hdkv/yuzu
1
Fork 0
Code Issues Pull Requests Releases 1 Activity

another try

Browse Source
This commit is contained in:
mgthepro
2022-11-05 13:58:44 +01:00
parent 4a9f2bbf2a
commit 9f63fbe700
2002 changed files with 671171 additions and 671092 deletions
Download Patch File Download Diff File Expand all files Collapse all files

620
src/video_core/host1x/codecs/codec.cpp
View File

@@ -1,310 +1,310 @@
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <algorithm>
#include <fstream>
#include <vector>
#include "common/assert.h"
#include "common/settings.h"
#include "video_core/host1x/codecs/codec.h"
#include "video_core/host1x/codecs/h264.h"
#include "video_core/host1x/codecs/vp8.h"
#include "video_core/host1x/codecs/vp9.h"
#include "video_core/host1x/host1x.h"
#include "video_core/memory_manager.h"
extern "C" {
#include <libavutil/opt.h>
#ifdef LIBVA_FOUND
// for querying VAAPI driver information
#include <libavutil/hwcontext_vaapi.h>
#endif
}
namespace Tegra {
namespace {
constexpr AVPixelFormat PREFERRED_GPU_FMT = AV_PIX_FMT_NV12;
constexpr AVPixelFormat PREFERRED_CPU_FMT = AV_PIX_FMT_YUV420P;
constexpr std::array PREFERRED_GPU_DECODERS = {
AV_HWDEVICE_TYPE_CUDA,
#ifdef _WIN32
AV_HWDEVICE_TYPE_D3D11VA,
AV_HWDEVICE_TYPE_DXVA2,
#elif defined(__unix__)
AV_HWDEVICE_TYPE_VAAPI,
AV_HWDEVICE_TYPE_VDPAU,
#endif
// last resort for Linux Flatpak (w/ NVIDIA)
AV_HWDEVICE_TYPE_VULKAN,
};
void AVPacketDeleter(AVPacket* ptr) {
av_packet_free(&ptr);
}
using AVPacketPtr = std::unique_ptr<AVPacket, decltype(&AVPacketDeleter)>;
AVPixelFormat GetGpuFormat(AVCodecContext* av_codec_ctx, const AVPixelFormat* pix_fmts) {
for (const AVPixelFormat* p = pix_fmts; *p != AV_PIX_FMT_NONE; ++p) {
if (*p == av_codec_ctx->pix_fmt) {
return av_codec_ctx->pix_fmt;
}
}
LOG_INFO(Service_NVDRV, "Could not find compatible GPU AV format, falling back to CPU");
av_buffer_unref(&av_codec_ctx->hw_device_ctx);
av_codec_ctx->pix_fmt = PREFERRED_CPU_FMT;
return PREFERRED_CPU_FMT;
}
// List all the currently available hwcontext in ffmpeg
std::vector<AVHWDeviceType> ListSupportedContexts() {
std::vector<AVHWDeviceType> contexts{};
AVHWDeviceType current_device_type = AV_HWDEVICE_TYPE_NONE;
do {
current_device_type = av_hwdevice_iterate_types(current_device_type);
contexts.push_back(current_device_type);
} while (current_device_type != AV_HWDEVICE_TYPE_NONE);
return contexts;
}
} // namespace
void AVFrameDeleter(AVFrame* ptr) {
av_frame_free(&ptr);
}
Codec::Codec(Host1x::Host1x& host1x_, const Host1x::NvdecCommon::NvdecRegisters& regs)
: host1x(host1x_), state{regs}, h264_decoder(std::make_unique<Decoder::H264>(host1x)),
vp8_decoder(std::make_unique<Decoder::VP8>(host1x)),
vp9_decoder(std::make_unique<Decoder::VP9>(host1x)) {}
Codec::~Codec() {
if (!initialized) {
return;
}
// Free libav memory
avcodec_free_context(&av_codec_ctx);
av_buffer_unref(&av_gpu_decoder);
}
bool Codec::CreateGpuAvDevice() {
static constexpr auto HW_CONFIG_METHOD = AV_CODEC_HW_CONFIG_METHOD_HW_DEVICE_CTX;
static const auto supported_contexts = ListSupportedContexts();
for (const auto& type : PREFERRED_GPU_DECODERS) {
if (std::none_of(supported_contexts.begin(), supported_contexts.end(),
[&type](const auto& context) { return context == type; })) {
LOG_DEBUG(Service_NVDRV, "{} explicitly unsupported", av_hwdevice_get_type_name(type));
continue;
}
// Avoid memory leak from not cleaning up after av_hwdevice_ctx_create
av_buffer_unref(&av_gpu_decoder);
const int hwdevice_res = av_hwdevice_ctx_create(&av_gpu_decoder, type, nullptr, nullptr, 0);
if (hwdevice_res < 0) {
LOG_DEBUG(Service_NVDRV, "{} av_hwdevice_ctx_create failed {}",
av_hwdevice_get_type_name(type), hwdevice_res);
continue;
}
#ifdef LIBVA_FOUND
if (type == AV_HWDEVICE_TYPE_VAAPI) {
// we need to determine if this is an impersonated VAAPI driver
AVHWDeviceContext* hwctx =
static_cast<AVHWDeviceContext*>(static_cast<void*>(av_gpu_decoder->data));
AVVAAPIDeviceContext* vactx = static_cast<AVVAAPIDeviceContext*>(hwctx->hwctx);
const char* vendor_name = vaQueryVendorString(vactx->display);
if (strstr(vendor_name, "VDPAU backend")) {
// VDPAU impersonated VAAPI impl's are super buggy, we need to skip them
LOG_DEBUG(Service_NVDRV, "Skipping vdapu impersonated VAAPI driver");
continue;
} else {
// according to some user testing, certain vaapi driver (Intel?) could be buggy
// so let's log the driver name which may help the developers/supporters
LOG_DEBUG(Service_NVDRV, "Using VAAPI driver: {}", vendor_name);
}
}
#endif
for (int i = 0;; i++) {
const AVCodecHWConfig* config = avcodec_get_hw_config(av_codec, i);
if (!config) {
LOG_DEBUG(Service_NVDRV, "{} decoder does not support device type {}.",
av_codec->name, av_hwdevice_get_type_name(type));
break;
}
if ((config->methods & HW_CONFIG_METHOD) != 0 && config->device_type == type) {
#if defined(__unix__)
// Some linux decoding backends are reported to crash with this config method
// TODO(ameerj): Properly support this method
if ((config->methods & AV_CODEC_HW_CONFIG_METHOD_HW_FRAMES_CTX) != 0) {
// skip zero-copy decoders, we don't currently support them
LOG_DEBUG(Service_NVDRV, "Skipping decoder {} with unsupported capability {}.",
av_hwdevice_get_type_name(type), config->methods);
continue;
}
#endif
LOG_INFO(Service_NVDRV, "Using {} GPU decoder", av_hwdevice_get_type_name(type));
av_codec_ctx->pix_fmt = config->pix_fmt;
return true;
}
}
}
return false;
}
void Codec::InitializeAvCodecContext() {
av_codec_ctx = avcodec_alloc_context3(av_codec);
av_opt_set(av_codec_ctx->priv_data, "tune", "zerolatency", 0);
}
void Codec::InitializeGpuDecoder() {
if (!CreateGpuAvDevice()) {
av_buffer_unref(&av_gpu_decoder);
return;
}
auto* hw_device_ctx = av_buffer_ref(av_gpu_decoder);
ASSERT_MSG(hw_device_ctx, "av_buffer_ref failed");
av_codec_ctx->hw_device_ctx = hw_device_ctx;
av_codec_ctx->get_format = GetGpuFormat;
}
void Codec::Initialize() {
const AVCodecID codec = [&] {
switch (current_codec) {
case Host1x::NvdecCommon::VideoCodec::H264:
return AV_CODEC_ID_H264;
case Host1x::NvdecCommon::VideoCodec::VP8:
return AV_CODEC_ID_VP8;
case Host1x::NvdecCommon::VideoCodec::VP9:
return AV_CODEC_ID_VP9;
default:
UNIMPLEMENTED_MSG("Unknown codec {}", current_codec);
return AV_CODEC_ID_NONE;
}
}();
av_codec = avcodec_find_decoder(codec);
InitializeAvCodecContext();
if (Settings::values.nvdec_emulation.GetValue() == Settings::NvdecEmulation::GPU) {
InitializeGpuDecoder();
}
if (const int res = avcodec_open2(av_codec_ctx, av_codec, nullptr); res < 0) {
LOG_ERROR(Service_NVDRV, "avcodec_open2() Failed with result {}", res);
avcodec_free_context(&av_codec_ctx);
av_buffer_unref(&av_gpu_decoder);
return;
}
if (!av_codec_ctx->hw_device_ctx) {
LOG_INFO(Service_NVDRV, "Using FFmpeg software decoding");
}
initialized = true;
}
void Codec::SetTargetCodec(Host1x::NvdecCommon::VideoCodec codec) {
if (current_codec != codec) {
current_codec = codec;
LOG_INFO(Service_NVDRV, "NVDEC video codec initialized to {}", GetCurrentCodecName());
}
}
void Codec::Decode() {
const bool is_first_frame = !initialized;
if (is_first_frame) {
Initialize();
}
if (!initialized) {
return;
}
bool vp9_hidden_frame = false;
const auto& frame_data = [&]() {
switch (current_codec) {
case Tegra::Host1x::NvdecCommon::VideoCodec::H264:
return h264_decoder->ComposeFrame(state, is_first_frame);
case Tegra::Host1x::NvdecCommon::VideoCodec::VP8:
return vp8_decoder->ComposeFrame(state);
case Tegra::Host1x::NvdecCommon::VideoCodec::VP9:
vp9_decoder->ComposeFrame(state);
vp9_hidden_frame = vp9_decoder->WasFrameHidden();
return vp9_decoder->GetFrameBytes();
default:
ASSERT(false);
return std::vector<u8>{};
}
}();
AVPacketPtr packet{av_packet_alloc(), AVPacketDeleter};
if (!packet) {
LOG_ERROR(Service_NVDRV, "av_packet_alloc failed");
return;
}
packet->data = const_cast<u8*>(frame_data.data());
packet->size = static_cast<s32>(frame_data.size());
if (const int res = avcodec_send_packet(av_codec_ctx, packet.get()); res != 0) {
LOG_DEBUG(Service_NVDRV, "avcodec_send_packet error {}", res);
return;
}
// Only receive/store visible frames
if (vp9_hidden_frame) {
return;
}
AVFramePtr initial_frame{av_frame_alloc(), AVFrameDeleter};
AVFramePtr final_frame{nullptr, AVFrameDeleter};
ASSERT_MSG(initial_frame, "av_frame_alloc initial_frame failed");
if (const int ret = avcodec_receive_frame(av_codec_ctx, initial_frame.get()); ret) {
LOG_DEBUG(Service_NVDRV, "avcodec_receive_frame error {}", ret);
return;
}
if (initial_frame->width == 0 || initial_frame->height == 0) {
LOG_WARNING(Service_NVDRV, "Zero width or height in frame");
return;
}
if (av_codec_ctx->hw_device_ctx) {
final_frame = AVFramePtr{av_frame_alloc(), AVFrameDeleter};
ASSERT_MSG(final_frame, "av_frame_alloc final_frame failed");
// Can't use AV_PIX_FMT_YUV420P and share code with software decoding in vic.cpp
// because Intel drivers crash unless using AV_PIX_FMT_NV12
final_frame->format = PREFERRED_GPU_FMT;
const int ret = av_hwframe_transfer_data(final_frame.get(), initial_frame.get(), 0);
ASSERT_MSG(!ret, "av_hwframe_transfer_data error {}", ret);
} else {
final_frame = std::move(initial_frame);
}
if (final_frame->format != PREFERRED_CPU_FMT && final_frame->format != PREFERRED_GPU_FMT) {
UNIMPLEMENTED_MSG("Unexpected video format: {}", final_frame->format);
return;
}
av_frames.push(std::move(final_frame));
if (av_frames.size() > 10) {
LOG_TRACE(Service_NVDRV, "av_frames.push overflow dropped frame");
av_frames.pop();
}
}
AVFramePtr Codec::GetCurrentFrame() {
// Sometimes VIC will request more frames than have been decoded.
// in this case, return a nullptr and don't overwrite previous frame data
if (av_frames.empty()) {
return AVFramePtr{nullptr, AVFrameDeleter};
}
AVFramePtr frame = std::move(av_frames.front());
av_frames.pop();
return frame;
}
Host1x::NvdecCommon::VideoCodec Codec::GetCurrentCodec() const {
return current_codec;
}
std::string_view Codec::GetCurrentCodecName() const {
switch (current_codec) {
case Host1x::NvdecCommon::VideoCodec::None:
return "None";
case Host1x::NvdecCommon::VideoCodec::H264:
return "H264";
case Host1x::NvdecCommon::VideoCodec::VP8:
return "VP8";
case Host1x::NvdecCommon::VideoCodec::H265:
return "H265";
case Host1x::NvdecCommon::VideoCodec::VP9:
return "VP9";
default:
return "Unknown";
}
}
} // namespace Tegra
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <algorithm>
#include <fstream>
#include <vector>
#include "common/assert.h"
#include "common/settings.h"
#include "video_core/host1x/codecs/codec.h"
#include "video_core/host1x/codecs/h264.h"
#include "video_core/host1x/codecs/vp8.h"
#include "video_core/host1x/codecs/vp9.h"
#include "video_core/host1x/host1x.h"
#include "video_core/memory_manager.h"
extern "C" {
#include <libavutil/opt.h>
#ifdef LIBVA_FOUND
// for querying VAAPI driver information
#include <libavutil/hwcontext_vaapi.h>
#endif
}
namespace Tegra {
namespace {
constexpr AVPixelFormat PREFERRED_GPU_FMT = AV_PIX_FMT_NV12;
constexpr AVPixelFormat PREFERRED_CPU_FMT = AV_PIX_FMT_YUV420P;
constexpr std::array PREFERRED_GPU_DECODERS = {
AV_HWDEVICE_TYPE_CUDA,
#ifdef _WIN32
AV_HWDEVICE_TYPE_D3D11VA,
AV_HWDEVICE_TYPE_DXVA2,
#elif defined(__unix__)
AV_HWDEVICE_TYPE_VAAPI,
AV_HWDEVICE_TYPE_VDPAU,
#endif
// last resort for Linux Flatpak (w/ NVIDIA)
AV_HWDEVICE_TYPE_VULKAN,
};
void AVPacketDeleter(AVPacket* ptr) {
av_packet_free(&ptr);
}
using AVPacketPtr = std::unique_ptr<AVPacket, decltype(&AVPacketDeleter)>;
AVPixelFormat GetGpuFormat(AVCodecContext* av_codec_ctx, const AVPixelFormat* pix_fmts) {
for (const AVPixelFormat* p = pix_fmts; *p != AV_PIX_FMT_NONE; ++p) {
if (*p == av_codec_ctx->pix_fmt) {
return av_codec_ctx->pix_fmt;
}
}
LOG_INFO(Service_NVDRV, "Could not find compatible GPU AV format, falling back to CPU");
av_buffer_unref(&av_codec_ctx->hw_device_ctx);
av_codec_ctx->pix_fmt = PREFERRED_CPU_FMT;
return PREFERRED_CPU_FMT;
}
// List all the currently available hwcontext in ffmpeg
std::vector<AVHWDeviceType> ListSupportedContexts() {
std::vector<AVHWDeviceType> contexts{};
AVHWDeviceType current_device_type = AV_HWDEVICE_TYPE_NONE;
do {
current_device_type = av_hwdevice_iterate_types(current_device_type);
contexts.push_back(current_device_type);
} while (current_device_type != AV_HWDEVICE_TYPE_NONE);
return contexts;
}
} // namespace
void AVFrameDeleter(AVFrame* ptr) {
av_frame_free(&ptr);
}
Codec::Codec(Host1x::Host1x& host1x_, const Host1x::NvdecCommon::NvdecRegisters& regs)
: host1x(host1x_), state{regs}, h264_decoder(std::make_unique<Decoder::H264>(host1x)),
vp8_decoder(std::make_unique<Decoder::VP8>(host1x)),
vp9_decoder(std::make_unique<Decoder::VP9>(host1x)) {}
Codec::~Codec() {
if (!initialized) {
return;
}
// Free libav memory
avcodec_free_context(&av_codec_ctx);
av_buffer_unref(&av_gpu_decoder);
}
bool Codec::CreateGpuAvDevice() {
static constexpr auto HW_CONFIG_METHOD = AV_CODEC_HW_CONFIG_METHOD_HW_DEVICE_CTX;
static const auto supported_contexts = ListSupportedContexts();
for (const auto& type : PREFERRED_GPU_DECODERS) {
if (std::none_of(supported_contexts.begin(), supported_contexts.end(),
[&type](const auto& context) { return context == type; })) {
LOG_DEBUG(Service_NVDRV, "{} explicitly unsupported", av_hwdevice_get_type_name(type));
continue;
}
// Avoid memory leak from not cleaning up after av_hwdevice_ctx_create
av_buffer_unref(&av_gpu_decoder);
const int hwdevice_res = av_hwdevice_ctx_create(&av_gpu_decoder, type, nullptr, nullptr, 0);
if (hwdevice_res < 0) {
LOG_DEBUG(Service_NVDRV, "{} av_hwdevice_ctx_create failed {}",
av_hwdevice_get_type_name(type), hwdevice_res);
continue;
}
#ifdef LIBVA_FOUND
if (type == AV_HWDEVICE_TYPE_VAAPI) {
// we need to determine if this is an impersonated VAAPI driver
AVHWDeviceContext* hwctx =
static_cast<AVHWDeviceContext*>(static_cast<void*>(av_gpu_decoder->data));
AVVAAPIDeviceContext* vactx = static_cast<AVVAAPIDeviceContext*>(hwctx->hwctx);
const char* vendor_name = vaQueryVendorString(vactx->display);
if (strstr(vendor_name, "VDPAU backend")) {
// VDPAU impersonated VAAPI impl's are super buggy, we need to skip them
LOG_DEBUG(Service_NVDRV, "Skipping vdapu impersonated VAAPI driver");
continue;
} else {
// according to some user testing, certain vaapi driver (Intel?) could be buggy
// so let's log the driver name which may help the developers/supporters
LOG_DEBUG(Service_NVDRV, "Using VAAPI driver: {}", vendor_name);
}
}
#endif
for (int i = 0;; i++) {
const AVCodecHWConfig* config = avcodec_get_hw_config(av_codec, i);
if (!config) {
LOG_DEBUG(Service_NVDRV, "{} decoder does not support device type {}.",
av_codec->name, av_hwdevice_get_type_name(type));
break;
}
if ((config->methods & HW_CONFIG_METHOD) != 0 && config->device_type == type) {
#if defined(__unix__)
// Some linux decoding backends are reported to crash with this config method
// TODO(ameerj): Properly support this method
if ((config->methods & AV_CODEC_HW_CONFIG_METHOD_HW_FRAMES_CTX) != 0) {
// skip zero-copy decoders, we don't currently support them
LOG_DEBUG(Service_NVDRV, "Skipping decoder {} with unsupported capability {}.",
av_hwdevice_get_type_name(type), config->methods);
continue;
}
#endif
LOG_INFO(Service_NVDRV, "Using {} GPU decoder", av_hwdevice_get_type_name(type));
av_codec_ctx->pix_fmt = config->pix_fmt;
return true;
}
}
}
return false;
}
void Codec::InitializeAvCodecContext() {
av_codec_ctx = avcodec_alloc_context3(av_codec);
av_opt_set(av_codec_ctx->priv_data, "tune", "zerolatency", 0);
}
void Codec::InitializeGpuDecoder() {
if (!CreateGpuAvDevice()) {
av_buffer_unref(&av_gpu_decoder);
return;
}
auto* hw_device_ctx = av_buffer_ref(av_gpu_decoder);
ASSERT_MSG(hw_device_ctx, "av_buffer_ref failed");
av_codec_ctx->hw_device_ctx = hw_device_ctx;
av_codec_ctx->get_format = GetGpuFormat;
}
void Codec::Initialize() {
const AVCodecID codec = [&] {
switch (current_codec) {
case Host1x::NvdecCommon::VideoCodec::H264:
return AV_CODEC_ID_H264;
case Host1x::NvdecCommon::VideoCodec::VP8:
return AV_CODEC_ID_VP8;
case Host1x::NvdecCommon::VideoCodec::VP9:
return AV_CODEC_ID_VP9;
default:
UNIMPLEMENTED_MSG("Unknown codec {}", current_codec);
return AV_CODEC_ID_NONE;
}
}();
av_codec = avcodec_find_decoder(codec);
InitializeAvCodecContext();
if (Settings::values.nvdec_emulation.GetValue() == Settings::NvdecEmulation::GPU) {
InitializeGpuDecoder();
}
if (const int res = avcodec_open2(av_codec_ctx, av_codec, nullptr); res < 0) {
LOG_ERROR(Service_NVDRV, "avcodec_open2() Failed with result {}", res);
avcodec_free_context(&av_codec_ctx);
av_buffer_unref(&av_gpu_decoder);
return;
}
if (!av_codec_ctx->hw_device_ctx) {
LOG_INFO(Service_NVDRV, "Using FFmpeg software decoding");
}
initialized = true;
}
void Codec::SetTargetCodec(Host1x::NvdecCommon::VideoCodec codec) {
if (current_codec != codec) {
current_codec = codec;
LOG_INFO(Service_NVDRV, "NVDEC video codec initialized to {}", GetCurrentCodecName());
}
}
void Codec::Decode() {
const bool is_first_frame = !initialized;
if (is_first_frame) {
Initialize();
}
if (!initialized) {
return;
}
bool vp9_hidden_frame = false;
const auto& frame_data = [&]() {
switch (current_codec) {
case Tegra::Host1x::NvdecCommon::VideoCodec::H264:
return h264_decoder->ComposeFrame(state, is_first_frame);
case Tegra::Host1x::NvdecCommon::VideoCodec::VP8:
return vp8_decoder->ComposeFrame(state);
case Tegra::Host1x::NvdecCommon::VideoCodec::VP9:
vp9_decoder->ComposeFrame(state);
vp9_hidden_frame = vp9_decoder->WasFrameHidden();
return vp9_decoder->GetFrameBytes();
default:
ASSERT(false);
return std::vector<u8>{};
}
}();
AVPacketPtr packet{av_packet_alloc(), AVPacketDeleter};
if (!packet) {
LOG_ERROR(Service_NVDRV, "av_packet_alloc failed");
return;
}
packet->data = const_cast<u8*>(frame_data.data());
packet->size = static_cast<s32>(frame_data.size());
if (const int res = avcodec_send_packet(av_codec_ctx, packet.get()); res != 0) {
LOG_DEBUG(Service_NVDRV, "avcodec_send_packet error {}", res);
return;
}
// Only receive/store visible frames
if (vp9_hidden_frame) {
return;
}
AVFramePtr initial_frame{av_frame_alloc(), AVFrameDeleter};
AVFramePtr final_frame{nullptr, AVFrameDeleter};
ASSERT_MSG(initial_frame, "av_frame_alloc initial_frame failed");
if (const int ret = avcodec_receive_frame(av_codec_ctx, initial_frame.get()); ret) {
LOG_DEBUG(Service_NVDRV, "avcodec_receive_frame error {}", ret);
return;
}
if (initial_frame->width == 0 || initial_frame->height == 0) {
LOG_WARNING(Service_NVDRV, "Zero width or height in frame");
return;
}
if (av_codec_ctx->hw_device_ctx) {
final_frame = AVFramePtr{av_frame_alloc(), AVFrameDeleter};
ASSERT_MSG(final_frame, "av_frame_alloc final_frame failed");
// Can't use AV_PIX_FMT_YUV420P and share code with software decoding in vic.cpp
// because Intel drivers crash unless using AV_PIX_FMT_NV12
final_frame->format = PREFERRED_GPU_FMT;
const int ret = av_hwframe_transfer_data(final_frame.get(), initial_frame.get(), 0);
ASSERT_MSG(!ret, "av_hwframe_transfer_data error {}", ret);
} else {
final_frame = std::move(initial_frame);
}
if (final_frame->format != PREFERRED_CPU_FMT && final_frame->format != PREFERRED_GPU_FMT) {
UNIMPLEMENTED_MSG("Unexpected video format: {}", final_frame->format);
return;
}
av_frames.push(std::move(final_frame));
if (av_frames.size() > 10) {
LOG_TRACE(Service_NVDRV, "av_frames.push overflow dropped frame");
av_frames.pop();
}
}
AVFramePtr Codec::GetCurrentFrame() {
// Sometimes VIC will request more frames than have been decoded.
// in this case, return a nullptr and don't overwrite previous frame data
if (av_frames.empty()) {
return AVFramePtr{nullptr, AVFrameDeleter};
}
AVFramePtr frame = std::move(av_frames.front());
av_frames.pop();
return frame;
}
Host1x::NvdecCommon::VideoCodec Codec::GetCurrentCodec() const {
return current_codec;
}
std::string_view Codec::GetCurrentCodecName() const {
switch (current_codec) {
case Host1x::NvdecCommon::VideoCodec::None:
return "None";
case Host1x::NvdecCommon::VideoCodec::H264:
return "H264";
case Host1x::NvdecCommon::VideoCodec::VP8:
return "VP8";
case Host1x::NvdecCommon::VideoCodec::H265:
return "H265";
case Host1x::NvdecCommon::VideoCodec::VP9:
return "VP9";
default:
return "Unknown";
}
}
} // namespace Tegra

