val[0], 6),
- vrshrn_n_s32(out[10].val[1], 6));
- o[11] = vcombine_s16(vrshrn_n_s32(out[11].val[0], 6),
- vrshrn_n_s32(out[11].val[1], 6));
- o[12] = vcombine_s16(vrshrn_n_s32(out[12].val[0], 6),
- vrshrn_n_s32(out[12].val[1], 6));
- o[13] = vcombine_s16(vrshrn_n_s32(out[13].val[0], 6),
- vrshrn_n_s32(out[13].val[1], 6));
- o[14] = vcombine_s16(vrshrn_n_s32(out[14].val[0], 6),
- vrshrn_n_s32(out[14].val[1], 6));
- o[15] = vcombine_s16(vrshrn_n_s32(out[15].val[0], 6),
- vrshrn_n_s32(out[15].val[1], 6));
- highbd_idct16x16_add8x1(o[0], max, &dest, stride);
- highbd_idct16x16_add8x1(o[1], max, &dest, stride);
- highbd_idct16x16_add8x1(o[2], max, &dest, stride);
- highbd_idct16x16_add8x1(o[3], max, &dest, stride);
- highbd_idct16x16_add8x1(o[4], max, &dest, stride);
- highbd_idct16x16_add8x1(o[5], max, &dest, stride);
- highbd_idct16x16_add8x1(o[6], max, &dest, stride);
- highbd_idct16x16_add8x1(o[7], max, &dest, stride);
- highbd_idct16x16_add8x1(o[8], max, &dest, stride);
- highbd_idct16x16_add8x1(o[9], max, &dest, stride);
- highbd_idct16x16_add8x1(o[10], max, &dest, stride);
- highbd_idct16x16_add8x1(o[11], max, &dest, stride);
- highbd_idct16x16_add8x1(o[12], max, &dest, stride);
- highbd_idct16x16_add8x1(o[13], max, &dest, stride);
- highbd_idct16x16_add8x1(o[14], max, &dest, stride);
- highbd_idct16x16_add8x1(o[15], max, &dest, stride);
-}
-
void vpx_idct16x16_256_add_half1d(const void *const input, int16_t *output,
void *const dest, const int stride,
const int highbd_flag);
--- /dev/null
+/*
+ * Copyright (c) 2018 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 <mmintrin.h>
+
+void vpx_clear_system_state() { _mm_empty(); }
%include "vpx_ports/x86_abi_support.asm"
section .text
-global sym(vpx_reset_mmx_state) PRIVATE
-sym(vpx_reset_mmx_state):
- emms
- ret
-
%if LIBVPX_YASM_WIN64
global sym(vpx_winx64_fldcw) PRIVATE
#include "./vpx_config.h"
-#if ARCH_X86 || ARCH_X86_64
-void vpx_reset_mmx_state(void);
-#define vpx_clear_system_state() vpx_reset_mmx_state()
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#if ARCH_X86 && HAVE_MMX
+extern void vpx_clear_system_state();
#else
#define vpx_clear_system_state()
-#endif // ARCH_X86 || ARCH_X86_64
+#endif // ARCH_X86 && HAVE_MMX
+
+#ifdef __cplusplus
+} // extern "C"
+#endif
+
#endif // VPX_PORTS_SYSTEM_STATE_H_
PORTS_SRCS-yes += system_state.h
PORTS_SRCS-yes += vpx_timer.h
+ifeq ($(ARCH_X86),yes)
+PORTS_SRCS-$(HAVE_MMX) += emms_mmx.c
+endif
+
+ifeq ($(ARCH_X86_64),yes)
+PORTS_SRCS-$(CONFIG_MSVS) += float_control_word.asm
+endif
+
ifeq ($(ARCH_X86)$(ARCH_X86_64),yes)
-PORTS_SRCS-yes += emms.asm
PORTS_SRCS-yes += x86.h
PORTS_SRCS-yes += x86_abi_support.asm
endif
return mode;
}
-extern void vpx_reset_mmx_state(void);
-
#ifdef __cplusplus
} // extern "C"
#endif
projects / libvpx / commitdiff