168
src/video_core/host1x/codecs/codec.h
View File

@@ -1,84 +1,84 @@
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <memory>
#include <string_view>
#include <queue>
#include "common/common_types.h"
#include "video_core/host1x/nvdec_common.h"
extern "C" {
#if defined(__GNUC__) || defined(__clang__)
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wconversion"
#endif
#include <libavcodec/avcodec.h>
#if defined(__GNUC__) || defined(__clang__)
#pragma GCC diagnostic pop
#endif
}
namespace Tegra {
void AVFrameDeleter(AVFrame* ptr);
using AVFramePtr = std::unique_ptr<AVFrame, decltype(&AVFrameDeleter)>;
namespace Decoder {
class H264;
class VP8;
class VP9;
} // namespace Decoder
namespace Host1x {
class Host1x;
} // namespace Host1x
class Codec {
public:
explicit Codec(Host1x::Host1x& host1x, const Host1x::NvdecCommon::NvdecRegisters& regs);
~Codec();
/// Initialize the codec, returning success or failure
void Initialize();
/// Sets NVDEC video stream codec
void SetTargetCodec(Host1x::NvdecCommon::VideoCodec codec);
/// Call decoders to construct headers, decode AVFrame with ffmpeg
void Decode();
/// Returns next decoded frame
[[nodiscard]] AVFramePtr GetCurrentFrame();
/// Returns the value of current_codec
[[nodiscard]] Host1x::NvdecCommon::VideoCodec GetCurrentCodec() const;
/// Return name of the current codec
[[nodiscard]] std::string_view GetCurrentCodecName() const;
private:
void InitializeAvCodecContext();
void InitializeGpuDecoder();
bool CreateGpuAvDevice();
bool initialized{};
Host1x::NvdecCommon::VideoCodec current_codec{Host1x::NvdecCommon::VideoCodec::None};
const AVCodec* av_codec{nullptr};
AVCodecContext* av_codec_ctx{nullptr};
AVBufferRef* av_gpu_decoder{nullptr};
Host1x::Host1x& host1x;
const Host1x::NvdecCommon::NvdecRegisters& state;
std::unique_ptr<Decoder::H264> h264_decoder;
std::unique_ptr<Decoder::VP8> vp8_decoder;
std::unique_ptr<Decoder::VP9> vp9_decoder;
std::queue<AVFramePtr> av_frames{};
};
} // namespace Tegra
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <memory>
#include <string_view>
#include <queue>
#include "common/common_types.h"
#include "video_core/host1x/nvdec_common.h"
extern "C" {
#if defined(__GNUC__) || defined(__clang__)
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wconversion"
#endif
#include <libavcodec/avcodec.h>
#if defined(__GNUC__) || defined(__clang__)
#pragma GCC diagnostic pop
#endif
}
namespace Tegra {
void AVFrameDeleter(AVFrame* ptr);
using AVFramePtr = std::unique_ptr<AVFrame, decltype(&AVFrameDeleter)>;
namespace Decoder {
class H264;
class VP8;
class VP9;
} // namespace Decoder
namespace Host1x {
class Host1x;
} // namespace Host1x
class Codec {
public:
explicit Codec(Host1x::Host1x& host1x, const Host1x::NvdecCommon::NvdecRegisters& regs);
~Codec();
/// Initialize the codec, returning success or failure
void Initialize();
/// Sets NVDEC video stream codec
void SetTargetCodec(Host1x::NvdecCommon::VideoCodec codec);
/// Call decoders to construct headers, decode AVFrame with ffmpeg
void Decode();
/// Returns next decoded frame
[[nodiscard]] AVFramePtr GetCurrentFrame();
/// Returns the value of current_codec
[[nodiscard]] Host1x::NvdecCommon::VideoCodec GetCurrentCodec() const;
/// Return name of the current codec
[[nodiscard]] std::string_view GetCurrentCodecName() const;
private:
void InitializeAvCodecContext();
void InitializeGpuDecoder();
bool CreateGpuAvDevice();
bool initialized{};
Host1x::NvdecCommon::VideoCodec current_codec{Host1x::NvdecCommon::VideoCodec::None};
const AVCodec* av_codec{nullptr};
AVCodecContext* av_codec_ctx{nullptr};
AVBufferRef* av_gpu_decoder{nullptr};
Host1x::Host1x& host1x;
const Host1x::NvdecCommon::NvdecRegisters& state;
std::unique_ptr<Decoder::H264> h264_decoder;
std::unique_ptr<Decoder::VP8> vp8_decoder;
std::unique_ptr<Decoder::VP9> vp9_decoder;
std::queue<AVFramePtr> av_frames{};
};
} // namespace Tegra

556
src/video_core/host1x/codecs/h264.cpp
View File

@@ -1,278 +1,278 @@
// SPDX-FileCopyrightText: Ryujinx Team and Contributors
// SPDX-License-Identifier: MIT
#include <array>
#include <bit>
#include "common/settings.h"
#include "video_core/host1x/codecs/h264.h"
#include "video_core/host1x/host1x.h"
#include "video_core/memory_manager.h"
namespace Tegra::Decoder {
namespace {
// ZigZag LUTs from libavcodec.
constexpr std::array<u8, 64> zig_zag_direct{
0, 1, 8, 16, 9, 2, 3, 10, 17, 24, 32, 25, 18, 11, 4, 5, 12, 19, 26, 33, 40, 48,
41, 34, 27, 20, 13, 6, 7, 14, 21, 28, 35, 42, 49, 56, 57, 50, 43, 36, 29, 22, 15, 23,
30, 37, 44, 51, 58, 59, 52, 45, 38, 31, 39, 46, 53, 60, 61, 54, 47, 55, 62, 63,
};
constexpr std::array<u8, 16> zig_zag_scan{
0 + 0 * 4, 1 + 0 * 4, 0 + 1 * 4, 0 + 2 * 4, 1 + 1 * 4, 2 + 0 * 4, 3 + 0 * 4, 2 + 1 * 4,
1 + 2 * 4, 0 + 3 * 4, 1 + 3 * 4, 2 + 2 * 4, 3 + 1 * 4, 3 + 2 * 4, 2 + 3 * 4, 3 + 3 * 4,
};
} // Anonymous namespace
H264::H264(Host1x::Host1x& host1x_) : host1x{host1x_} {}
H264::~H264() = default;
const std::vector<u8>& H264::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state,
bool is_first_frame) {
H264DecoderContext context;
host1x.MemoryManager().ReadBlock(state.picture_info_offset, &context,
sizeof(H264DecoderContext));
const s64 frame_number = context.h264_parameter_set.frame_number.Value();
if (!is_first_frame && frame_number != 0) {
frame.resize(context.stream_len);
host1x.MemoryManager().ReadBlock(state.frame_bitstream_offset, frame.data(), frame.size());
return frame;
}
// Encode header
H264BitWriter writer{};
writer.WriteU(1, 24);
writer.WriteU(0, 1);
writer.WriteU(3, 2);
writer.WriteU(7, 5);
writer.WriteU(100, 8);
writer.WriteU(0, 8);
writer.WriteU(31, 8);
writer.WriteUe(0);
const u32 chroma_format_idc =
static_cast<u32>(context.h264_parameter_set.chroma_format_idc.Value());
writer.WriteUe(chroma_format_idc);
if (chroma_format_idc == 3) {
writer.WriteBit(false);
}
writer.WriteUe(0);
writer.WriteUe(0);
writer.WriteBit(false); // QpprimeYZeroTransformBypassFlag
writer.WriteBit(false); // Scaling matrix present flag
writer.WriteUe(static_cast<u32>(context.h264_parameter_set.log2_max_frame_num_minus4.Value()));
const auto order_cnt_type =
static_cast<u32>(context.h264_parameter_set.pic_order_cnt_type.Value());
writer.WriteUe(order_cnt_type);
if (order_cnt_type == 0) {
writer.WriteUe(context.h264_parameter_set.log2_max_pic_order_cnt_lsb_minus4);
} else if (order_cnt_type == 1) {
writer.WriteBit(context.h264_parameter_set.delta_pic_order_always_zero_flag != 0);
writer.WriteSe(0);
writer.WriteSe(0);
writer.WriteUe(0);
}
const s32 pic_height = context.h264_parameter_set.frame_height_in_map_units /
(context.h264_parameter_set.frame_mbs_only_flag ? 1 : 2);
// TODO (ameerj): Where do we get this number, it seems to be particular for each stream
const auto nvdec_decoding = Settings::values.nvdec_emulation.GetValue();
const bool uses_gpu_decoding = nvdec_decoding == Settings::NvdecEmulation::GPU;
const u32 max_num_ref_frames = uses_gpu_decoding ? 6u : 16u;
writer.WriteUe(max_num_ref_frames);
writer.WriteBit(false);
writer.WriteUe(context.h264_parameter_set.pic_width_in_mbs - 1);
writer.WriteUe(pic_height - 1);
writer.WriteBit(context.h264_parameter_set.frame_mbs_only_flag != 0);
if (!context.h264_parameter_set.frame_mbs_only_flag) {
writer.WriteBit(context.h264_parameter_set.flags.mbaff_frame.Value() != 0);
}
writer.WriteBit(context.h264_parameter_set.flags.direct_8x8_inference.Value() != 0);
writer.WriteBit(false); // Frame cropping flag
writer.WriteBit(false); // VUI parameter present flag
writer.End();
// H264 PPS
writer.WriteU(1, 24);
writer.WriteU(0, 1);
writer.WriteU(3, 2);
writer.WriteU(8, 5);
writer.WriteUe(0);
writer.WriteUe(0);
writer.WriteBit(context.h264_parameter_set.entropy_coding_mode_flag != 0);
writer.WriteBit(false);
writer.WriteUe(0);
writer.WriteUe(context.h264_parameter_set.num_refidx_l0_default_active);
writer.WriteUe(context.h264_parameter_set.num_refidx_l1_default_active);
writer.WriteBit(context.h264_parameter_set.flags.weighted_pred.Value() != 0);
writer.WriteU(static_cast<s32>(context.h264_parameter_set.weighted_bipred_idc.Value()), 2);
s32 pic_init_qp = static_cast<s32>(context.h264_parameter_set.pic_init_qp_minus26.Value());
writer.WriteSe(pic_init_qp);
writer.WriteSe(0);
s32 chroma_qp_index_offset =
static_cast<s32>(context.h264_parameter_set.chroma_qp_index_offset.Value());
writer.WriteSe(chroma_qp_index_offset);
writer.WriteBit(context.h264_parameter_set.deblocking_filter_control_present_flag != 0);
writer.WriteBit(context.h264_parameter_set.flags.constrained_intra_pred.Value() != 0);
writer.WriteBit(context.h264_parameter_set.redundant_pic_cnt_present_flag != 0);
writer.WriteBit(context.h264_parameter_set.transform_8x8_mode_flag != 0);
writer.WriteBit(true);
for (s32 index = 0; index < 6; index++) {
writer.WriteBit(true);
std::span<const u8> matrix{context.weight_scale};
writer.WriteScalingList(matrix, index * 16, 16);
}
if (context.h264_parameter_set.transform_8x8_mode_flag) {
for (s32 index = 0; index < 2; index++) {
writer.WriteBit(true);
std::span<const u8> matrix{context.weight_scale_8x8};
writer.WriteScalingList(matrix, index * 64, 64);
}
}
s32 chroma_qp_index_offset2 =
static_cast<s32>(context.h264_parameter_set.second_chroma_qp_index_offset.Value());
writer.WriteSe(chroma_qp_index_offset2);
writer.End();
const auto& encoded_header = writer.GetByteArray();
frame.resize(encoded_header.size() + context.stream_len);
std::memcpy(frame.data(), encoded_header.data(), encoded_header.size());
host1x.MemoryManager().ReadBlock(state.frame_bitstream_offset,
frame.data() + encoded_header.size(), context.stream_len);
return frame;
}
H264BitWriter::H264BitWriter() = default;
H264BitWriter::~H264BitWriter() = default;
void H264BitWriter::WriteU(s32 value, s32 value_sz) {
WriteBits(value, value_sz);
}
void H264BitWriter::WriteSe(s32 value) {
WriteExpGolombCodedInt(value);
}
void H264BitWriter::WriteUe(u32 value) {
WriteExpGolombCodedUInt(value);
}
void H264BitWriter::End() {
WriteBit(true);
Flush();
}
void H264BitWriter::WriteBit(bool state) {
WriteBits(state ? 1 : 0, 1);
}
void H264BitWriter::WriteScalingList(std::span<const u8> list, s32 start, s32 count) {
std::vector<u8> scan(count);
if (count == 16) {
std::memcpy(scan.data(), zig_zag_scan.data(), scan.size());
} else {
std::memcpy(scan.data(), zig_zag_direct.data(), scan.size());
}
u8 last_scale = 8;
for (s32 index = 0; index < count; index++) {
const u8 value = list[start + scan[index]];
const s32 delta_scale = static_cast<s32>(value - last_scale);
WriteSe(delta_scale);
last_scale = value;
}
}
std::vector<u8>& H264BitWriter::GetByteArray() {
return byte_array;
}
const std::vector<u8>& H264BitWriter::GetByteArray() const {
return byte_array;
}
void H264BitWriter::WriteBits(s32 value, s32 bit_count) {
s32 value_pos = 0;
s32 remaining = bit_count;
while (remaining > 0) {
s32 copy_size = remaining;
const s32 free_bits = GetFreeBufferBits();
if (copy_size > free_bits) {
copy_size = free_bits;
}
const s32 mask = (1 << copy_size) - 1;
const s32 src_shift = (bit_count - value_pos) - copy_size;
const s32 dst_shift = (buffer_size - buffer_pos) - copy_size;
buffer |= ((value >> src_shift) & mask) << dst_shift;
value_pos += copy_size;
buffer_pos += copy_size;
remaining -= copy_size;
}
}
void H264BitWriter::WriteExpGolombCodedInt(s32 value) {
const s32 sign = value <= 0 ? 0 : 1;
if (value < 0) {
value = -value;
}
value = (value << 1) - sign;
WriteExpGolombCodedUInt(value);
}
void H264BitWriter::WriteExpGolombCodedUInt(u32 value) {
const s32 size = 32 - std::countl_zero(value + 1);
WriteBits(1, size);
value -= (1U << (size - 1)) - 1;
WriteBits(static_cast<s32>(value), size - 1);
}
s32 H264BitWriter::GetFreeBufferBits() {
if (buffer_pos == buffer_size) {
Flush();
}
return buffer_size - buffer_pos;
}
void H264BitWriter::Flush() {
if (buffer_pos == 0) {
return;
}
byte_array.push_back(static_cast<u8>(buffer));
buffer = 0;
buffer_pos = 0;
}
} // namespace Tegra::Decoder
// SPDX-FileCopyrightText: Ryujinx Team and Contributors
// SPDX-License-Identifier: MIT
#include <array>
#include <bit>
#include "common/settings.h"
#include "video_core/host1x/codecs/h264.h"
#include "video_core/host1x/host1x.h"
#include "video_core/memory_manager.h"
namespace Tegra::Decoder {
namespace {
// ZigZag LUTs from libavcodec.
constexpr std::array<u8, 64> zig_zag_direct{
0, 1, 8, 16, 9, 2, 3, 10, 17, 24, 32, 25, 18, 11, 4, 5, 12, 19, 26, 33, 40, 48,
41, 34, 27, 20, 13, 6, 7, 14, 21, 28, 35, 42, 49, 56, 57, 50, 43, 36, 29, 22, 15, 23,
30, 37, 44, 51, 58, 59, 52, 45, 38, 31, 39, 46, 53, 60, 61, 54, 47, 55, 62, 63,
};
constexpr std::array<u8, 16> zig_zag_scan{
0 + 0 * 4, 1 + 0 * 4, 0 + 1 * 4, 0 + 2 * 4, 1 + 1 * 4, 2 + 0 * 4, 3 + 0 * 4, 2 + 1 * 4,
1 + 2 * 4, 0 + 3 * 4, 1 + 3 * 4, 2 + 2 * 4, 3 + 1 * 4, 3 + 2 * 4, 2 + 3 * 4, 3 + 3 * 4,
};
} // Anonymous namespace
H264::H264(Host1x::Host1x& host1x_) : host1x{host1x_} {}
H264::~H264() = default;
const std::vector<u8>& H264::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state,
bool is_first_frame) {
H264DecoderContext context;
host1x.MemoryManager().ReadBlock(state.picture_info_offset, &context,
sizeof(H264DecoderContext));
const s64 frame_number = context.h264_parameter_set.frame_number.Value();
if (!is_first_frame && frame_number != 0) {
frame.resize(context.stream_len);
host1x.MemoryManager().ReadBlock(state.frame_bitstream_offset, frame.data(), frame.size());
return frame;
}
// Encode header
H264BitWriter writer{};
writer.WriteU(1, 24);
writer.WriteU(0, 1);
writer.WriteU(3, 2);
writer.WriteU(7, 5);
writer.WriteU(100, 8);
writer.WriteU(0, 8);
writer.WriteU(31, 8);
writer.WriteUe(0);
const u32 chroma_format_idc =
static_cast<u32>(context.h264_parameter_set.chroma_format_idc.Value());
writer.WriteUe(chroma_format_idc);
if (chroma_format_idc == 3) {
writer.WriteBit(false);
}
writer.WriteUe(0);
writer.WriteUe(0);
writer.WriteBit(false); // QpprimeYZeroTransformBypassFlag
writer.WriteBit(false); // Scaling matrix present flag
writer.WriteUe(static_cast<u32>(context.h264_parameter_set.log2_max_frame_num_minus4.Value()));
const auto order_cnt_type =
static_cast<u32>(context.h264_parameter_set.pic_order_cnt_type.Value());
writer.WriteUe(order_cnt_type);
if (order_cnt_type == 0) {
writer.WriteUe(context.h264_parameter_set.log2_max_pic_order_cnt_lsb_minus4);
} else if (order_cnt_type == 1) {
writer.WriteBit(context.h264_parameter_set.delta_pic_order_always_zero_flag != 0);
writer.WriteSe(0);
writer.WriteSe(0);
writer.WriteUe(0);
}
const s32 pic_height = context.h264_parameter_set.frame_height_in_map_units /
(context.h264_parameter_set.frame_mbs_only_flag ? 1 : 2);
// TODO (ameerj): Where do we get this number, it seems to be particular for each stream
const auto nvdec_decoding = Settings::values.nvdec_emulation.GetValue();
const bool uses_gpu_decoding = nvdec_decoding == Settings::NvdecEmulation::GPU;
const u32 max_num_ref_frames = uses_gpu_decoding ? 6u : 16u;
writer.WriteUe(max_num_ref_frames);
writer.WriteBit(false);
writer.WriteUe(context.h264_parameter_set.pic_width_in_mbs - 1);
writer.WriteUe(pic_height - 1);
writer.WriteBit(context.h264_parameter_set.frame_mbs_only_flag != 0);
if (!context.h264_parameter_set.frame_mbs_only_flag) {
writer.WriteBit(context.h264_parameter_set.flags.mbaff_frame.Value() != 0);
}
writer.WriteBit(context.h264_parameter_set.flags.direct_8x8_inference.Value() != 0);
writer.WriteBit(false); // Frame cropping flag
writer.WriteBit(false); // VUI parameter present flag
writer.End();
// H264 PPS
writer.WriteU(1, 24);
writer.WriteU(0, 1);
writer.WriteU(3, 2);
writer.WriteU(8, 5);
writer.WriteUe(0);
writer.WriteUe(0);
writer.WriteBit(context.h264_parameter_set.entropy_coding_mode_flag != 0);
writer.WriteBit(false);
writer.WriteUe(0);
writer.WriteUe(context.h264_parameter_set.num_refidx_l0_default_active);
writer.WriteUe(context.h264_parameter_set.num_refidx_l1_default_active);
writer.WriteBit(context.h264_parameter_set.flags.weighted_pred.Value() != 0);
writer.WriteU(static_cast<s32>(context.h264_parameter_set.weighted_bipred_idc.Value()), 2);
s32 pic_init_qp = static_cast<s32>(context.h264_parameter_set.pic_init_qp_minus26.Value());
writer.WriteSe(pic_init_qp);
writer.WriteSe(0);
s32 chroma_qp_index_offset =
static_cast<s32>(context.h264_parameter_set.chroma_qp_index_offset.Value());
writer.WriteSe(chroma_qp_index_offset);
writer.WriteBit(context.h264_parameter_set.deblocking_filter_control_present_flag != 0);
writer.WriteBit(context.h264_parameter_set.flags.constrained_intra_pred.Value() != 0);
writer.WriteBit(context.h264_parameter_set.redundant_pic_cnt_present_flag != 0);
writer.WriteBit(context.h264_parameter_set.transform_8x8_mode_flag != 0);
writer.WriteBit(true);
for (s32 index = 0; index < 6; index++) {
writer.WriteBit(true);
std::span<const u8> matrix{context.weight_scale};
writer.WriteScalingList(matrix, index * 16, 16);
}
if (context.h264_parameter_set.transform_8x8_mode_flag) {
for (s32 index = 0; index < 2; index++) {
writer.WriteBit(true);
std::span<const u8> matrix{context.weight_scale_8x8};
writer.WriteScalingList(matrix, index * 64, 64);
}
}
s32 chroma_qp_index_offset2 =
static_cast<s32>(context.h264_parameter_set.second_chroma_qp_index_offset.Value());
writer.WriteSe(chroma_qp_index_offset2);
writer.End();
const auto& encoded_header = writer.GetByteArray();
frame.resize(encoded_header.size() + context.stream_len);
std::memcpy(frame.data(), encoded_header.data(), encoded_header.size());
host1x.MemoryManager().ReadBlock(state.frame_bitstream_offset,
frame.data() + encoded_header.size(), context.stream_len);
return frame;
}
H264BitWriter::H264BitWriter() = default;
H264BitWriter::~H264BitWriter() = default;
void H264BitWriter::WriteU(s32 value, s32 value_sz) {
WriteBits(value, value_sz);
}
void H264BitWriter::WriteSe(s32 value) {
WriteExpGolombCodedInt(value);
}
void H264BitWriter::WriteUe(u32 value) {
WriteExpGolombCodedUInt(value);
}
void H264BitWriter::End() {
WriteBit(true);
Flush();
}
void H264BitWriter::WriteBit(bool state) {
WriteBits(state ? 1 : 0, 1);
}
void H264BitWriter::WriteScalingList(std::span<const u8> list, s32 start, s32 count) {
std::vector<u8> scan(count);
if (count == 16) {
std::memcpy(scan.data(), zig_zag_scan.data(), scan.size());
} else {
std::memcpy(scan.data(), zig_zag_direct.data(), scan.size());
}
u8 last_scale = 8;
for (s32 index = 0; index < count; index++) {
const u8 value = list[start + scan[index]];
const s32 delta_scale = static_cast<s32>(value - last_scale);
WriteSe(delta_scale);
last_scale = value;
}
}
std::vector<u8>& H264BitWriter::GetByteArray() {
return byte_array;
}
const std::vector<u8>& H264BitWriter::GetByteArray() const {
return byte_array;
}
void H264BitWriter::WriteBits(s32 value, s32 bit_count) {
s32 value_pos = 0;
s32 remaining = bit_count;
while (remaining > 0) {
s32 copy_size = remaining;
const s32 free_bits = GetFreeBufferBits();
if (copy_size > free_bits) {
copy_size = free_bits;
}
const s32 mask = (1 << copy_size) - 1;
const s32 src_shift = (bit_count - value_pos) - copy_size;
const s32 dst_shift = (buffer_size - buffer_pos) - copy_size;
buffer |= ((value >> src_shift) & mask) << dst_shift;
value_pos += copy_size;
buffer_pos += copy_size;
remaining -= copy_size;
}
}
void H264BitWriter::WriteExpGolombCodedInt(s32 value) {
const s32 sign = value <= 0 ? 0 : 1;
if (value < 0) {
value = -value;
}
value = (value << 1) - sign;
WriteExpGolombCodedUInt(value);
}
void H264BitWriter::WriteExpGolombCodedUInt(u32 value) {
const s32 size = 32 - std::countl_zero(value + 1);
WriteBits(1, size);
value -= (1U << (size - 1)) - 1;
WriteBits(static_cast<s32>(value), size - 1);
}
s32 H264BitWriter::GetFreeBufferBits() {
if (buffer_pos == buffer_size) {
Flush();
}
return buffer_size - buffer_pos;
}
void H264BitWriter::Flush() {
if (buffer_pos == 0) {
return;
}
byte_array.push_back(static_cast<u8>(buffer));
buffer = 0;
buffer_pos = 0;
}
} // namespace Tegra::Decoder

354
src/video_core/host1x/codecs/h264.h
View File

@@ -1,177 +1,177 @@
// SPDX-FileCopyrightText: Ryujinx Team and Contributors
// SPDX-License-Identifier: MIT
#pragma once
#include <span>
#include <vector>
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "video_core/host1x/nvdec_common.h"
namespace Tegra {
namespace Host1x {
class Host1x;
} // namespace Host1x
namespace Decoder {
class H264BitWriter {
public:
H264BitWriter();
~H264BitWriter();
/// The following Write methods are based on clause 9.1 in the H.264 specification.
/// WriteSe and WriteUe write in the Exp-Golomb-coded syntax
void WriteU(s32 value, s32 value_sz);
void WriteSe(s32 value);
void WriteUe(u32 value);
/// Finalize the bitstream
void End();
/// append a bit to the stream, equivalent value to the state parameter
void WriteBit(bool state);
/// Based on section 7.3.2.1.1.1 and Table 7-4 in the H.264 specification
/// Writes the scaling matrices of the sream
void WriteScalingList(std::span<const u8> list, s32 start, s32 count);
/// Return the bitstream as a vector.
[[nodiscard]] std::vector<u8>& GetByteArray();
[[nodiscard]] const std::vector<u8>& GetByteArray() const;
private:
void WriteBits(s32 value, s32 bit_count);
void WriteExpGolombCodedInt(s32 value);
void WriteExpGolombCodedUInt(u32 value);
[[nodiscard]] s32 GetFreeBufferBits();
void Flush();
s32 buffer_size{8};
s32 buffer{};
s32 buffer_pos{};
std::vector<u8> byte_array;
};
class H264 {
public:
explicit H264(Host1x::Host1x& host1x);
~H264();
/// Compose the H264 frame for FFmpeg decoding
[[nodiscard]] const std::vector<u8>& ComposeFrame(
const Host1x::NvdecCommon::NvdecRegisters& state, bool is_first_frame = false);
private:
std::vector<u8> frame;
Host1x::Host1x& host1x;
struct H264ParameterSet {
s32 log2_max_pic_order_cnt_lsb_minus4; ///< 0x00
s32 delta_pic_order_always_zero_flag; ///< 0x04
s32 frame_mbs_only_flag; ///< 0x08
u32 pic_width_in_mbs; ///< 0x0C
u32 frame_height_in_map_units; ///< 0x10
union { ///< 0x14
BitField<0, 2, u32> tile_format;
BitField<2, 3, u32> gob_height;
};
u32 entropy_coding_mode_flag; ///< 0x18
s32 pic_order_present_flag; ///< 0x1C
s32 num_refidx_l0_default_active; ///< 0x20
s32 num_refidx_l1_default_active; ///< 0x24
s32 deblocking_filter_control_present_flag; ///< 0x28
s32 redundant_pic_cnt_present_flag; ///< 0x2C
u32 transform_8x8_mode_flag; ///< 0x30
u32 pitch_luma; ///< 0x34
u32 pitch_chroma; ///< 0x38
u32 luma_top_offset; ///< 0x3C
u32 luma_bot_offset; ///< 0x40
u32 luma_frame_offset; ///< 0x44
u32 chroma_top_offset; ///< 0x48
u32 chroma_bot_offset; ///< 0x4C
u32 chroma_frame_offset; ///< 0x50
u32 hist_buffer_size; ///< 0x54
union { ///< 0x58
union {
BitField<0, 1, u64> mbaff_frame;
BitField<1, 1, u64> direct_8x8_inference;
BitField<2, 1, u64> weighted_pred;
BitField<3, 1, u64> constrained_intra_pred;
BitField<4, 1, u64> ref_pic;
BitField<5, 1, u64> field_pic;
BitField<6, 1, u64> bottom_field;
BitField<7, 1, u64> second_field;
} flags;
BitField<8, 4, u64> log2_max_frame_num_minus4;
BitField<12, 2, u64> chroma_format_idc;
BitField<14, 2, u64> pic_order_cnt_type;
BitField<16, 6, s64> pic_init_qp_minus26;
BitField<22, 5, s64> chroma_qp_index_offset;
BitField<27, 5, s64> second_chroma_qp_index_offset;
BitField<32, 2, u64> weighted_bipred_idc;
BitField<34, 7, u64> curr_pic_idx;
BitField<41, 5, u64> curr_col_idx;
BitField<46, 16, u64> frame_number;
BitField<62, 1, u64> frame_surfaces;
BitField<63, 1, u64> output_memory_layout;
};
};
static_assert(sizeof(H264ParameterSet) == 0x60, "H264ParameterSet is an invalid size");
struct H264DecoderContext {
INSERT_PADDING_WORDS_NOINIT(18); ///< 0x0000
u32 stream_len; ///< 0x0048
INSERT_PADDING_WORDS_NOINIT(3); ///< 0x004C
H264ParameterSet h264_parameter_set; ///< 0x0058
INSERT_PADDING_WORDS_NOINIT(66); ///< 0x00B8
std::array<u8, 0x60> weight_scale; ///< 0x01C0
std::array<u8, 0x80> weight_scale_8x8; ///< 0x0220
};
static_assert(sizeof(H264DecoderContext) == 0x2A0, "H264DecoderContext is an invalid size");
#define ASSERT_POSITION(field_name, position) \
static_assert(offsetof(H264ParameterSet, field_name) == position, \
"Field " #field_name " has invalid position")
ASSERT_POSITION(log2_max_pic_order_cnt_lsb_minus4, 0x00);
ASSERT_POSITION(delta_pic_order_always_zero_flag, 0x04);
ASSERT_POSITION(frame_mbs_only_flag, 0x08);
ASSERT_POSITION(pic_width_in_mbs, 0x0C);
ASSERT_POSITION(frame_height_in_map_units, 0x10);
ASSERT_POSITION(tile_format, 0x14);
ASSERT_POSITION(entropy_coding_mode_flag, 0x18);
ASSERT_POSITION(pic_order_present_flag, 0x1C);
ASSERT_POSITION(num_refidx_l0_default_active, 0x20);
ASSERT_POSITION(num_refidx_l1_default_active, 0x24);
ASSERT_POSITION(deblocking_filter_control_present_flag, 0x28);
ASSERT_POSITION(redundant_pic_cnt_present_flag, 0x2C);
ASSERT_POSITION(transform_8x8_mode_flag, 0x30);
ASSERT_POSITION(pitch_luma, 0x34);
ASSERT_POSITION(pitch_chroma, 0x38);
ASSERT_POSITION(luma_top_offset, 0x3C);
ASSERT_POSITION(luma_bot_offset, 0x40);
ASSERT_POSITION(luma_frame_offset, 0x44);
ASSERT_POSITION(chroma_top_offset, 0x48);
ASSERT_POSITION(chroma_bot_offset, 0x4C);
ASSERT_POSITION(chroma_frame_offset, 0x50);
ASSERT_POSITION(hist_buffer_size, 0x54);
ASSERT_POSITION(flags, 0x58);
#undef ASSERT_POSITION
#define ASSERT_POSITION(field_name, position) \
static_assert(offsetof(H264DecoderContext, field_name) == position, \
"Field " #field_name " has invalid position")
ASSERT_POSITION(stream_len, 0x48);
ASSERT_POSITION(h264_parameter_set, 0x58);
ASSERT_POSITION(weight_scale, 0x1C0);
#undef ASSERT_POSITION
};
} // namespace Decoder
} // namespace Tegra
// SPDX-FileCopyrightText: Ryujinx Team and Contributors
// SPDX-License-Identifier: MIT
#pragma once
#include <span>
#include <vector>
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "video_core/host1x/nvdec_common.h"
namespace Tegra {
namespace Host1x {
class Host1x;
} // namespace Host1x
namespace Decoder {
class H264BitWriter {
public:
H264BitWriter();
~H264BitWriter();
/// The following Write methods are based on clause 9.1 in the H.264 specification.
/// WriteSe and WriteUe write in the Exp-Golomb-coded syntax
void WriteU(s32 value, s32 value_sz);
void WriteSe(s32 value);
void WriteUe(u32 value);
/// Finalize the bitstream
void End();
/// append a bit to the stream, equivalent value to the state parameter
void WriteBit(bool state);
/// Based on section 7.3.2.1.1.1 and Table 7-4 in the H.264 specification
/// Writes the scaling matrices of the sream
void WriteScalingList(std::span<const u8> list, s32 start, s32 count);
/// Return the bitstream as a vector.
[[nodiscard]] std::vector<u8>& GetByteArray();
[[nodiscard]] const std::vector<u8>& GetByteArray() const;
private:
void WriteBits(s32 value, s32 bit_count);
void WriteExpGolombCodedInt(s32 value);
void WriteExpGolombCodedUInt(u32 value);
[[nodiscard]] s32 GetFreeBufferBits();
void Flush();
s32 buffer_size{8};
s32 buffer{};
s32 buffer_pos{};
std::vector<u8> byte_array;
};
class H264 {
public:
explicit H264(Host1x::Host1x& host1x);
~H264();
/// Compose the H264 frame for FFmpeg decoding
[[nodiscard]] const std::vector<u8>& ComposeFrame(
const Host1x::NvdecCommon::NvdecRegisters& state, bool is_first_frame = false);
private:
std::vector<u8> frame;
Host1x::Host1x& host1x;
struct H264ParameterSet {
s32 log2_max_pic_order_cnt_lsb_minus4; ///< 0x00
s32 delta_pic_order_always_zero_flag; ///< 0x04
s32 frame_mbs_only_flag; ///< 0x08
u32 pic_width_in_mbs; ///< 0x0C
u32 frame_height_in_map_units; ///< 0x10
union { ///< 0x14
BitField<0, 2, u32> tile_format;
BitField<2, 3, u32> gob_height;
};
u32 entropy_coding_mode_flag; ///< 0x18
s32 pic_order_present_flag; ///< 0x1C
s32 num_refidx_l0_default_active; ///< 0x20
s32 num_refidx_l1_default_active; ///< 0x24
s32 deblocking_filter_control_present_flag; ///< 0x28
s32 redundant_pic_cnt_present_flag; ///< 0x2C
u32 transform_8x8_mode_flag; ///< 0x30
u32 pitch_luma; ///< 0x34
u32 pitch_chroma; ///< 0x38
u32 luma_top_offset; ///< 0x3C
u32 luma_bot_offset; ///< 0x40
u32 luma_frame_offset; ///< 0x44
u32 chroma_top_offset; ///< 0x48
u32 chroma_bot_offset; ///< 0x4C
u32 chroma_frame_offset; ///< 0x50
u32 hist_buffer_size; ///< 0x54
union { ///< 0x58
union {
BitField<0, 1, u64> mbaff_frame;
BitField<1, 1, u64> direct_8x8_inference;
BitField<2, 1, u64> weighted_pred;
BitField<3, 1, u64> constrained_intra_pred;
BitField<4, 1, u64> ref_pic;
BitField<5, 1, u64> field_pic;
BitField<6, 1, u64> bottom_field;
BitField<7, 1, u64> second_field;
} flags;
BitField<8, 4, u64> log2_max_frame_num_minus4;
BitField<12, 2, u64> chroma_format_idc;
BitField<14, 2, u64> pic_order_cnt_type;
BitField<16, 6, s64> pic_init_qp_minus26;
BitField<22, 5, s64> chroma_qp_index_offset;
BitField<27, 5, s64> second_chroma_qp_index_offset;
BitField<32, 2, u64> weighted_bipred_idc;
BitField<34, 7, u64> curr_pic_idx;
BitField<41, 5, u64> curr_col_idx;
BitField<46, 16, u64> frame_number;
BitField<62, 1, u64> frame_surfaces;
BitField<63, 1, u64> output_memory_layout;
};
};
static_assert(sizeof(H264ParameterSet) == 0x60, "H264ParameterSet is an invalid size");
struct H264DecoderContext {
INSERT_PADDING_WORDS_NOINIT(18); ///< 0x0000
u32 stream_len; ///< 0x0048
INSERT_PADDING_WORDS_NOINIT(3); ///< 0x004C
H264ParameterSet h264_parameter_set; ///< 0x0058
INSERT_PADDING_WORDS_NOINIT(66); ///< 0x00B8
std::array<u8, 0x60> weight_scale; ///< 0x01C0
std::array<u8, 0x80> weight_scale_8x8; ///< 0x0220
};
static_assert(sizeof(H264DecoderContext) == 0x2A0, "H264DecoderContext is an invalid size");
#define ASSERT_POSITION(field_name, position) \
static_assert(offsetof(H264ParameterSet, field_name) == position, \
"Field " #field_name " has invalid position")
ASSERT_POSITION(log2_max_pic_order_cnt_lsb_minus4, 0x00);
ASSERT_POSITION(delta_pic_order_always_zero_flag, 0x04);
ASSERT_POSITION(frame_mbs_only_flag, 0x08);
ASSERT_POSITION(pic_width_in_mbs, 0x0C);
ASSERT_POSITION(frame_height_in_map_units, 0x10);
ASSERT_POSITION(tile_format, 0x14);
ASSERT_POSITION(entropy_coding_mode_flag, 0x18);
ASSERT_POSITION(pic_order_present_flag, 0x1C);
ASSERT_POSITION(num_refidx_l0_default_active, 0x20);
ASSERT_POSITION(num_refidx_l1_default_active, 0x24);
ASSERT_POSITION(deblocking_filter_control_present_flag, 0x28);
ASSERT_POSITION(redundant_pic_cnt_present_flag, 0x2C);
ASSERT_POSITION(transform_8x8_mode_flag, 0x30);
ASSERT_POSITION(pitch_luma, 0x34);
ASSERT_POSITION(pitch_chroma, 0x38);
ASSERT_POSITION(luma_top_offset, 0x3C);
ASSERT_POSITION(luma_bot_offset, 0x40);
ASSERT_POSITION(luma_frame_offset, 0x44);
ASSERT_POSITION(chroma_top_offset, 0x48);
ASSERT_POSITION(chroma_bot_offset, 0x4C);
ASSERT_POSITION(chroma_frame_offset, 0x50);
ASSERT_POSITION(hist_buffer_size, 0x54);
ASSERT_POSITION(flags, 0x58);
#undef ASSERT_POSITION
#define ASSERT_POSITION(field_name, position) \
static_assert(offsetof(H264DecoderContext, field_name) == position, \
"Field " #field_name " has invalid position")
ASSERT_POSITION(stream_len, 0x48);
ASSERT_POSITION(h264_parameter_set, 0x58);
ASSERT_POSITION(weight_scale, 0x1C0);
#undef ASSERT_POSITION
};
} // namespace Decoder
} // namespace Tegra

106
src/video_core/host1x/codecs/vp8.cpp
View File

@@ -1,53 +1,53 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <vector>
#include "video_core/host1x/codecs/vp8.h"
#include "video_core/host1x/host1x.h"
#include "video_core/memory_manager.h"
namespace Tegra::Decoder {
VP8::VP8(Host1x::Host1x& host1x_) : host1x{host1x_} {}
VP8::~VP8() = default;
const std::vector<u8>& VP8::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state) {
VP8PictureInfo info;
host1x.MemoryManager().ReadBlock(state.picture_info_offset, &info, sizeof(VP8PictureInfo));
const bool is_key_frame = info.key_frame == 1u;
const auto bitstream_size = static_cast<size_t>(info.vld_buffer_size);
const size_t header_size = is_key_frame ? 10u : 3u;
frame.resize(header_size + bitstream_size);
// Based on page 30 of the VP8 specification.
// https://datatracker.ietf.org/doc/rfc6386/
frame[0] = is_key_frame ? 0u : 1u; // 1-bit frame type (0: keyframe, 1: interframes).
frame[0] |= static_cast<u8>((info.version & 7u) << 1u); // 3-bit version number
frame[0] |= static_cast<u8>(1u << 4u); // 1-bit show_frame flag
// The next 19-bits are the first partition size
frame[0] |= static_cast<u8>((info.first_part_size & 7u) << 5u);
frame[1] = static_cast<u8>((info.first_part_size & 0x7f8u) >> 3u);
frame[2] = static_cast<u8>((info.first_part_size & 0x7f800u) >> 11u);
if (is_key_frame) {
frame[3] = 0x9du;
frame[4] = 0x01u;
frame[5] = 0x2au;
// TODO(ameerj): Horizontal/Vertical Scale
// 16 bits: (2 bits Horizontal Scale << 14) | Width (14 bits)
frame[6] = static_cast<u8>(info.frame_width & 0xff);
frame[7] = static_cast<u8>(((info.frame_width >> 8) & 0x3f));
// 16 bits:(2 bits Vertical Scale << 14) | Height (14 bits)
frame[8] = static_cast<u8>(info.frame_height & 0xff);
frame[9] = static_cast<u8>(((info.frame_height >> 8) & 0x3f));
}
const u64 bitstream_offset = state.frame_bitstream_offset;
host1x.MemoryManager().ReadBlock(bitstream_offset, frame.data() + header_size, bitstream_size);
return frame;
}
} // namespace Tegra::Decoder
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <vector>
#include "video_core/host1x/codecs/vp8.h"
#include "video_core/host1x/host1x.h"
#include "video_core/memory_manager.h"
namespace Tegra::Decoder {
VP8::VP8(Host1x::Host1x& host1x_) : host1x{host1x_} {}
VP8::~VP8() = default;
const std::vector<u8>& VP8::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state) {
VP8PictureInfo info;
host1x.MemoryManager().ReadBlock(state.picture_info_offset, &info, sizeof(VP8PictureInfo));
const bool is_key_frame = info.key_frame == 1u;
const auto bitstream_size = static_cast<size_t>(info.vld_buffer_size);
const size_t header_size = is_key_frame ? 10u : 3u;
frame.resize(header_size + bitstream_size);
// Based on page 30 of the VP8 specification.
// https://datatracker.ietf.org/doc/rfc6386/
frame[0] = is_key_frame ? 0u : 1u; // 1-bit frame type (0: keyframe, 1: interframes).
frame[0] |= static_cast<u8>((info.version & 7u) << 1u); // 3-bit version number
frame[0] |= static_cast<u8>(1u << 4u); // 1-bit show_frame flag
// The next 19-bits are the first partition size
frame[0] |= static_cast<u8>((info.first_part_size & 7u) << 5u);
frame[1] = static_cast<u8>((info.first_part_size & 0x7f8u) >> 3u);
frame[2] = static_cast<u8>((info.first_part_size & 0x7f800u) >> 11u);
if (is_key_frame) {
frame[3] = 0x9du;
frame[4] = 0x01u;
frame[5] = 0x2au;
// TODO(ameerj): Horizontal/Vertical Scale
// 16 bits: (2 bits Horizontal Scale << 14) | Width (14 bits)
frame[6] = static_cast<u8>(info.frame_width & 0xff);
frame[7] = static_cast<u8>(((info.frame_width >> 8) & 0x3f));
// 16 bits:(2 bits Vertical Scale << 14) | Height (14 bits)
frame[8] = static_cast<u8>(info.frame_height & 0xff);
frame[9] = static_cast<u8>(((info.frame_height >> 8) & 0x3f));
}
const u64 bitstream_offset = state.frame_bitstream_offset;
host1x.MemoryManager().ReadBlock(bitstream_offset, frame.data() + header_size, bitstream_size);
return frame;
}
} // namespace Tegra::Decoder

156
src/video_core/host1x/codecs/vp8.h
View File

@@ -1,78 +1,78 @@
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <array>
#include <vector>
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "video_core/host1x/nvdec_common.h"
namespace Tegra {
namespace Host1x {
class Host1x;
} // namespace Host1x
namespace Decoder {
class VP8 {
public:
explicit VP8(Host1x::Host1x& host1x);
~VP8();
/// Compose the VP8 frame for FFmpeg decoding
[[nodiscard]] const std::vector<u8>& ComposeFrame(
const Host1x::NvdecCommon::NvdecRegisters& state);
private:
std::vector<u8> frame;
Host1x::Host1x& host1x;
struct VP8PictureInfo {
INSERT_PADDING_WORDS_NOINIT(14);
u16 frame_width; // actual frame width
u16 frame_height; // actual frame height
u8 key_frame;
u8 version;
union {
u8 raw;
BitField<0, 2, u8> tile_format;
BitField<2, 3, u8> gob_height;
BitField<5, 3, u8> reserverd_surface_format;
};
u8 error_conceal_on; // 1: error conceal on; 0: off
u32 first_part_size; // the size of first partition(frame header and mb header partition)
u32 hist_buffer_size; // in units of 256
u32 vld_buffer_size; // in units of 1
// Current frame buffers
std::array<u32, 2> frame_stride; // [y_c]
u32 luma_top_offset; // offset of luma top field in units of 256
u32 luma_bot_offset; // offset of luma bottom field in units of 256
u32 luma_frame_offset; // offset of luma frame in units of 256
u32 chroma_top_offset; // offset of chroma top field in units of 256
u32 chroma_bot_offset; // offset of chroma bottom field in units of 256
u32 chroma_frame_offset; // offset of chroma frame in units of 256
INSERT_PADDING_BYTES_NOINIT(0x1c); // NvdecDisplayParams
// Decode picture buffer related
s8 current_output_memory_layout;
// output NV12/NV24 setting. index 0: golden; 1: altref; 2: last
std::array<s8, 3> output_memory_layout;
u8 segmentation_feature_data_update;
INSERT_PADDING_BYTES_NOINIT(3);
// ucode return result
u32 result_value;
std::array<u32, 8> partition_offset;
INSERT_PADDING_WORDS_NOINIT(3);
};
static_assert(sizeof(VP8PictureInfo) == 0xc0, "PictureInfo is an invalid size");
};
} // namespace Decoder
} // namespace Tegra
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <array>
#include <vector>
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "video_core/host1x/nvdec_common.h"
namespace Tegra {
namespace Host1x {
class Host1x;
} // namespace Host1x
namespace Decoder {
class VP8 {
public:
explicit VP8(Host1x::Host1x& host1x);
~VP8();
/// Compose the VP8 frame for FFmpeg decoding
[[nodiscard]] const std::vector<u8>& ComposeFrame(
const Host1x::NvdecCommon::NvdecRegisters& state);
private:
std::vector<u8> frame;
Host1x::Host1x& host1x;
struct VP8PictureInfo {
INSERT_PADDING_WORDS_NOINIT(14);
u16 frame_width; // actual frame width
u16 frame_height; // actual frame height
u8 key_frame;
u8 version;
union {
u8 raw;
BitField<0, 2, u8> tile_format;
BitField<2, 3, u8> gob_height;
BitField<5, 3, u8> reserverd_surface_format;
};
u8 error_conceal_on; // 1: error conceal on; 0: off
u32 first_part_size; // the size of first partition(frame header and mb header partition)
u32 hist_buffer_size; // in units of 256
u32 vld_buffer_size; // in units of 1
// Current frame buffers
std::array<u32, 2> frame_stride; // [y_c]
u32 luma_top_offset; // offset of luma top field in units of 256
u32 luma_bot_offset; // offset of luma bottom field in units of 256
u32 luma_frame_offset; // offset of luma frame in units of 256
u32 chroma_top_offset; // offset of chroma top field in units of 256
u32 chroma_bot_offset; // offset of chroma bottom field in units of 256
u32 chroma_frame_offset; // offset of chroma frame in units of 256
INSERT_PADDING_BYTES_NOINIT(0x1c); // NvdecDisplayParams
// Decode picture buffer related
s8 current_output_memory_layout;
// output NV12/NV24 setting. index 0: golden; 1: altref; 2: last
std::array<s8, 3> output_memory_layout;
u8 segmentation_feature_data_update;
INSERT_PADDING_BYTES_NOINIT(3);
// ucode return result
u32 result_value;
std::array<u32, 8> partition_offset;
INSERT_PADDING_WORDS_NOINIT(3);
};
static_assert(sizeof(VP8PictureInfo) == 0xc0, "PictureInfo is an invalid size");
};
} // namespace Decoder
} // namespace Tegra

1894
src/video_core/host1x/codecs/vp9.cpp
View File

File diff suppressed because it is too large Load Diff

396
src/video_core/host1x/codecs/vp9.h
View File

@@ -1,198 +1,198 @@
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <array>
#include <vector>
#include "common/common_types.h"
#include "common/stream.h"
#include "video_core/host1x/codecs/vp9_types.h"
#include "video_core/host1x/nvdec_common.h"
namespace Tegra {
namespace Host1x {
class Host1x;
} // namespace Host1x
namespace Decoder {
/// The VpxRangeEncoder, and VpxBitStreamWriter classes are used to compose the
/// VP9 header bitstreams.
class VpxRangeEncoder {
public:
VpxRangeEncoder();
~VpxRangeEncoder();
VpxRangeEncoder(const VpxRangeEncoder&) = delete;
VpxRangeEncoder& operator=(const VpxRangeEncoder&) = delete;
VpxRangeEncoder(VpxRangeEncoder&&) = default;
VpxRangeEncoder& operator=(VpxRangeEncoder&&) = default;
/// Writes the rightmost value_size bits from value into the stream
void Write(s32 value, s32 value_size);
/// Writes a single bit with half probability
void Write(bool bit);
/// Writes a bit to the base_stream encoded with probability
void Write(bool bit, s32 probability);
/// Signal the end of the bitstream
void End();
[[nodiscard]] std::vector<u8>& GetBuffer() {
return base_stream.GetBuffer();
}
[[nodiscard]] const std::vector<u8>& GetBuffer() const {
return base_stream.GetBuffer();
}
private:
u8 PeekByte();
Common::Stream base_stream{};
u32 low_value{};
u32 range{0xff};
s32 count{-24};
s32 half_probability{128};
};
class VpxBitStreamWriter {
public:
VpxBitStreamWriter();
~VpxBitStreamWriter();
VpxBitStreamWriter(const VpxBitStreamWriter&) = delete;
VpxBitStreamWriter& operator=(const VpxBitStreamWriter&) = delete;
VpxBitStreamWriter(VpxBitStreamWriter&&) = default;
VpxBitStreamWriter& operator=(VpxBitStreamWriter&&) = default;
/// Write an unsigned integer value
void WriteU(u32 value, u32 value_size);
/// Write a signed integer value
void WriteS(s32 value, u32 value_size);
/// Based on 6.2.10 of VP9 Spec, writes a delta coded value
void WriteDeltaQ(u32 value);
/// Write a single bit.
void WriteBit(bool state);
/// Pushes current buffer into buffer_array, resets buffer
void Flush();
/// Returns byte_array
[[nodiscard]] std::vector<u8>& GetByteArray();
/// Returns const byte_array
[[nodiscard]] const std::vector<u8>& GetByteArray() const;
private:
/// Write bit_count bits from value into buffer
void WriteBits(u32 value, u32 bit_count);
/// Gets next available position in buffer, invokes Flush() if buffer is full
s32 GetFreeBufferBits();
s32 buffer_size{8};
s32 buffer{};
s32 buffer_pos{};
std::vector<u8> byte_array;
};
class VP9 {
public:
explicit VP9(Host1x::Host1x& host1x);
~VP9();
VP9(const VP9&) = delete;
VP9& operator=(const VP9&) = delete;
VP9(VP9&&) = default;
VP9& operator=(VP9&&) = delete;
/// Composes the VP9 frame from the GPU state information.
/// Based on the official VP9 spec documentation
void ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state);
/// Returns true if the most recent frame was a hidden frame.
[[nodiscard]] bool WasFrameHidden() const {
return !current_frame_info.show_frame;
}
/// Returns a const reference to the composed frame data.
[[nodiscard]] const std::vector<u8>& GetFrameBytes() const {
return frame;
}
private:
/// Generates compressed header probability updates in the bitstream writer
template <typename T, std::size_t N>
void WriteProbabilityUpdate(VpxRangeEncoder& writer, const std::array<T, N>& new_prob,
const std::array<T, N>& old_prob);
/// Generates compressed header probability updates in the bitstream writer
/// If probs are not equal, WriteProbabilityDelta is invoked
void WriteProbabilityUpdate(VpxRangeEncoder& writer, u8 new_prob, u8 old_prob);
/// Generates compressed header probability deltas in the bitstream writer
void WriteProbabilityDelta(VpxRangeEncoder& writer, u8 new_prob, u8 old_prob);
/// Inverse of 6.3.4 Decode term subexp
void EncodeTermSubExp(VpxRangeEncoder& writer, s32 value);
/// Writes if the value is less than the test value
bool WriteLessThan(VpxRangeEncoder& writer, s32 value, s32 test);
/// Writes probability updates for the Coef probabilities
void WriteCoefProbabilityUpdate(VpxRangeEncoder& writer, s32 tx_mode,
const std::array<u8, 1728>& new_prob,
const std::array<u8, 1728>& old_prob);
/// Write probabilities for 4-byte aligned structures
template <typename T, std::size_t N>
void WriteProbabilityUpdateAligned4(VpxRangeEncoder& writer, const std::array<T, N>& new_prob,
const std::array<T, N>& old_prob);
/// Write motion vector probability updates. 6.3.17 in the spec
void WriteMvProbabilityUpdate(VpxRangeEncoder& writer, u8 new_prob, u8 old_prob);
/// Returns VP9 information from NVDEC provided offset and size
[[nodiscard]] Vp9PictureInfo GetVp9PictureInfo(
const Host1x::NvdecCommon::NvdecRegisters& state);
/// Read and convert NVDEC provided entropy probs to Vp9EntropyProbs struct
void InsertEntropy(u64 offset, Vp9EntropyProbs& dst);
/// Returns frame to be decoded after buffering
[[nodiscard]] Vp9FrameContainer GetCurrentFrame(
const Host1x::NvdecCommon::NvdecRegisters& state);
/// Use NVDEC providied information to compose the headers for the current frame
[[nodiscard]] std::vector<u8> ComposeCompressedHeader();
[[nodiscard]] VpxBitStreamWriter ComposeUncompressedHeader();
Host1x::Host1x& host1x;
std::vector<u8> frame;
std::array<s8, 4> loop_filter_ref_deltas{};
std::array<s8, 2> loop_filter_mode_deltas{};
Vp9FrameContainer next_frame{};
std::array<Vp9EntropyProbs, 4> frame_ctxs{};
bool swap_ref_indices{};
Vp9PictureInfo current_frame_info{};
Vp9EntropyProbs prev_frame_probs{};
};
} // namespace Decoder
} // namespace Tegra
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <array>
#include <vector>
#include "common/common_types.h"
#include "common/stream.h"
#include "video_core/host1x/codecs/vp9_types.h"
#include "video_core/host1x/nvdec_common.h"
namespace Tegra {
namespace Host1x {
class Host1x;
} // namespace Host1x
namespace Decoder {
/// The VpxRangeEncoder, and VpxBitStreamWriter classes are used to compose the
/// VP9 header bitstreams.
class VpxRangeEncoder {
public:
VpxRangeEncoder();
~VpxRangeEncoder();
VpxRangeEncoder(const VpxRangeEncoder&) = delete;
VpxRangeEncoder& operator=(const VpxRangeEncoder&) = delete;
VpxRangeEncoder(VpxRangeEncoder&&) = default;
VpxRangeEncoder& operator=(VpxRangeEncoder&&) = default;
/// Writes the rightmost value_size bits from value into the stream
void Write(s32 value, s32 value_size);
/// Writes a single bit with half probability
void Write(bool bit);
/// Writes a bit to the base_stream encoded with probability
void Write(bool bit, s32 probability);
/// Signal the end of the bitstream
void End();
[[nodiscard]] std::vector<u8>& GetBuffer() {
return base_stream.GetBuffer();
}
[[nodiscard]] const std::vector<u8>& GetBuffer() const {
return base_stream.GetBuffer();
}
private:
u8 PeekByte();
Common::Stream base_stream{};
u32 low_value{};
u32 range{0xff};
s32 count{-24};
s32 half_probability{128};
};
class VpxBitStreamWriter {
public:
VpxBitStreamWriter();
~VpxBitStreamWriter();
VpxBitStreamWriter(const VpxBitStreamWriter&) = delete;
VpxBitStreamWriter& operator=(const VpxBitStreamWriter&) = delete;
VpxBitStreamWriter(VpxBitStreamWriter&&) = default;
VpxBitStreamWriter& operator=(VpxBitStreamWriter&&) = default;
/// Write an unsigned integer value
void WriteU(u32 value, u32 value_size);
/// Write a signed integer value
void WriteS(s32 value, u32 value_size);
/// Based on 6.2.10 of VP9 Spec, writes a delta coded value
void WriteDeltaQ(u32 value);
/// Write a single bit.
void WriteBit(bool state);
/// Pushes current buffer into buffer_array, resets buffer
void Flush();
/// Returns byte_array
[[nodiscard]] std::vector<u8>& GetByteArray();
/// Returns const byte_array
[[nodiscard]] const std::vector<u8>& GetByteArray() const;
private:
/// Write bit_count bits from value into buffer
void WriteBits(u32 value, u32 bit_count);
/// Gets next available position in buffer, invokes Flush() if buffer is full
s32 GetFreeBufferBits();
s32 buffer_size{8};
s32 buffer{};
s32 buffer_pos{};
std::vector<u8> byte_array;
};
class VP9 {
public:
explicit VP9(Host1x::Host1x& host1x);
~VP9();
VP9(const VP9&) = delete;
VP9& operator=(const VP9&) = delete;
VP9(VP9&&) = default;
VP9& operator=(VP9&&) = delete;
/// Composes the VP9 frame from the GPU state information.
/// Based on the official VP9 spec documentation
void ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state);
/// Returns true if the most recent frame was a hidden frame.
[[nodiscard]] bool WasFrameHidden() const {
return !current_frame_info.show_frame;
}
/// Returns a const reference to the composed frame data.
[[nodiscard]] const std::vector<u8>& GetFrameBytes() const {
return frame;
}
private:
/// Generates compressed header probability updates in the bitstream writer
template <typename T, std::size_t N>
void WriteProbabilityUpdate(VpxRangeEncoder& writer, const std::array<T, N>& new_prob,
const std::array<T, N>& old_prob);
/// Generates compressed header probability updates in the bitstream writer
/// If probs are not equal, WriteProbabilityDelta is invoked
void WriteProbabilityUpdate(VpxRangeEncoder& writer, u8 new_prob, u8 old_prob);
/// Generates compressed header probability deltas in the bitstream writer
void WriteProbabilityDelta(VpxRangeEncoder& writer, u8 new_prob, u8 old_prob);
/// Inverse of 6.3.4 Decode term subexp
void EncodeTermSubExp(VpxRangeEncoder& writer, s32 value);
/// Writes if the value is less than the test value
bool WriteLessThan(VpxRangeEncoder& writer, s32 value, s32 test);
/// Writes probability updates for the Coef probabilities
void WriteCoefProbabilityUpdate(VpxRangeEncoder& writer, s32 tx_mode,
const std::array<u8, 1728>& new_prob,
const std::array<u8, 1728>& old_prob);
/// Write probabilities for 4-byte aligned structures
template <typename T, std::size_t N>
void WriteProbabilityUpdateAligned4(VpxRangeEncoder& writer, const std::array<T, N>& new_prob,
const std::array<T, N>& old_prob);
/// Write motion vector probability updates. 6.3.17 in the spec
void WriteMvProbabilityUpdate(VpxRangeEncoder& writer, u8 new_prob, u8 old_prob);
/// Returns VP9 information from NVDEC provided offset and size
[[nodiscard]] Vp9PictureInfo GetVp9PictureInfo(
const Host1x::NvdecCommon::NvdecRegisters& state);
/// Read and convert NVDEC provided entropy probs to Vp9EntropyProbs struct
void InsertEntropy(u64 offset, Vp9EntropyProbs& dst);
/// Returns frame to be decoded after buffering
[[nodiscard]] Vp9FrameContainer GetCurrentFrame(
const Host1x::NvdecCommon::NvdecRegisters& state);
/// Use NVDEC providied information to compose the headers for the current frame
[[nodiscard]] std::vector<u8> ComposeCompressedHeader();
[[nodiscard]] VpxBitStreamWriter ComposeUncompressedHeader();
Host1x::Host1x& host1x;
std::vector<u8> frame;
std::array<s8, 4> loop_filter_ref_deltas{};
std::array<s8, 2> loop_filter_mode_deltas{};
Vp9FrameContainer next_frame{};
std::array<Vp9EntropyProbs, 4> frame_ctxs{};
bool swap_ref_indices{};
Vp9PictureInfo current_frame_info{};
Vp9EntropyProbs prev_frame_probs{};
};
} // namespace Decoder
} // namespace Tegra

610
src/video_core/host1x/codecs/vp9_types.h
View File

@@ -1,305 +1,305 @@
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <array>
#include <vector>
#include "common/common_funcs.h"
#include "common/common_types.h"
namespace Tegra {
namespace Decoder {
struct Vp9FrameDimensions {
s16 width;
s16 height;
s16 luma_pitch;
s16 chroma_pitch;
};
static_assert(sizeof(Vp9FrameDimensions) == 0x8, "Vp9 Vp9FrameDimensions is an invalid size");
enum class FrameFlags : u32 {
IsKeyFrame = 1 << 0,
LastFrameIsKeyFrame = 1 << 1,
FrameSizeChanged = 1 << 2,
ErrorResilientMode = 1 << 3,
LastShowFrame = 1 << 4,
IntraOnly = 1 << 5,
};
DECLARE_ENUM_FLAG_OPERATORS(FrameFlags)
enum class TxSize {
Tx4x4 = 0, // 4x4 transform
Tx8x8 = 1, // 8x8 transform
Tx16x16 = 2, // 16x16 transform
Tx32x32 = 3, // 32x32 transform
TxSizes = 4
};
enum class TxMode {
Only4X4 = 0, // Only 4x4 transform used
Allow8X8 = 1, // Allow block transform size up to 8x8
Allow16X16 = 2, // Allow block transform size up to 16x16
Allow32X32 = 3, // Allow block transform size up to 32x32
TxModeSelect = 4, // Transform specified for each block
TxModes = 5
};
struct Segmentation {
u8 enabled;
u8 update_map;
u8 temporal_update;
u8 abs_delta;
std::array<u32, 8> feature_mask;
std::array<std::array<s16, 4>, 8> feature_data;
};
static_assert(sizeof(Segmentation) == 0x64, "Segmentation is an invalid size");
struct LoopFilter {
u8 mode_ref_delta_enabled;
std::array<s8, 4> ref_deltas;
std::array<s8, 2> mode_deltas;
};
static_assert(sizeof(LoopFilter) == 0x7, "LoopFilter is an invalid size");
struct Vp9EntropyProbs {
std::array<u8, 36> y_mode_prob; ///< 0x0000
std::array<u8, 64> partition_prob; ///< 0x0024
std::array<u8, 1728> coef_probs; ///< 0x0064
std::array<u8, 8> switchable_interp_prob; ///< 0x0724
std::array<u8, 28> inter_mode_prob; ///< 0x072C
std::array<u8, 4> intra_inter_prob; ///< 0x0748
std::array<u8, 5> comp_inter_prob; ///< 0x074C
std::array<u8, 10> single_ref_prob; ///< 0x0751
std::array<u8, 5> comp_ref_prob; ///< 0x075B
std::array<u8, 6> tx_32x32_prob; ///< 0x0760
std::array<u8, 4> tx_16x16_prob; ///< 0x0766
std::array<u8, 2> tx_8x8_prob; ///< 0x076A
std::array<u8, 3> skip_probs; ///< 0x076C
std::array<u8, 3> joints; ///< 0x076F
std::array<u8, 2> sign; ///< 0x0772
std::array<u8, 20> classes; ///< 0x0774
std::array<u8, 2> class_0; ///< 0x0788
std::array<u8, 20> prob_bits; ///< 0x078A
std::array<u8, 12> class_0_fr; ///< 0x079E
std::array<u8, 6> fr; ///< 0x07AA
std::array<u8, 2> class_0_hp; ///< 0x07B0
std::array<u8, 2> high_precision; ///< 0x07B2
};
static_assert(sizeof(Vp9EntropyProbs) == 0x7B4, "Vp9EntropyProbs is an invalid size");
struct Vp9PictureInfo {
u32 bitstream_size;
std::array<u64, 4> frame_offsets;
std::array<s8, 4> ref_frame_sign_bias;
s32 base_q_index;
s32 y_dc_delta_q;
s32 uv_dc_delta_q;
s32 uv_ac_delta_q;
s32 transform_mode;
s32 interp_filter;
s32 reference_mode;
s32 log2_tile_cols;
s32 log2_tile_rows;
std::array<s8, 4> ref_deltas;
std::array<s8, 2> mode_deltas;
Vp9EntropyProbs entropy;
Vp9FrameDimensions frame_size;
u8 first_level;
u8 sharpness_level;
bool is_key_frame;
bool intra_only;
bool last_frame_was_key;
bool error_resilient_mode;
bool last_frame_shown;
bool show_frame;
bool lossless;
bool allow_high_precision_mv;
bool segment_enabled;
bool mode_ref_delta_enabled;
};
struct Vp9FrameContainer {
Vp9PictureInfo info{};
std::vector<u8> bit_stream;
};
struct PictureInfo {
INSERT_PADDING_WORDS_NOINIT(12); ///< 0x00
u32 bitstream_size; ///< 0x30
INSERT_PADDING_WORDS_NOINIT(5); ///< 0x34
Vp9FrameDimensions last_frame_size; ///< 0x48
Vp9FrameDimensions golden_frame_size; ///< 0x50
Vp9FrameDimensions alt_frame_size; ///< 0x58
Vp9FrameDimensions current_frame_size; ///< 0x60
FrameFlags vp9_flags; ///< 0x68
std::array<s8, 4> ref_frame_sign_bias; ///< 0x6C
u8 first_level; ///< 0x70
u8 sharpness_level; ///< 0x71
u8 base_q_index; ///< 0x72
u8 y_dc_delta_q; ///< 0x73
u8 uv_ac_delta_q; ///< 0x74
u8 uv_dc_delta_q; ///< 0x75
u8 lossless; ///< 0x76
u8 tx_mode; ///< 0x77
u8 allow_high_precision_mv; ///< 0x78
u8 interp_filter; ///< 0x79
u8 reference_mode; ///< 0x7A
INSERT_PADDING_BYTES_NOINIT(3); ///< 0x7B
u8 log2_tile_cols; ///< 0x7E
u8 log2_tile_rows; ///< 0x7F
Segmentation segmentation; ///< 0x80
LoopFilter loop_filter; ///< 0xE4
INSERT_PADDING_BYTES_NOINIT(21); ///< 0xEB
[[nodiscard]] Vp9PictureInfo Convert() const {
return {
.bitstream_size = bitstream_size,
.frame_offsets{},
.ref_frame_sign_bias = ref_frame_sign_bias,
.base_q_index = base_q_index,
.y_dc_delta_q = y_dc_delta_q,
.uv_dc_delta_q = uv_dc_delta_q,
.uv_ac_delta_q = uv_ac_delta_q,
.transform_mode = tx_mode,
.interp_filter = interp_filter,
.reference_mode = reference_mode,
.log2_tile_cols = log2_tile_cols,
.log2_tile_rows = log2_tile_rows,
.ref_deltas = loop_filter.ref_deltas,
.mode_deltas = loop_filter.mode_deltas,
.entropy{},
.frame_size = current_frame_size,
.first_level = first_level,
.sharpness_level = sharpness_level,
.is_key_frame = True(vp9_flags & FrameFlags::IsKeyFrame),
.intra_only = True(vp9_flags & FrameFlags::IntraOnly),
.last_frame_was_key = True(vp9_flags & FrameFlags::LastFrameIsKeyFrame),
.error_resilient_mode = True(vp9_flags & FrameFlags::ErrorResilientMode),
.last_frame_shown = True(vp9_flags & FrameFlags::LastShowFrame),
.show_frame = true,
.lossless = lossless != 0,
.allow_high_precision_mv = allow_high_precision_mv != 0,
.segment_enabled = segmentation.enabled != 0,
.mode_ref_delta_enabled = loop_filter.mode_ref_delta_enabled != 0,
};
}
};
static_assert(sizeof(PictureInfo) == 0x100, "PictureInfo is an invalid size");
struct EntropyProbs {
INSERT_PADDING_BYTES_NOINIT(1024); ///< 0x0000
std::array<u8, 28> inter_mode_prob; ///< 0x0400
std::array<u8, 4> intra_inter_prob; ///< 0x041C
INSERT_PADDING_BYTES_NOINIT(80); ///< 0x0420
std::array<u8, 2> tx_8x8_prob; ///< 0x0470
std::array<u8, 4> tx_16x16_prob; ///< 0x0472
std::array<u8, 6> tx_32x32_prob; ///< 0x0476
std::array<u8, 4> y_mode_prob_e8; ///< 0x047C
std::array<std::array<u8, 8>, 4> y_mode_prob_e0e7; ///< 0x0480
INSERT_PADDING_BYTES_NOINIT(64); ///< 0x04A0
std::array<u8, 64> partition_prob; ///< 0x04E0
INSERT_PADDING_BYTES_NOINIT(10); ///< 0x0520
std::array<u8, 8> switchable_interp_prob; ///< 0x052A
std::array<u8, 5> comp_inter_prob; ///< 0x0532
std::array<u8, 3> skip_probs; ///< 0x0537
INSERT_PADDING_BYTES_NOINIT(1); ///< 0x053A
std::array<u8, 3> joints; ///< 0x053B
std::array<u8, 2> sign; ///< 0x053E
std::array<u8, 2> class_0; ///< 0x0540
std::array<u8, 6> fr; ///< 0x0542
std::array<u8, 2> class_0_hp; ///< 0x0548
std::array<u8, 2> high_precision; ///< 0x054A
std::array<u8, 20> classes; ///< 0x054C
std::array<u8, 12> class_0_fr; ///< 0x0560
std::array<u8, 20> pred_bits; ///< 0x056C
std::array<u8, 10> single_ref_prob; ///< 0x0580
std::array<u8, 5> comp_ref_prob; ///< 0x058A
INSERT_PADDING_BYTES_NOINIT(17); ///< 0x058F
std::array<u8, 2304> coef_probs; ///< 0x05A0
void Convert(Vp9EntropyProbs& fc) {
fc.inter_mode_prob = inter_mode_prob;
fc.intra_inter_prob = intra_inter_prob;
fc.tx_8x8_prob = tx_8x8_prob;
fc.tx_16x16_prob = tx_16x16_prob;
fc.tx_32x32_prob = tx_32x32_prob;
for (std::size_t i = 0; i < 4; i++) {
for (std::size_t j = 0; j < 9; j++) {
fc.y_mode_prob[j + 9 * i] = j < 8 ? y_mode_prob_e0e7[i][j] : y_mode_prob_e8[i];
}
}
fc.partition_prob = partition_prob;
fc.switchable_interp_prob = switchable_interp_prob;
fc.comp_inter_prob = comp_inter_prob;
fc.skip_probs = skip_probs;
fc.joints = joints;
fc.sign = sign;
fc.class_0 = class_0;
fc.fr = fr;
fc.class_0_hp = class_0_hp;
fc.high_precision = high_precision;
fc.classes = classes;
fc.class_0_fr = class_0_fr;
fc.prob_bits = pred_bits;
fc.single_ref_prob = single_ref_prob;
fc.comp_ref_prob = comp_ref_prob;
// Skip the 4th element as it goes unused
for (std::size_t i = 0; i < coef_probs.size(); i += 4) {
const std::size_t j = i - i / 4;
fc.coef_probs[j] = coef_probs[i];
fc.coef_probs[j + 1] = coef_probs[i + 1];
fc.coef_probs[j + 2] = coef_probs[i + 2];
}
}
};
static_assert(sizeof(EntropyProbs) == 0xEA0, "EntropyProbs is an invalid size");
enum class Ref { Last, Golden, AltRef };
struct RefPoolElement {
s64 frame{};
Ref ref{};
bool refresh{};
};
#define ASSERT_POSITION(field_name, position) \
static_assert(offsetof(Vp9EntropyProbs, field_name) == position, \
"Field " #field_name " has invalid position")
ASSERT_POSITION(partition_prob, 0x0024);
ASSERT_POSITION(switchable_interp_prob, 0x0724);
ASSERT_POSITION(sign, 0x0772);
ASSERT_POSITION(class_0_fr, 0x079E);
ASSERT_POSITION(high_precision, 0x07B2);
#undef ASSERT_POSITION
#define ASSERT_POSITION(field_name, position) \
static_assert(offsetof(PictureInfo, field_name) == position, \
"Field " #field_name " has invalid position")
ASSERT_POSITION(bitstream_size, 0x30);
ASSERT_POSITION(last_frame_size, 0x48);
ASSERT_POSITION(first_level, 0x70);
ASSERT_POSITION(segmentation, 0x80);
ASSERT_POSITION(loop_filter, 0xE4);
#undef ASSERT_POSITION
#define ASSERT_POSITION(field_name, position) \
static_assert(offsetof(EntropyProbs, field_name) == position, \
"Field " #field_name " has invalid position")
ASSERT_POSITION(inter_mode_prob, 0x400);
ASSERT_POSITION(tx_8x8_prob, 0x470);
ASSERT_POSITION(partition_prob, 0x4E0);
ASSERT_POSITION(class_0, 0x540);
ASSERT_POSITION(class_0_fr, 0x560);
ASSERT_POSITION(coef_probs, 0x5A0);
#undef ASSERT_POSITION
}; // namespace Decoder
}; // namespace Tegra
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <array>
#include <vector>
#include "common/common_funcs.h"
#include "common/common_types.h"
namespace Tegra {
namespace Decoder {
struct Vp9FrameDimensions {
s16 width;
s16 height;
s16 luma_pitch;
s16 chroma_pitch;
};
static_assert(sizeof(Vp9FrameDimensions) == 0x8, "Vp9 Vp9FrameDimensions is an invalid size");
enum class FrameFlags : u32 {
IsKeyFrame = 1 << 0,
LastFrameIsKeyFrame = 1 << 1,
FrameSizeChanged = 1 << 2,
ErrorResilientMode = 1 << 3,
LastShowFrame = 1 << 4,
IntraOnly = 1 << 5,
};
DECLARE_ENUM_FLAG_OPERATORS(FrameFlags)
enum class TxSize {
Tx4x4 = 0, // 4x4 transform
Tx8x8 = 1, // 8x8 transform
Tx16x16 = 2, // 16x16 transform
Tx32x32 = 3, // 32x32 transform
TxSizes = 4
};
enum class TxMode {
Only4X4 = 0, // Only 4x4 transform used
Allow8X8 = 1, // Allow block transform size up to 8x8
Allow16X16 = 2, // Allow block transform size up to 16x16
Allow32X32 = 3, // Allow block transform size up to 32x32
TxModeSelect = 4, // Transform specified for each block
TxModes = 5
};
struct Segmentation {
u8 enabled;
u8 update_map;
u8 temporal_update;
u8 abs_delta;
std::array<u32, 8> feature_mask;
std::array<std::array<s16, 4>, 8> feature_data;
};
static_assert(sizeof(Segmentation) == 0x64, "Segmentation is an invalid size");
struct LoopFilter {
u8 mode_ref_delta_enabled;
std::array<s8, 4> ref_deltas;
std::array<s8, 2> mode_deltas;
};
static_assert(sizeof(LoopFilter) == 0x7, "LoopFilter is an invalid size");
struct Vp9EntropyProbs {
std::array<u8, 36> y_mode_prob; ///< 0x0000
std::array<u8, 64> partition_prob; ///< 0x0024
std::array<u8, 1728> coef_probs; ///< 0x0064
std::array<u8, 8> switchable_interp_prob; ///< 0x0724
std::array<u8, 28> inter_mode_prob; ///< 0x072C
std::array<u8, 4> intra_inter_prob; ///< 0x0748
std::array<u8, 5> comp_inter_prob; ///< 0x074C
std::array<u8, 10> single_ref_prob; ///< 0x0751
std::array<u8, 5> comp_ref_prob; ///< 0x075B
std::array<u8, 6> tx_32x32_prob; ///< 0x0760
std::array<u8, 4> tx_16x16_prob; ///< 0x0766
std::array<u8, 2> tx_8x8_prob; ///< 0x076A
std::array<u8, 3> skip_probs; ///< 0x076C
std::array<u8, 3> joints; ///< 0x076F
std::array<u8, 2> sign; ///< 0x0772
std::array<u8, 20> classes; ///< 0x0774
std::array<u8, 2> class_0; ///< 0x0788
std::array<u8, 20> prob_bits; ///< 0x078A
std::array<u8, 12> class_0_fr; ///< 0x079E
std::array<u8, 6> fr; ///< 0x07AA
std::array<u8, 2> class_0_hp; ///< 0x07B0
std::array<u8, 2> high_precision; ///< 0x07B2
};
static_assert(sizeof(Vp9EntropyProbs) == 0x7B4, "Vp9EntropyProbs is an invalid size");
struct Vp9PictureInfo {
u32 bitstream_size;
std::array<u64, 4> frame_offsets;
std::array<s8, 4> ref_frame_sign_bias;
s32 base_q_index;
s32 y_dc_delta_q;
s32 uv_dc_delta_q;
s32 uv_ac_delta_q;
s32 transform_mode;
s32 interp_filter;
s32 reference_mode;
s32 log2_tile_cols;
s32 log2_tile_rows;
std::array<s8, 4> ref_deltas;
std::array<s8, 2> mode_deltas;
Vp9EntropyProbs entropy;
Vp9FrameDimensions frame_size;
u8 first_level;
u8 sharpness_level;
bool is_key_frame;
bool intra_only;
bool last_frame_was_key;
bool error_resilient_mode;
bool last_frame_shown;
bool show_frame;
bool lossless;
bool allow_high_precision_mv;
bool segment_enabled;
bool mode_ref_delta_enabled;
};
struct Vp9FrameContainer {
Vp9PictureInfo info{};
std::vector<u8> bit_stream;
};
struct PictureInfo {
INSERT_PADDING_WORDS_NOINIT(12); ///< 0x00
u32 bitstream_size; ///< 0x30
INSERT_PADDING_WORDS_NOINIT(5); ///< 0x34
Vp9FrameDimensions last_frame_size; ///< 0x48
Vp9FrameDimensions golden_frame_size; ///< 0x50
Vp9FrameDimensions alt_frame_size; ///< 0x58
Vp9FrameDimensions current_frame_size; ///< 0x60
FrameFlags vp9_flags; ///< 0x68
std::array<s8, 4> ref_frame_sign_bias; ///< 0x6C
u8 first_level; ///< 0x70
u8 sharpness_level; ///< 0x71
u8 base_q_index; ///< 0x72
u8 y_dc_delta_q; ///< 0x73
u8 uv_ac_delta_q; ///< 0x74
u8 uv_dc_delta_q; ///< 0x75
u8 lossless; ///< 0x76
u8 tx_mode; ///< 0x77
u8 allow_high_precision_mv; ///< 0x78
u8 interp_filter; ///< 0x79
u8 reference_mode; ///< 0x7A
INSERT_PADDING_BYTES_NOINIT(3); ///< 0x7B
u8 log2_tile_cols; ///< 0x7E
u8 log2_tile_rows; ///< 0x7F
Segmentation segmentation; ///< 0x80
LoopFilter loop_filter; ///< 0xE4
INSERT_PADDING_BYTES_NOINIT(21); ///< 0xEB
[[nodiscard]] Vp9PictureInfo Convert() const {
return {
.bitstream_size = bitstream_size,
.frame_offsets{},
.ref_frame_sign_bias = ref_frame_sign_bias,
.base_q_index = base_q_index,
.y_dc_delta_q = y_dc_delta_q,
.uv_dc_delta_q = uv_dc_delta_q,
.uv_ac_delta_q = uv_ac_delta_q,
.transform_mode = tx_mode,
.interp_filter = interp_filter,
.reference_mode = reference_mode,
.log2_tile_cols = log2_tile_cols,
.log2_tile_rows = log2_tile_rows,
.ref_deltas = loop_filter.ref_deltas,
.mode_deltas = loop_filter.mode_deltas,
.entropy{},
.frame_size = current_frame_size,
.first_level = first_level,
.sharpness_level = sharpness_level,
.is_key_frame = True(vp9_flags & FrameFlags::IsKeyFrame),
.intra_only = True(vp9_flags & FrameFlags::IntraOnly),
.last_frame_was_key = True(vp9_flags & FrameFlags::LastFrameIsKeyFrame),
.error_resilient_mode = True(vp9_flags & FrameFlags::ErrorResilientMode),
.last_frame_shown = True(vp9_flags & FrameFlags::LastShowFrame),
.show_frame = true,
.lossless = lossless != 0,
.allow_high_precision_mv = allow_high_precision_mv != 0,
.segment_enabled = segmentation.enabled != 0,
.mode_ref_delta_enabled = loop_filter.mode_ref_delta_enabled != 0,
};
}
};
static_assert(sizeof(PictureInfo) == 0x100, "PictureInfo is an invalid size");
struct EntropyProbs {
INSERT_PADDING_BYTES_NOINIT(1024); ///< 0x0000
std::array<u8, 28> inter_mode_prob; ///< 0x0400
std::array<u8, 4> intra_inter_prob; ///< 0x041C
INSERT_PADDING_BYTES_NOINIT(80); ///< 0x0420
std::array<u8, 2> tx_8x8_prob; ///< 0x0470
std::array<u8, 4> tx_16x16_prob; ///< 0x0472
std::array<u8, 6> tx_32x32_prob; ///< 0x0476
std::array<u8, 4> y_mode_prob_e8; ///< 0x047C
std::array<std::array<u8, 8>, 4> y_mode_prob_e0e7; ///< 0x0480
INSERT_PADDING_BYTES_NOINIT(64); ///< 0x04A0
std::array<u8, 64> partition_prob; ///< 0x04E0
INSERT_PADDING_BYTES_NOINIT(10); ///< 0x0520
std::array<u8, 8> switchable_interp_prob; ///< 0x052A
std::array<u8, 5> comp_inter_prob; ///< 0x0532
std::array<u8, 3> skip_probs; ///< 0x0537
INSERT_PADDING_BYTES_NOINIT(1); ///< 0x053A
std::array<u8, 3> joints; ///< 0x053B
std::array<u8, 2> sign; ///< 0x053E
std::array<u8, 2> class_0; ///< 0x0540
std::array<u8, 6> fr; ///< 0x0542
std::array<u8, 2> class_0_hp; ///< 0x0548
std::array<u8, 2> high_precision; ///< 0x054A
std::array<u8, 20> classes; ///< 0x054C
std::array<u8, 12> class_0_fr; ///< 0x0560
std::array<u8, 20> pred_bits; ///< 0x056C
std::array<u8, 10> single_ref_prob; ///< 0x0580
std::array<u8, 5> comp_ref_prob; ///< 0x058A
INSERT_PADDING_BYTES_NOINIT(17); ///< 0x058F
std::array<u8, 2304> coef_probs; ///< 0x05A0
void Convert(Vp9EntropyProbs& fc) {
fc.inter_mode_prob = inter_mode_prob;
fc.intra_inter_prob = intra_inter_prob;
fc.tx_8x8_prob = tx_8x8_prob;
fc.tx_16x16_prob = tx_16x16_prob;
fc.tx_32x32_prob = tx_32x32_prob;
for (std::size_t i = 0; i < 4; i++) {
for (std::size_t j = 0; j < 9; j++) {
fc.y_mode_prob[j + 9 * i] = j < 8 ? y_mode_prob_e0e7[i][j] : y_mode_prob_e8[i];
}
}
fc.partition_prob = partition_prob;
fc.switchable_interp_prob = switchable_interp_prob;
fc.comp_inter_prob = comp_inter_prob;
fc.skip_probs = skip_probs;
fc.joints = joints;
fc.sign = sign;
fc.class_0 = class_0;
fc.fr = fr;
fc.class_0_hp = class_0_hp;
fc.high_precision = high_precision;
fc.classes = classes;
fc.class_0_fr = class_0_fr;
fc.prob_bits = pred_bits;
fc.single_ref_prob = single_ref_prob;
fc.comp_ref_prob = comp_ref_prob;
// Skip the 4th element as it goes unused
for (std::size_t i = 0; i < coef_probs.size(); i += 4) {
const std::size_t j = i - i / 4;
fc.coef_probs[j] = coef_probs[i];
fc.coef_probs[j + 1] = coef_probs[i + 1];
fc.coef_probs[j + 2] = coef_probs[i + 2];
}
}
};
static_assert(sizeof(EntropyProbs) == 0xEA0, "EntropyProbs is an invalid size");
enum class Ref { Last, Golden, AltRef };
struct RefPoolElement {
s64 frame{};
Ref ref{};
bool refresh{};
};
#define ASSERT_POSITION(field_name, position) \
static_assert(offsetof(Vp9EntropyProbs, field_name) == position, \
"Field " #field_name " has invalid position")
ASSERT_POSITION(partition_prob, 0x0024);
ASSERT_POSITION(switchable_interp_prob, 0x0724);
ASSERT_POSITION(sign, 0x0772);
ASSERT_POSITION(class_0_fr, 0x079E);
ASSERT_POSITION(high_precision, 0x07B2);
#undef ASSERT_POSITION
#define ASSERT_POSITION(field_name, position) \
static_assert(offsetof(PictureInfo, field_name) == position, \
"Field " #field_name " has invalid position")
ASSERT_POSITION(bitstream_size, 0x30);
ASSERT_POSITION(last_frame_size, 0x48);
ASSERT_POSITION(first_level, 0x70);
ASSERT_POSITION(segmentation, 0x80);
ASSERT_POSITION(loop_filter, 0xE4);
#undef ASSERT_POSITION
#define ASSERT_POSITION(field_name, position) \
static_assert(offsetof(EntropyProbs, field_name) == position, \
"Field " #field_name " has invalid position")
ASSERT_POSITION(inter_mode_prob, 0x400);
ASSERT_POSITION(tx_8x8_prob, 0x470);
ASSERT_POSITION(partition_prob, 0x4E0);
ASSERT_POSITION(class_0, 0x540);
ASSERT_POSITION(class_0_fr, 0x560);
ASSERT_POSITION(coef_probs, 0x5A0);
#undef ASSERT_POSITION
}; // namespace Decoder
}; // namespace Tegra

66
src/video_core/host1x/control.cpp
View File

@@ -1,33 +1,33 @@
// SPDX-FileCopyrightText: 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#include "common/assert.h"
#include "video_core/host1x/control.h"
#include "video_core/host1x/host1x.h"
namespace Tegra::Host1x {
Control::Control(Host1x& host1x_) : host1x(host1x_) {}
Control::~Control() = default;
void Control::ProcessMethod(Method method, u32 argument) {
switch (method) {
case Method::LoadSyncptPayload32:
syncpoint_value = argument;
break;
case Method::WaitSyncpt:
case Method::WaitSyncpt32:
Execute(argument);
break;
default:
UNIMPLEMENTED_MSG("Control method 0x{:X}", static_cast<u32>(method));
break;
}
}
void Control::Execute(u32 data) {
host1x.GetSyncpointManager().WaitHost(data, syncpoint_value);
}
} // namespace Tegra::Host1x
// SPDX-FileCopyrightText: 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#include "common/assert.h"
#include "video_core/host1x/control.h"
#include "video_core/host1x/host1x.h"
namespace Tegra::Host1x {
Control::Control(Host1x& host1x_) : host1x(host1x_) {}
Control::~Control() = default;
void Control::ProcessMethod(Method method, u32 argument) {
switch (method) {
case Method::LoadSyncptPayload32:
syncpoint_value = argument;
break;
case Method::WaitSyncpt:
case Method::WaitSyncpt32:
Execute(argument);
break;
default:
UNIMPLEMENTED_MSG("Control method 0x{:X}", static_cast<u32>(method));
break;
}
}
void Control::Execute(u32 data) {
host1x.GetSyncpointManager().WaitHost(data, syncpoint_value);
}
} // namespace Tegra::Host1x

80
src/video_core/host1x/control.h
View File

@@ -1,40 +1,40 @@
// SPDX-FileCopyrightText: 2021 yuzu Emulator Project
// SPDX-FileCopyrightText: 2021 Skyline Team and Contributors
// SPDX-License-Identifier: GPL-3.0-or-later
#pragma once
#include "common/common_types.h"
namespace Tegra {
namespace Host1x {
class Host1x;
class Nvdec;
class Control {
public:
enum class Method : u32 {
WaitSyncpt = 0x8,
LoadSyncptPayload32 = 0x4e,
WaitSyncpt32 = 0x50,
};
explicit Control(Host1x& host1x);
~Control();
/// Writes the method into the state, Invoke Execute() if encountered
void ProcessMethod(Method method, u32 argument);
private:
/// For Host1x, execute is waiting on a syncpoint previously written into the state
void Execute(u32 data);
u32 syncpoint_value{};
Host1x& host1x;
};
} // namespace Host1x
} // namespace Tegra
// SPDX-FileCopyrightText: 2021 yuzu Emulator Project
// SPDX-FileCopyrightText: 2021 Skyline Team and Contributors
// SPDX-License-Identifier: GPL-3.0-or-later
#pragma once
#include "common/common_types.h"
namespace Tegra {
namespace Host1x {
class Host1x;
class Nvdec;
class Control {
public:
enum class Method : u32 {
WaitSyncpt = 0x8,
LoadSyncptPayload32 = 0x4e,
WaitSyncpt32 = 0x50,
};
explicit Control(Host1x& host1x);
~Control();
/// Writes the method into the state, Invoke Execute() if encountered
void ProcessMethod(Method method, u32 argument);
private:
/// For Host1x, execute is waiting on a syncpoint previously written into the state
void Execute(u32 data);
u32 syncpoint_value{};
Host1x& host1x;
};
} // namespace Host1x
} // namespace Tegra

34
src/video_core/host1x/host1x.cpp
View File

@@ -1,17 +1,17 @@
// SPDX-FileCopyrightText: 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#include "core/core.h"
#include "video_core/host1x/host1x.h"
namespace Tegra {
namespace Host1x {
Host1x::Host1x(Core::System& system_)
: system{system_}, syncpoint_manager{}, memory_manager{system, 32, 12},
allocator{std::make_unique<Common::FlatAllocator<u32, 0, 32>>(1 << 12)} {}
} // namespace Host1x
} // namespace Tegra
// SPDX-FileCopyrightText: 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#include "core/core.h"
#include "video_core/host1x/host1x.h"
namespace Tegra {
namespace Host1x {
Host1x::Host1x(Core::System& system_)
: system{system_}, syncpoint_manager{}, memory_manager{system, 32, 12},
allocator{std::make_unique<Common::FlatAllocator<u32, 0, 32>>(1 << 12)} {}
} // namespace Host1x
} // namespace Tegra

114
src/video_core/host1x/host1x.h
View File

@@ -1,57 +1,57 @@
// SPDX-FileCopyrightText: 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#pragma once
#include "common/common_types.h"
#include "common/address_space.h"
#include "video_core/host1x/syncpoint_manager.h"
#include "video_core/memory_manager.h"
namespace Core {
class System;
} // namespace Core
namespace Tegra {
namespace Host1x {
class Host1x {
public:
explicit Host1x(Core::System& system);
SyncpointManager& GetSyncpointManager() {
return syncpoint_manager;
}
const SyncpointManager& GetSyncpointManager() const {
return syncpoint_manager;
}
Tegra::MemoryManager& MemoryManager() {
return memory_manager;
}
const Tegra::MemoryManager& MemoryManager() const {
return memory_manager;
}
Common::FlatAllocator<u32, 0, 32>& Allocator() {
return *allocator;
}
const Common::FlatAllocator<u32, 0, 32>& Allocator() const {
return *allocator;
}
private:
Core::System& system;
SyncpointManager syncpoint_manager;
Tegra::MemoryManager memory_manager;
std::unique_ptr<Common::FlatAllocator<u32, 0, 32>> allocator;
};
} // namespace Host1x
} // namespace Tegra
// SPDX-FileCopyrightText: 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#pragma once
#include "common/common_types.h"
#include "common/address_space.h"
#include "video_core/host1x/syncpoint_manager.h"
#include "video_core/memory_manager.h"
namespace Core {
class System;
} // namespace Core
namespace Tegra {
namespace Host1x {
class Host1x {
public:
explicit Host1x(Core::System& system);
SyncpointManager& GetSyncpointManager() {
return syncpoint_manager;
}
const SyncpointManager& GetSyncpointManager() const {
return syncpoint_manager;
}
Tegra::MemoryManager& MemoryManager() {
return memory_manager;
}
const Tegra::MemoryManager& MemoryManager() const {
return memory_manager;
}
Common::FlatAllocator<u32, 0, 32>& Allocator() {
return *allocator;
}
const Common::FlatAllocator<u32, 0, 32>& Allocator() const {
return *allocator;
}
private:
Core::System& system;
SyncpointManager syncpoint_manager;
Tegra::MemoryManager memory_manager;
std::unique_ptr<Common::FlatAllocator<u32, 0, 32>> allocator;
};
} // namespace Host1x
} // namespace Tegra

96
src/video_core/host1x/nvdec.cpp
View File

@@ -1,48 +1,48 @@
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "common/assert.h"
#include "video_core/host1x/host1x.h"
#include "video_core/host1x/nvdec.h"
namespace Tegra::Host1x {
#define NVDEC_REG_INDEX(field_name) \
(offsetof(NvdecCommon::NvdecRegisters, field_name) / sizeof(u64))
Nvdec::Nvdec(Host1x& host1x_)
: host1x(host1x_), state{}, codec(std::make_unique<Codec>(host1x, state)) {}
Nvdec::~Nvdec() = default;
void Nvdec::ProcessMethod(u32 method, u32 argument) {
state.reg_array[method] = static_cast<u64>(argument) << 8;
switch (method) {
case NVDEC_REG_INDEX(set_codec_id):
codec->SetTargetCodec(static_cast<NvdecCommon::VideoCodec>(argument));
break;
case NVDEC_REG_INDEX(execute):
Execute();
break;
}
}
AVFramePtr Nvdec::GetFrame() {
return codec->GetCurrentFrame();
}
void Nvdec::Execute() {
switch (codec->GetCurrentCodec()) {
case NvdecCommon::VideoCodec::H264:
case NvdecCommon::VideoCodec::VP8:
case NvdecCommon::VideoCodec::VP9:
codec->Decode();
break;
default:
UNIMPLEMENTED_MSG("Codec {}", codec->GetCurrentCodecName());
break;
}
}
} // namespace Tegra::Host1x
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "common/assert.h"
#include "video_core/host1x/host1x.h"
#include "video_core/host1x/nvdec.h"
namespace Tegra::Host1x {
#define NVDEC_REG_INDEX(field_name) \
(offsetof(NvdecCommon::NvdecRegisters, field_name) / sizeof(u64))
Nvdec::Nvdec(Host1x& host1x_)
: host1x(host1x_), state{}, codec(std::make_unique<Codec>(host1x, state)) {}
Nvdec::~Nvdec() = default;
void Nvdec::ProcessMethod(u32 method, u32 argument) {
state.reg_array[method] = static_cast<u64>(argument) << 8;
switch (method) {
case NVDEC_REG_INDEX(set_codec_id):
codec->SetTargetCodec(static_cast<NvdecCommon::VideoCodec>(argument));
break;
case NVDEC_REG_INDEX(execute):
Execute();
break;
}
}
AVFramePtr Nvdec::GetFrame() {
return codec->GetCurrentFrame();
}
void Nvdec::Execute() {
switch (codec->GetCurrentCodec()) {
case NvdecCommon::VideoCodec::H264:
case NvdecCommon::VideoCodec::VP8:
case NvdecCommon::VideoCodec::VP9:
codec->Decode();
break;
default:
UNIMPLEMENTED_MSG("Codec {}", codec->GetCurrentCodecName());
break;
}
}
} // namespace Tegra::Host1x

78
src/video_core/host1x/nvdec.h
View File

@@ -1,39 +1,39 @@
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <memory>
#include <vector>
#include "common/common_types.h"
#include "video_core/host1x/codecs/codec.h"
namespace Tegra {
namespace Host1x {
class Host1x;
class Nvdec {
public:
explicit Nvdec(Host1x& host1x);
~Nvdec();
/// Writes the method into the state, Invoke Execute() if encountered
void ProcessMethod(u32 method, u32 argument);
/// Return most recently decoded frame
[[nodiscard]] AVFramePtr GetFrame();
private:
/// Invoke codec to decode a frame
void Execute();
Host1x& host1x;
NvdecCommon::NvdecRegisters state;
std::unique_ptr<Codec> codec;
};
} // namespace Host1x
} // namespace Tegra
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <memory>
#include <vector>
#include "common/common_types.h"
#include "video_core/host1x/codecs/codec.h"
namespace Tegra {
namespace Host1x {
class Host1x;
class Nvdec {
public:
explicit Nvdec(Host1x& host1x);
~Nvdec();
/// Writes the method into the state, Invoke Execute() if encountered
void ProcessMethod(u32 method, u32 argument);
/// Return most recently decoded frame
[[nodiscard]] AVFramePtr GetFrame();
private:
/// Invoke codec to decode a frame
void Execute();
Host1x& host1x;
NvdecCommon::NvdecRegisters state;
std::unique_ptr<Codec> codec;
};
} // namespace Host1x
} // namespace Tegra

194
src/video_core/host1x/nvdec_common.h
View File

@@ -1,97 +1,97 @@
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
namespace Tegra::Host1x::NvdecCommon {
enum class VideoCodec : u64 {
None = 0x0,
H264 = 0x3,
VP8 = 0x5,
H265 = 0x7,
VP9 = 0x9,
};
// NVDEC should use a 32-bit address space, but is mapped to 64-bit,
// doubling the sizes here is compensating for that.
struct NvdecRegisters {
static constexpr std::size_t NUM_REGS = 0x178;
union {
struct {
INSERT_PADDING_WORDS_NOINIT(256); ///< 0x0000
VideoCodec set_codec_id; ///< 0x0400
INSERT_PADDING_WORDS_NOINIT(126); ///< 0x0408
u64 execute; ///< 0x0600
INSERT_PADDING_WORDS_NOINIT(126); ///< 0x0608
struct { ///< 0x0800
union {
BitField<0, 3, VideoCodec> codec;
BitField<4, 1, u64> gp_timer_on;
BitField<13, 1, u64> mb_timer_on;
BitField<14, 1, u64> intra_frame_pslc;
BitField<17, 1, u64> all_intra_frame;
};
} control_params;
u64 picture_info_offset; ///< 0x0808
u64 frame_bitstream_offset; ///< 0x0810
u64 frame_number; ///< 0x0818
u64 h264_slice_data_offsets; ///< 0x0820
u64 h264_mv_dump_offset; ///< 0x0828
INSERT_PADDING_WORDS_NOINIT(6); ///< 0x0830
u64 frame_stats_offset; ///< 0x0848
u64 h264_last_surface_luma_offset; ///< 0x0850
u64 h264_last_surface_chroma_offset; ///< 0x0858
std::array<u64, 17> surface_luma_offset; ///< 0x0860
std::array<u64, 17> surface_chroma_offset; ///< 0x08E8
INSERT_PADDING_WORDS_NOINIT(68); ///< 0x0970
u64 vp8_prob_data_offset; ///< 0x0A80
u64 vp8_header_partition_buf_offset; ///< 0x0A88
INSERT_PADDING_WORDS_NOINIT(60); ///< 0x0A90
u64 vp9_entropy_probs_offset; ///< 0x0B80
u64 vp9_backward_updates_offset; ///< 0x0B88
u64 vp9_last_frame_segmap_offset; ///< 0x0B90
u64 vp9_curr_frame_segmap_offset; ///< 0x0B98
INSERT_PADDING_WORDS_NOINIT(2); ///< 0x0BA0
u64 vp9_last_frame_mvs_offset; ///< 0x0BA8
u64 vp9_curr_frame_mvs_offset; ///< 0x0BB0
INSERT_PADDING_WORDS_NOINIT(2); ///< 0x0BB8
};
std::array<u64, NUM_REGS> reg_array;
};
};
static_assert(sizeof(NvdecRegisters) == (0xBC0), "NvdecRegisters is incorrect size");
#define ASSERT_REG_POSITION(field_name, position) \
static_assert(offsetof(NvdecRegisters, field_name) == position * sizeof(u64), \
"Field " #field_name " has invalid position")
ASSERT_REG_POSITION(set_codec_id, 0x80);
ASSERT_REG_POSITION(execute, 0xC0);
ASSERT_REG_POSITION(control_params, 0x100);
ASSERT_REG_POSITION(picture_info_offset, 0x101);
ASSERT_REG_POSITION(frame_bitstream_offset, 0x102);
ASSERT_REG_POSITION(frame_number, 0x103);
ASSERT_REG_POSITION(h264_slice_data_offsets, 0x104);
ASSERT_REG_POSITION(frame_stats_offset, 0x109);
ASSERT_REG_POSITION(h264_last_surface_luma_offset, 0x10A);
ASSERT_REG_POSITION(h264_last_surface_chroma_offset, 0x10B);
ASSERT_REG_POSITION(surface_luma_offset, 0x10C);
ASSERT_REG_POSITION(surface_chroma_offset, 0x11D);
ASSERT_REG_POSITION(vp8_prob_data_offset, 0x150);
ASSERT_REG_POSITION(vp8_header_partition_buf_offset, 0x151);
ASSERT_REG_POSITION(vp9_entropy_probs_offset, 0x170);
ASSERT_REG_POSITION(vp9_backward_updates_offset, 0x171);
ASSERT_REG_POSITION(vp9_last_frame_segmap_offset, 0x172);
ASSERT_REG_POSITION(vp9_curr_frame_segmap_offset, 0x173);
ASSERT_REG_POSITION(vp9_last_frame_mvs_offset, 0x175);
ASSERT_REG_POSITION(vp9_curr_frame_mvs_offset, 0x176);
#undef ASSERT_REG_POSITION
} // namespace Tegra::Host1x::NvdecCommon
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
namespace Tegra::Host1x::NvdecCommon {
enum class VideoCodec : u64 {
None = 0x0,
H264 = 0x3,
VP8 = 0x5,
H265 = 0x7,
VP9 = 0x9,
};
// NVDEC should use a 32-bit address space, but is mapped to 64-bit,
// doubling the sizes here is compensating for that.
struct NvdecRegisters {
static constexpr std::size_t NUM_REGS = 0x178;
union {
struct {
INSERT_PADDING_WORDS_NOINIT(256); ///< 0x0000
VideoCodec set_codec_id; ///< 0x0400
INSERT_PADDING_WORDS_NOINIT(126); ///< 0x0408
u64 execute; ///< 0x0600
INSERT_PADDING_WORDS_NOINIT(126); ///< 0x0608
struct { ///< 0x0800
union {
BitField<0, 3, VideoCodec> codec;
BitField<4, 1, u64> gp_timer_on;
BitField<13, 1, u64> mb_timer_on;
BitField<14, 1, u64> intra_frame_pslc;
BitField<17, 1, u64> all_intra_frame;
};
} control_params;
u64 picture_info_offset; ///< 0x0808
u64 frame_bitstream_offset; ///< 0x0810
u64 frame_number; ///< 0x0818
u64 h264_slice_data_offsets; ///< 0x0820
u64 h264_mv_dump_offset; ///< 0x0828
INSERT_PADDING_WORDS_NOINIT(6); ///< 0x0830
u64 frame_stats_offset; ///< 0x0848
u64 h264_last_surface_luma_offset; ///< 0x0850
u64 h264_last_surface_chroma_offset; ///< 0x0858
std::array<u64, 17> surface_luma_offset; ///< 0x0860
std::array<u64, 17> surface_chroma_offset; ///< 0x08E8
INSERT_PADDING_WORDS_NOINIT(68); ///< 0x0970
u64 vp8_prob_data_offset; ///< 0x0A80
u64 vp8_header_partition_buf_offset; ///< 0x0A88
INSERT_PADDING_WORDS_NOINIT(60); ///< 0x0A90
u64 vp9_entropy_probs_offset; ///< 0x0B80
u64 vp9_backward_updates_offset; ///< 0x0B88
u64 vp9_last_frame_segmap_offset; ///< 0x0B90
u64 vp9_curr_frame_segmap_offset; ///< 0x0B98
INSERT_PADDING_WORDS_NOINIT(2); ///< 0x0BA0
u64 vp9_last_frame_mvs_offset; ///< 0x0BA8
u64 vp9_curr_frame_mvs_offset; ///< 0x0BB0
INSERT_PADDING_WORDS_NOINIT(2); ///< 0x0BB8
};
std::array<u64, NUM_REGS> reg_array;
};
};
static_assert(sizeof(NvdecRegisters) == (0xBC0), "NvdecRegisters is incorrect size");
#define ASSERT_REG_POSITION(field_name, position) \
static_assert(offsetof(NvdecRegisters, field_name) == position * sizeof(u64), \
"Field " #field_name " has invalid position")
ASSERT_REG_POSITION(set_codec_id, 0x80);
ASSERT_REG_POSITION(execute, 0xC0);
ASSERT_REG_POSITION(control_params, 0x100);
ASSERT_REG_POSITION(picture_info_offset, 0x101);
ASSERT_REG_POSITION(frame_bitstream_offset, 0x102);
ASSERT_REG_POSITION(frame_number, 0x103);
ASSERT_REG_POSITION(h264_slice_data_offsets, 0x104);
ASSERT_REG_POSITION(frame_stats_offset, 0x109);
ASSERT_REG_POSITION(h264_last_surface_luma_offset, 0x10A);
ASSERT_REG_POSITION(h264_last_surface_chroma_offset, 0x10B);
ASSERT_REG_POSITION(surface_luma_offset, 0x10C);
ASSERT_REG_POSITION(surface_chroma_offset, 0x11D);
ASSERT_REG_POSITION(vp8_prob_data_offset, 0x150);
ASSERT_REG_POSITION(vp8_header_partition_buf_offset, 0x151);
ASSERT_REG_POSITION(vp9_entropy_probs_offset, 0x170);
ASSERT_REG_POSITION(vp9_backward_updates_offset, 0x171);
ASSERT_REG_POSITION(vp9_last_frame_segmap_offset, 0x172);
ASSERT_REG_POSITION(vp9_curr_frame_segmap_offset, 0x173);
ASSERT_REG_POSITION(vp9_last_frame_mvs_offset, 0x175);
ASSERT_REG_POSITION(vp9_curr_frame_mvs_offset, 0x176);
#undef ASSERT_REG_POSITION
} // namespace Tegra::Host1x::NvdecCommon

100
src/video_core/host1x/sync_manager.cpp
View File

@@ -1,50 +1,50 @@
// SPDX-FileCopyrightText: Ryujinx Team and Contributors
// SPDX-License-Identifier: MIT
#include <algorithm>
#include "sync_manager.h"
#include "video_core/host1x/host1x.h"
#include "video_core/host1x/syncpoint_manager.h"
namespace Tegra {
namespace Host1x {
SyncptIncrManager::SyncptIncrManager(Host1x& host1x_) : host1x(host1x_) {}
SyncptIncrManager::~SyncptIncrManager() = default;
void SyncptIncrManager::Increment(u32 id) {
increments.emplace_back(0, 0, id, true);
IncrementAllDone();
}
u32 SyncptIncrManager::IncrementWhenDone(u32 class_id, u32 id) {
const u32 handle = current_id++;
increments.emplace_back(handle, class_id, id);
return handle;
}
void SyncptIncrManager::SignalDone(u32 handle) {
const auto done_incr =
std::find_if(increments.begin(), increments.end(),
[handle](const SyncptIncr& incr) { return incr.id == handle; });
if (done_incr != increments.cend()) {
done_incr->complete = true;
}
IncrementAllDone();
}
void SyncptIncrManager::IncrementAllDone() {
std::size_t done_count = 0;
for (; done_count < increments.size(); ++done_count) {
if (!increments[done_count].complete) {
break;
}
auto& syncpoint_manager = host1x.GetSyncpointManager();
syncpoint_manager.IncrementGuest(increments[done_count].syncpt_id);
syncpoint_manager.IncrementHost(increments[done_count].syncpt_id);
}
increments.erase(increments.begin(), increments.begin() + done_count);
}
} // namespace Host1x
} // namespace Tegra
// SPDX-FileCopyrightText: Ryujinx Team and Contributors
// SPDX-License-Identifier: MIT
#include <algorithm>
#include "sync_manager.h"
#include "video_core/host1x/host1x.h"
#include "video_core/host1x/syncpoint_manager.h"
namespace Tegra {
namespace Host1x {
SyncptIncrManager::SyncptIncrManager(Host1x& host1x_) : host1x(host1x_) {}
SyncptIncrManager::~SyncptIncrManager() = default;
void SyncptIncrManager::Increment(u32 id) {
increments.emplace_back(0, 0, id, true);
IncrementAllDone();
}
u32 SyncptIncrManager::IncrementWhenDone(u32 class_id, u32 id) {
const u32 handle = current_id++;
increments.emplace_back(handle, class_id, id);
return handle;
}
void SyncptIncrManager::SignalDone(u32 handle) {
const auto done_incr =
std::find_if(increments.begin(), increments.end(),
[handle](const SyncptIncr& incr) { return incr.id == handle; });
if (done_incr != increments.cend()) {
done_incr->complete = true;
}
IncrementAllDone();
}
void SyncptIncrManager::IncrementAllDone() {
std::size_t done_count = 0;
for (; done_count < increments.size(); ++done_count) {
if (!increments[done_count].complete) {
break;
}
auto& syncpoint_manager = host1x.GetSyncpointManager();
syncpoint_manager.IncrementGuest(increments[done_count].syncpt_id);
syncpoint_manager.IncrementHost(increments[done_count].syncpt_id);
}
increments.erase(increments.begin(), increments.begin() + done_count);
}
} // namespace Host1x
} // namespace Tegra

106
src/video_core/host1x/sync_manager.h
View File

@@ -1,53 +1,53 @@
// SPDX-FileCopyrightText: Ryujinx Team and Contributors
// SPDX-License-Identifier: MIT
#pragma once
#include <mutex>
#include <vector>
#include "common/common_types.h"
namespace Tegra {
namespace Host1x {
class Host1x;
struct SyncptIncr {
u32 id;
u32 class_id;
u32 syncpt_id;
bool complete;
SyncptIncr(u32 id_, u32 class_id_, u32 syncpt_id_, bool done = false)
: id(id_), class_id(class_id_), syncpt_id(syncpt_id_), complete(done) {}
};
class SyncptIncrManager {
public:
explicit SyncptIncrManager(Host1x& host1x);
~SyncptIncrManager();
/// Add syncpoint id and increment all
void Increment(u32 id);
/// Returns a handle to increment later
u32 IncrementWhenDone(u32 class_id, u32 id);
/// IncrememntAllDone, including handle
void SignalDone(u32 handle);
/// Increment all sequential pending increments that are already done.
void IncrementAllDone();
private:
std::vector<SyncptIncr> increments;
std::mutex increment_lock;
u32 current_id{};
Host1x& host1x;
};
} // namespace Host1x
} // namespace Tegra
// SPDX-FileCopyrightText: Ryujinx Team and Contributors
// SPDX-License-Identifier: MIT
#pragma once
#include <mutex>
#include <vector>
#include "common/common_types.h"
namespace Tegra {
namespace Host1x {
class Host1x;
struct SyncptIncr {
u32 id;
u32 class_id;
u32 syncpt_id;
bool complete;
SyncptIncr(u32 id_, u32 class_id_, u32 syncpt_id_, bool done = false)
: id(id_), class_id(class_id_), syncpt_id(syncpt_id_), complete(done) {}
};
class SyncptIncrManager {
public:
explicit SyncptIncrManager(Host1x& host1x);
~SyncptIncrManager();
/// Add syncpoint id and increment all
void Increment(u32 id);
/// Returns a handle to increment later
u32 IncrementWhenDone(u32 class_id, u32 id);
/// IncrememntAllDone, including handle
void SignalDone(u32 handle);
/// Increment all sequential pending increments that are already done.
void IncrementAllDone();
private:
std::vector<SyncptIncr> increments;
std::mutex increment_lock;
u32 current_id{};
Host1x& host1x;
};
} // namespace Host1x
} // namespace Tegra

212
src/video_core/host1x/syncpoint_manager.cpp
View File

@@ -1,106 +1,106 @@
// SPDX-FileCopyrightText: 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#include "common/microprofile.h"
#include "video_core/host1x/syncpoint_manager.h"
namespace Tegra {
namespace Host1x {
MICROPROFILE_DEFINE(GPU_wait, "GPU", "Wait for the GPU", MP_RGB(128, 128, 192));
SyncpointManager::ActionHandle SyncpointManager::RegisterAction(
std::atomic<u32>& syncpoint, std::list<RegisteredAction>& action_storage, u32 expected_value,
std::function<void()>&& action) {
if (syncpoint.load(std::memory_order_acquire) >= expected_value) {
action();
return {};
}
std::unique_lock lk(guard);
if (syncpoint.load(std::memory_order_relaxed) >= expected_value) {
action();
return {};
}
auto it = action_storage.begin();
while (it != action_storage.end()) {
if (it->expected_value >= expected_value) {
break;
}
++it;
}
return action_storage.emplace(it, expected_value, std::move(action));
}
void SyncpointManager::DeregisterAction(std::list<RegisteredAction>& action_storage,
ActionHandle& handle) {
std::unique_lock lk(guard);
// We want to ensure the iterator still exists prior to erasing it
// Otherwise, if an invalid iterator was passed in then it could lead to UB
// It is important to avoid UB in that case since the deregister isn't called from a locked
// context
for (auto it = action_storage.begin(); it != action_storage.end(); it++) {
if (it == handle) {
action_storage.erase(it);
return;
}
}
}
void SyncpointManager::DeregisterGuestAction(u32 syncpoint_id, ActionHandle& handle) {
DeregisterAction(guest_action_storage[syncpoint_id], handle);
}
void SyncpointManager::DeregisterHostAction(u32 syncpoint_id, ActionHandle& handle) {
DeregisterAction(host_action_storage[syncpoint_id], handle);
}
void SyncpointManager::IncrementGuest(u32 syncpoint_id) {
Increment(syncpoints_guest[syncpoint_id], wait_guest_cv, guest_action_storage[syncpoint_id]);
}
void SyncpointManager::IncrementHost(u32 syncpoint_id) {
Increment(syncpoints_host[syncpoint_id], wait_host_cv, host_action_storage[syncpoint_id]);
}
void SyncpointManager::WaitGuest(u32 syncpoint_id, u32 expected_value) {
Wait(syncpoints_guest[syncpoint_id], wait_guest_cv, expected_value);
}
void SyncpointManager::WaitHost(u32 syncpoint_id, u32 expected_value) {
MICROPROFILE_SCOPE(GPU_wait);
Wait(syncpoints_host[syncpoint_id], wait_host_cv, expected_value);
}
void SyncpointManager::Increment(std::atomic<u32>& syncpoint, std::condition_variable& wait_cv,
std::list<RegisteredAction>& action_storage) {
auto new_value{syncpoint.fetch_add(1, std::memory_order_acq_rel) + 1};
std::unique_lock lk(guard);
auto it = action_storage.begin();
while (it != action_storage.end()) {
if (it->expected_value > new_value) {
break;
}
it->action();
it = action_storage.erase(it);
}
wait_cv.notify_all();
}
void SyncpointManager::Wait(std::atomic<u32>& syncpoint, std::condition_variable& wait_cv,
u32 expected_value) {
const auto pred = [&]() { return syncpoint.load(std::memory_order_acquire) >= expected_value; };
if (pred()) {
return;
}
std::unique_lock lk(guard);
wait_cv.wait(lk, pred);
}
} // namespace Host1x
} // namespace Tegra
// SPDX-FileCopyrightText: 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#include "common/microprofile.h"
#include "video_core/host1x/syncpoint_manager.h"
namespace Tegra {
namespace Host1x {
MICROPROFILE_DEFINE(GPU_wait, "GPU", "Wait for the GPU", MP_RGB(128, 128, 192));
SyncpointManager::ActionHandle SyncpointManager::RegisterAction(
std::atomic<u32>& syncpoint, std::list<RegisteredAction>& action_storage, u32 expected_value,
std::function<void()>&& action) {
if (syncpoint.load(std::memory_order_acquire) >= expected_value) {
action();
return {};
}
std::unique_lock lk(guard);
if (syncpoint.load(std::memory_order_relaxed) >= expected_value) {
action();
return {};
}
auto it = action_storage.begin();
while (it != action_storage.end()) {
if (it->expected_value >= expected_value) {
break;
}
++it;
}
return action_storage.emplace(it, expected_value, std::move(action));
}
void SyncpointManager::DeregisterAction(std::list<RegisteredAction>& action_storage,
ActionHandle& handle) {
std::unique_lock lk(guard);
// We want to ensure the iterator still exists prior to erasing it
// Otherwise, if an invalid iterator was passed in then it could lead to UB
// It is important to avoid UB in that case since the deregister isn't called from a locked
// context
for (auto it = action_storage.begin(); it != action_storage.end(); it++) {
if (it == handle) {
action_storage.erase(it);
return;
}
}
}
void SyncpointManager::DeregisterGuestAction(u32 syncpoint_id, ActionHandle& handle) {
DeregisterAction(guest_action_storage[syncpoint_id], handle);
}
void SyncpointManager::DeregisterHostAction(u32 syncpoint_id, ActionHandle& handle) {
DeregisterAction(host_action_storage[syncpoint_id], handle);
}
void SyncpointManager::IncrementGuest(u32 syncpoint_id) {
Increment(syncpoints_guest[syncpoint_id], wait_guest_cv, guest_action_storage[syncpoint_id]);
}
void SyncpointManager::IncrementHost(u32 syncpoint_id) {
Increment(syncpoints_host[syncpoint_id], wait_host_cv, host_action_storage[syncpoint_id]);
}
void SyncpointManager::WaitGuest(u32 syncpoint_id, u32 expected_value) {
Wait(syncpoints_guest[syncpoint_id], wait_guest_cv, expected_value);
}
void SyncpointManager::WaitHost(u32 syncpoint_id, u32 expected_value) {
MICROPROFILE_SCOPE(GPU_wait);
Wait(syncpoints_host[syncpoint_id], wait_host_cv, expected_value);
}
void SyncpointManager::Increment(std::atomic<u32>& syncpoint, std::condition_variable& wait_cv,
std::list<RegisteredAction>& action_storage) {
auto new_value{syncpoint.fetch_add(1, std::memory_order_acq_rel) + 1};
std::unique_lock lk(guard);
auto it = action_storage.begin();
while (it != action_storage.end()) {
if (it->expected_value > new_value) {
break;
}
it->action();
it = action_storage.erase(it);
}
wait_cv.notify_all();
}
void SyncpointManager::Wait(std::atomic<u32>& syncpoint, std::condition_variable& wait_cv,
u32 expected_value) {
const auto pred = [&]() { return syncpoint.load(std::memory_order_acquire) >= expected_value; };
if (pred()) {
return;
}
std::unique_lock lk(guard);
wait_cv.wait(lk, pred);
}
} // namespace Host1x
} // namespace Tegra

196
src/video_core/host1x/syncpoint_manager.h
View File

@@ -1,98 +1,98 @@
// SPDX-FileCopyrightText: 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#pragma once
#include <array>
#include <atomic>
#include <condition_variable>
#include <functional>
#include <list>
#include <mutex>
#include "common/common_types.h"
namespace Tegra {
namespace Host1x {
class SyncpointManager {
public:
u32 GetGuestSyncpointValue(u32 id) const {
return syncpoints_guest[id].load(std::memory_order_acquire);
}
u32 GetHostSyncpointValue(u32 id) const {
return syncpoints_host[id].load(std::memory_order_acquire);
}
struct RegisteredAction {
explicit RegisteredAction(u32 expected_value_, std::function<void()>&& action_)
: expected_value{expected_value_}, action{std::move(action_)} {}
u32 expected_value;
std::function<void()> action;
};
using ActionHandle = std::list<RegisteredAction>::iterator;
template <typename Func>
ActionHandle RegisterGuestAction(u32 syncpoint_id, u32 expected_value, Func&& action) {
std::function<void()> func(action);
return RegisterAction(syncpoints_guest[syncpoint_id], guest_action_storage[syncpoint_id],
expected_value, std::move(func));
}
template <typename Func>
ActionHandle RegisterHostAction(u32 syncpoint_id, u32 expected_value, Func&& action) {
std::function<void()> func(action);
return RegisterAction(syncpoints_host[syncpoint_id], host_action_storage[syncpoint_id],
expected_value, std::move(func));
}
void DeregisterGuestAction(u32 syncpoint_id, ActionHandle& handle);
void DeregisterHostAction(u32 syncpoint_id, ActionHandle& handle);
void IncrementGuest(u32 syncpoint_id);
void IncrementHost(u32 syncpoint_id);
void WaitGuest(u32 syncpoint_id, u32 expected_value);
void WaitHost(u32 syncpoint_id, u32 expected_value);
bool IsReadyGuest(u32 syncpoint_id, u32 expected_value) const {
return syncpoints_guest[syncpoint_id].load(std::memory_order_acquire) >= expected_value;
}
bool IsReadyHost(u32 syncpoint_id, u32 expected_value) const {
return syncpoints_host[syncpoint_id].load(std::memory_order_acquire) >= expected_value;
}
private:
void Increment(std::atomic<u32>& syncpoint, std::condition_variable& wait_cv,
std::list<RegisteredAction>& action_storage);
ActionHandle RegisterAction(std::atomic<u32>& syncpoint,
std::list<RegisteredAction>& action_storage, u32 expected_value,
std::function<void()>&& action);
void DeregisterAction(std::list<RegisteredAction>& action_storage, ActionHandle& handle);
void Wait(std::atomic<u32>& syncpoint, std::condition_variable& wait_cv, u32 expected_value);
static constexpr size_t NUM_MAX_SYNCPOINTS = 192;
std::array<std::atomic<u32>, NUM_MAX_SYNCPOINTS> syncpoints_guest{};
std::array<std::atomic<u32>, NUM_MAX_SYNCPOINTS> syncpoints_host{};
std::array<std::list<RegisteredAction>, NUM_MAX_SYNCPOINTS> guest_action_storage;
std::array<std::list<RegisteredAction>, NUM_MAX_SYNCPOINTS> host_action_storage;
std::mutex guard;
std::condition_variable wait_guest_cv;
std::condition_variable wait_host_cv;
};
} // namespace Host1x
} // namespace Tegra
// SPDX-FileCopyrightText: 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#pragma once
#include <array>
#include <atomic>
#include <condition_variable>
#include <functional>
#include <list>
#include <mutex>
#include "common/common_types.h"
namespace Tegra {
namespace Host1x {
class SyncpointManager {
public:
u32 GetGuestSyncpointValue(u32 id) const {
return syncpoints_guest[id].load(std::memory_order_acquire);
}
u32 GetHostSyncpointValue(u32 id) const {
return syncpoints_host[id].load(std::memory_order_acquire);
}
struct RegisteredAction {
explicit RegisteredAction(u32 expected_value_, std::function<void()>&& action_)
: expected_value{expected_value_}, action{std::move(action_)} {}
u32 expected_value;
std::function<void()> action;
};
using ActionHandle = std::list<RegisteredAction>::iterator;
template <typename Func>
ActionHandle RegisterGuestAction(u32 syncpoint_id, u32 expected_value, Func&& action) {
std::function<void()> func(action);
return RegisterAction(syncpoints_guest[syncpoint_id], guest_action_storage[syncpoint_id],
expected_value, std::move(func));
}
template <typename Func>
ActionHandle RegisterHostAction(u32 syncpoint_id, u32 expected_value, Func&& action) {
std::function<void()> func(action);
return RegisterAction(syncpoints_host[syncpoint_id], host_action_storage[syncpoint_id],
expected_value, std::move(func));
}
void DeregisterGuestAction(u32 syncpoint_id, ActionHandle& handle);
void DeregisterHostAction(u32 syncpoint_id, ActionHandle& handle);
void IncrementGuest(u32 syncpoint_id);
void IncrementHost(u32 syncpoint_id);
void WaitGuest(u32 syncpoint_id, u32 expected_value);
void WaitHost(u32 syncpoint_id, u32 expected_value);
bool IsReadyGuest(u32 syncpoint_id, u32 expected_value) const {
return syncpoints_guest[syncpoint_id].load(std::memory_order_acquire) >= expected_value;
}
bool IsReadyHost(u32 syncpoint_id, u32 expected_value) const {
return syncpoints_host[syncpoint_id].load(std::memory_order_acquire) >= expected_value;
}
private:
void Increment(std::atomic<u32>& syncpoint, std::condition_variable& wait_cv,
std::list<RegisteredAction>& action_storage);
ActionHandle RegisterAction(std::atomic<u32>& syncpoint,
std::list<RegisteredAction>& action_storage, u32 expected_value,
std::function<void()>&& action);
void DeregisterAction(std::list<RegisteredAction>& action_storage, ActionHandle& handle);
void Wait(std::atomic<u32>& syncpoint, std::condition_variable& wait_cv, u32 expected_value);
static constexpr size_t NUM_MAX_SYNCPOINTS = 192;
std::array<std::atomic<u32>, NUM_MAX_SYNCPOINTS> syncpoints_guest{};
std::array<std::atomic<u32>, NUM_MAX_SYNCPOINTS> syncpoints_host{};
std::array<std::list<RegisteredAction>, NUM_MAX_SYNCPOINTS> guest_action_storage;
std::array<std::list<RegisteredAction>, NUM_MAX_SYNCPOINTS> host_action_storage;
std::mutex guard;
std::condition_variable wait_guest_cv;
std::condition_variable wait_host_cv;
};
} // namespace Host1x
} // namespace Tegra

488
src/video_core/host1x/vic.cpp
View File

@@ -1,244 +1,244 @@
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <array>
extern "C" {
#if defined(__GNUC__) || defined(__clang__)
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wconversion"
#endif
#include <libswscale/swscale.h>
#if defined(__GNUC__) || defined(__clang__)
#pragma GCC diagnostic pop
#endif
}
#include "common/assert.h"
#include "common/bit_field.h"
#include "common/logging/log.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/host1x/host1x.h"
#include "video_core/host1x/nvdec.h"
#include "video_core/host1x/vic.h"
#include "video_core/memory_manager.h"
#include "video_core/textures/decoders.h"
namespace Tegra {
namespace Host1x {
namespace {
enum class VideoPixelFormat : u64_le {
RGBA8 = 0x1f,
BGRA8 = 0x20,
RGBX8 = 0x23,
YUV420 = 0x44,
};
} // Anonymous namespace
union VicConfig {
u64_le raw{};
BitField<0, 7, VideoPixelFormat> pixel_format;
BitField<7, 2, u64_le> chroma_loc_horiz;
BitField<9, 2, u64_le> chroma_loc_vert;
BitField<11, 4, u64_le> block_linear_kind;
BitField<15, 4, u64_le> block_linear_height_log2;
BitField<32, 14, u64_le> surface_width_minus1;
BitField<46, 14, u64_le> surface_height_minus1;
};
Vic::Vic(Host1x& host1x_, std::shared_ptr<Nvdec> nvdec_processor_)
: host1x(host1x_),
nvdec_processor(std::move(nvdec_processor_)), converted_frame_buffer{nullptr, av_free} {}
Vic::~Vic() = default;
void Vic::ProcessMethod(Method method, u32 argument) {
LOG_DEBUG(HW_GPU, "Vic method 0x{:X}", static_cast<u32>(method));
const u64 arg = static_cast<u64>(argument) << 8;
switch (method) {
case Method::Execute:
Execute();
break;
case Method::SetConfigStructOffset:
config_struct_address = arg;
break;
case Method::SetOutputSurfaceLumaOffset:
output_surface_luma_address = arg;
break;
case Method::SetOutputSurfaceChromaOffset:
output_surface_chroma_address = arg;
break;
default:
break;
}
}
void Vic::Execute() {
if (output_surface_luma_address == 0) {
LOG_ERROR(Service_NVDRV, "VIC Luma address not set.");
return;
}
const VicConfig config{host1x.MemoryManager().Read<u64>(config_struct_address + 0x20)};
const AVFramePtr frame_ptr = nvdec_processor->GetFrame();
const auto* frame = frame_ptr.get();
if (!frame) {
return;
}
const u64 surface_width = config.surface_width_minus1 + 1;
const u64 surface_height = config.surface_height_minus1 + 1;
if (static_cast<u64>(frame->width) != surface_width ||
static_cast<u64>(frame->height) != surface_height) {
// TODO: Properly support multiple video streams with differing frame dimensions
LOG_WARNING(Service_NVDRV, "Frame dimensions {}x{} don't match surface dimensions {}x{}",
frame->width, frame->height, surface_width, surface_height);
}
switch (config.pixel_format) {
case VideoPixelFormat::RGBA8:
case VideoPixelFormat::BGRA8:
case VideoPixelFormat::RGBX8:
WriteRGBFrame(frame, config);
break;
case VideoPixelFormat::YUV420:
WriteYUVFrame(frame, config);
break;
default:
UNIMPLEMENTED_MSG("Unknown video pixel format {:X}", config.pixel_format.Value());
break;
}
}
void Vic::WriteRGBFrame(const AVFrame* frame, const VicConfig& config) {
LOG_TRACE(Service_NVDRV, "Writing RGB Frame");
if (!scaler_ctx || frame->width != scaler_width || frame->height != scaler_height) {
const AVPixelFormat target_format = [pixel_format = config.pixel_format]() {
switch (pixel_format) {
case VideoPixelFormat::RGBA8:
return AV_PIX_FMT_RGBA;
case VideoPixelFormat::BGRA8:
return AV_PIX_FMT_BGRA;
case VideoPixelFormat::RGBX8:
return AV_PIX_FMT_RGB0;
default:
return AV_PIX_FMT_RGBA;
}
}();
sws_freeContext(scaler_ctx);
// Frames are decoded into either YUV420 or NV12 formats. Convert to desired RGB format
scaler_ctx = sws_getContext(frame->width, frame->height,
static_cast<AVPixelFormat>(frame->format), frame->width,
frame->height, target_format, 0, nullptr, nullptr, nullptr);
scaler_width = frame->width;
scaler_height = frame->height;
converted_frame_buffer.reset();
}
if (!converted_frame_buffer) {
const size_t frame_size = frame->width * frame->height * 4;
converted_frame_buffer = AVMallocPtr{static_cast<u8*>(av_malloc(frame_size)), av_free};
}
const std::array<int, 4> converted_stride{frame->width * 4, frame->height * 4, 0, 0};
u8* const converted_frame_buf_addr{converted_frame_buffer.get()};
sws_scale(scaler_ctx, frame->data, frame->linesize, 0, frame->height, &converted_frame_buf_addr,
converted_stride.data());
// Use the minimum of surface/frame dimensions to avoid buffer overflow.
const u32 surface_width = static_cast<u32>(config.surface_width_minus1) + 1;
const u32 surface_height = static_cast<u32>(config.surface_height_minus1) + 1;
const u32 width = std::min(surface_width, static_cast<u32>(frame->width));
const u32 height = std::min(surface_height, static_cast<u32>(frame->height));
const u32 blk_kind = static_cast<u32>(config.block_linear_kind);
if (blk_kind != 0) {
// swizzle pitch linear to block linear
const u32 block_height = static_cast<u32>(config.block_linear_height_log2);
const auto size = Texture::CalculateSize(true, 4, width, height, 1, block_height, 0);
luma_buffer.resize(size);
std::span<const u8> frame_buff(converted_frame_buf_addr, 4 * width * height);
Texture::SwizzleSubrect(luma_buffer, frame_buff, 4, width, height, 1, 0, 0, width, height,
block_height, 0, width * 4);
host1x.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(), size);
} else {
// send pitch linear frame
const size_t linear_size = width * height * 4;
host1x.MemoryManager().WriteBlock(output_surface_luma_address, converted_frame_buf_addr,
linear_size);
}
}
void Vic::WriteYUVFrame(const AVFrame* frame, const VicConfig& config) {
LOG_TRACE(Service_NVDRV, "Writing YUV420 Frame");
const std::size_t surface_width = config.surface_width_minus1 + 1;
const std::size_t surface_height = config.surface_height_minus1 + 1;
const std::size_t aligned_width = (surface_width + 0xff) & ~0xffUL;
// Use the minimum of surface/frame dimensions to avoid buffer overflow.
const auto frame_width = std::min(surface_width, static_cast<size_t>(frame->width));
const auto frame_height = std::min(surface_height, static_cast<size_t>(frame->height));
const auto stride = static_cast<size_t>(frame->linesize[0]);
luma_buffer.resize(aligned_width * surface_height);
chroma_buffer.resize(aligned_width * surface_height / 2);
// Populate luma buffer
const u8* luma_src = frame->data[0];
for (std::size_t y = 0; y < frame_height; ++y) {
const std::size_t src = y * stride;
const std::size_t dst = y * aligned_width;
for (std::size_t x = 0; x < frame_width; ++x) {
luma_buffer[dst + x] = luma_src[src + x];
}
}
host1x.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(),
luma_buffer.size());
// Chroma
const std::size_t half_height = frame_height / 2;
const auto half_stride = static_cast<size_t>(frame->linesize[1]);
switch (frame->format) {
case AV_PIX_FMT_YUV420P: {
// Frame from FFmpeg software
// Populate chroma buffer from both channels with interleaving.
const std::size_t half_width = frame_width / 2;
const u8* chroma_b_src = frame->data[1];
const u8* chroma_r_src = frame->data[2];
for (std::size_t y = 0; y < half_height; ++y) {
const std::size_t src = y * half_stride;
const std::size_t dst = y * aligned_width;
for (std::size_t x = 0; x < half_width; ++x) {
chroma_buffer[dst + x * 2] = chroma_b_src[src + x];
chroma_buffer[dst + x * 2 + 1] = chroma_r_src[src + x];
}
}
break;
}
case AV_PIX_FMT_NV12: {
// Frame from VA-API hardware
// This is already interleaved so just copy
const u8* chroma_src = frame->data[1];
for (std::size_t y = 0; y < half_height; ++y) {
const std::size_t src = y * stride;
const std::size_t dst = y * aligned_width;
for (std::size_t x = 0; x < frame_width; ++x) {
chroma_buffer[dst + x] = chroma_src[src + x];
}
}
break;
}
default:
ASSERT(false);
break;
}
host1x.MemoryManager().WriteBlock(output_surface_chroma_address, chroma_buffer.data(),
chroma_buffer.size());
}
} // namespace Host1x
} // namespace Tegra
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <array>
extern "C" {
#if defined(__GNUC__) || defined(__clang__)
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wconversion"
#endif
#include <libswscale/swscale.h>
#if defined(__GNUC__) || defined(__clang__)
#pragma GCC diagnostic pop
#endif
}
#include "common/assert.h"
#include "common/bit_field.h"
#include "common/logging/log.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/host1x/host1x.h"
#include "video_core/host1x/nvdec.h"
#include "video_core/host1x/vic.h"
#include "video_core/memory_manager.h"
#include "video_core/textures/decoders.h"
namespace Tegra {
namespace Host1x {
namespace {
enum class VideoPixelFormat : u64_le {
RGBA8 = 0x1f,
BGRA8 = 0x20,
RGBX8 = 0x23,
YUV420 = 0x44,
};
} // Anonymous namespace
union VicConfig {
u64_le raw{};
BitField<0, 7, VideoPixelFormat> pixel_format;
BitField<7, 2, u64_le> chroma_loc_horiz;
BitField<9, 2, u64_le> chroma_loc_vert;
BitField<11, 4, u64_le> block_linear_kind;
BitField<15, 4, u64_le> block_linear_height_log2;
BitField<32, 14, u64_le> surface_width_minus1;
BitField<46, 14, u64_le> surface_height_minus1;
};
Vic::Vic(Host1x& host1x_, std::shared_ptr<Nvdec> nvdec_processor_)
: host1x(host1x_),
nvdec_processor(std::move(nvdec_processor_)), converted_frame_buffer{nullptr, av_free} {}
Vic::~Vic() = default;
void Vic::ProcessMethod(Method method, u32 argument) {
LOG_DEBUG(HW_GPU, "Vic method 0x{:X}", static_cast<u32>(method));
const u64 arg = static_cast<u64>(argument) << 8;
switch (method) {
case Method::Execute:
Execute();
break;
case Method::SetConfigStructOffset:
config_struct_address = arg;
break;
case Method::SetOutputSurfaceLumaOffset:
output_surface_luma_address = arg;
break;
case Method::SetOutputSurfaceChromaOffset:
output_surface_chroma_address = arg;
break;
default:
break;
}
}
void Vic::Execute() {
if (output_surface_luma_address == 0) {
LOG_ERROR(Service_NVDRV, "VIC Luma address not set.");
return;
}
const VicConfig config{host1x.MemoryManager().Read<u64>(config_struct_address + 0x20)};
const AVFramePtr frame_ptr = nvdec_processor->GetFrame();
const auto* frame = frame_ptr.get();
if (!frame) {
return;
}
const u64 surface_width = config.surface_width_minus1 + 1;
const u64 surface_height = config.surface_height_minus1 + 1;
if (static_cast<u64>(frame->width) != surface_width ||
static_cast<u64>(frame->height) != surface_height) {
// TODO: Properly support multiple video streams with differing frame dimensions
LOG_WARNING(Service_NVDRV, "Frame dimensions {}x{} don't match surface dimensions {}x{}",
frame->width, frame->height, surface_width, surface_height);
}
switch (config.pixel_format) {
case VideoPixelFormat::RGBA8:
case VideoPixelFormat::BGRA8:
case VideoPixelFormat::RGBX8:
WriteRGBFrame(frame, config);
break;
case VideoPixelFormat::YUV420:
WriteYUVFrame(frame, config);
break;
default:
UNIMPLEMENTED_MSG("Unknown video pixel format {:X}", config.pixel_format.Value());
break;
}
}
void Vic::WriteRGBFrame(const AVFrame* frame, const VicConfig& config) {
LOG_TRACE(Service_NVDRV, "Writing RGB Frame");
if (!scaler_ctx || frame->width != scaler_width || frame->height != scaler_height) {
const AVPixelFormat target_format = [pixel_format = config.pixel_format]() {
switch (pixel_format) {
case VideoPixelFormat::RGBA8:
return AV_PIX_FMT_RGBA;
case VideoPixelFormat::BGRA8:
return AV_PIX_FMT_BGRA;
case VideoPixelFormat::RGBX8:
return AV_PIX_FMT_RGB0;
default:
return AV_PIX_FMT_RGBA;
}
}();
sws_freeContext(scaler_ctx);
// Frames are decoded into either YUV420 or NV12 formats. Convert to desired RGB format
scaler_ctx = sws_getContext(frame->width, frame->height,
static_cast<AVPixelFormat>(frame->format), frame->width,
frame->height, target_format, 0, nullptr, nullptr, nullptr);
scaler_width = frame->width;
scaler_height = frame->height;
converted_frame_buffer.reset();
}
if (!converted_frame_buffer) {
const size_t frame_size = frame->width * frame->height * 4;
converted_frame_buffer = AVMallocPtr{static_cast<u8*>(av_malloc(frame_size)), av_free};
}
const std::array<int, 4> converted_stride{frame->width * 4, frame->height * 4, 0, 0};
u8* const converted_frame_buf_addr{converted_frame_buffer.get()};
sws_scale(scaler_ctx, frame->data, frame->linesize, 0, frame->height, &converted_frame_buf_addr,
converted_stride.data());
// Use the minimum of surface/frame dimensions to avoid buffer overflow.
const u32 surface_width = static_cast<u32>(config.surface_width_minus1) + 1;
const u32 surface_height = static_cast<u32>(config.surface_height_minus1) + 1;
const u32 width = std::min(surface_width, static_cast<u32>(frame->width));
const u32 height = std::min(surface_height, static_cast<u32>(frame->height));
const u32 blk_kind = static_cast<u32>(config.block_linear_kind);
if (blk_kind != 0) {
// swizzle pitch linear to block linear
const u32 block_height = static_cast<u32>(config.block_linear_height_log2);
const auto size = Texture::CalculateSize(true, 4, width, height, 1, block_height, 0);
luma_buffer.resize(size);
std::span<const u8> frame_buff(converted_frame_buf_addr, 4 * width * height);
Texture::SwizzleSubrect(luma_buffer, frame_buff, 4, width, height, 1, 0, 0, width, height,
block_height, 0, width * 4);
host1x.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(), size);
} else {
// send pitch linear frame
const size_t linear_size = width * height * 4;
host1x.MemoryManager().WriteBlock(output_surface_luma_address, converted_frame_buf_addr,
linear_size);
}
}
void Vic::WriteYUVFrame(const AVFrame* frame, const VicConfig& config) {
LOG_TRACE(Service_NVDRV, "Writing YUV420 Frame");
const std::size_t surface_width = config.surface_width_minus1 + 1;
const std::size_t surface_height = config.surface_height_minus1 + 1;
const std::size_t aligned_width = (surface_width + 0xff) & ~0xffUL;
// Use the minimum of surface/frame dimensions to avoid buffer overflow.
const auto frame_width = std::min(surface_width, static_cast<size_t>(frame->width));
const auto frame_height = std::min(surface_height, static_cast<size_t>(frame->height));
const auto stride = static_cast<size_t>(frame->linesize[0]);
luma_buffer.resize(aligned_width * surface_height);
chroma_buffer.resize(aligned_width * surface_height / 2);
// Populate luma buffer
const u8* luma_src = frame->data[0];
for (std::size_t y = 0; y < frame_height; ++y) {
const std::size_t src = y * stride;
const std::size_t dst = y * aligned_width;
for (std::size_t x = 0; x < frame_width; ++x) {
luma_buffer[dst + x] = luma_src[src + x];
}
}
host1x.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(),
luma_buffer.size());
// Chroma
const std::size_t half_height = frame_height / 2;
const auto half_stride = static_cast<size_t>(frame->linesize[1]);
switch (frame->format) {
case AV_PIX_FMT_YUV420P: {
// Frame from FFmpeg software
// Populate chroma buffer from both channels with interleaving.
const std::size_t half_width = frame_width / 2;
const u8* chroma_b_src = frame->data[1];
const u8* chroma_r_src = frame->data[2];
for (std::size_t y = 0; y < half_height; ++y) {
const std::size_t src = y * half_stride;
const std::size_t dst = y * aligned_width;
for (std::size_t x = 0; x < half_width; ++x) {
chroma_buffer[dst + x * 2] = chroma_b_src[src + x];
chroma_buffer[dst + x * 2 + 1] = chroma_r_src[src + x];
}
}
break;
}
case AV_PIX_FMT_NV12: {
// Frame from VA-API hardware
// This is already interleaved so just copy
const u8* chroma_src = frame->data[1];
for (std::size_t y = 0; y < half_height; ++y) {
const std::size_t src = y * stride;
const std::size_t dst = y * aligned_width;
for (std::size_t x = 0; x < frame_width; ++x) {
chroma_buffer[dst + x] = chroma_src[src + x];
}
}
break;
}
default:
ASSERT(false);
break;
}
host1x.MemoryManager().WriteBlock(output_surface_chroma_address, chroma_buffer.data(),
chroma_buffer.size());
}
} // namespace Host1x
} // namespace Tegra

132
src/video_core/host1x/vic.h
View File

@@ -1,66 +1,66 @@
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <memory>
#include <vector>
#include "common/common_types.h"
struct SwsContext;
namespace Tegra {
namespace Host1x {
class Host1x;
class Nvdec;
union VicConfig;
class Vic {
public:
enum class Method : u32 {
Execute = 0xc0,
SetControlParams = 0x1c1,
SetConfigStructOffset = 0x1c2,
SetOutputSurfaceLumaOffset = 0x1c8,
SetOutputSurfaceChromaOffset = 0x1c9,
SetOutputSurfaceChromaUnusedOffset = 0x1ca
};
explicit Vic(Host1x& host1x, std::shared_ptr<Nvdec> nvdec_processor);
~Vic();
/// Write to the device state.
void ProcessMethod(Method method, u32 argument);
private:
void Execute();
void WriteRGBFrame(const AVFrame* frame, const VicConfig& config);
void WriteYUVFrame(const AVFrame* frame, const VicConfig& config);
Host1x& host1x;
std::shared_ptr<Tegra::Host1x::Nvdec> nvdec_processor;
/// Avoid reallocation of the following buffers every frame, as their
/// size does not change during a stream
using AVMallocPtr = std::unique_ptr<u8, decltype(&av_free)>;
AVMallocPtr converted_frame_buffer;
std::vector<u8> luma_buffer;
std::vector<u8> chroma_buffer;
GPUVAddr config_struct_address{};
GPUVAddr output_surface_luma_address{};
GPUVAddr output_surface_chroma_address{};
SwsContext* scaler_ctx{};
s32 scaler_width{};
s32 scaler_height{};
};
} // namespace Host1x
} // namespace Tegra
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <memory>
#include <vector>
#include "common/common_types.h"
struct SwsContext;
namespace Tegra {
namespace Host1x {
class Host1x;
class Nvdec;
union VicConfig;
class Vic {
public:
enum class Method : u32 {
Execute = 0xc0,
SetControlParams = 0x1c1,
SetConfigStructOffset = 0x1c2,
SetOutputSurfaceLumaOffset = 0x1c8,
SetOutputSurfaceChromaOffset = 0x1c9,
SetOutputSurfaceChromaUnusedOffset = 0x1ca
};
explicit Vic(Host1x& host1x, std::shared_ptr<Nvdec> nvdec_processor);
~Vic();
/// Write to the device state.
void ProcessMethod(Method method, u32 argument);
private:
void Execute();
void WriteRGBFrame(const AVFrame* frame, const VicConfig& config);
void WriteYUVFrame(const AVFrame* frame, const VicConfig& config);
Host1x& host1x;
std::shared_ptr<Tegra::Host1x::Nvdec> nvdec_processor;
/// Avoid reallocation of the following buffers every frame, as their
/// size does not change during a stream
using AVMallocPtr = std::unique_ptr<u8, decltype(&av_free)>;
AVMallocPtr converted_frame_buffer;
std::vector<u8> luma_buffer;
std::vector<u8> chroma_buffer;
GPUVAddr config_struct_address{};
GPUVAddr output_surface_luma_address{};
GPUVAddr output_surface_chroma_address{};
SwsContext* scaler_ctx{};
s32 scaler_width{};
s32 scaler_height{};
};
} // namespace Host1x
} // namespace Tegra