From ac265a72ce4176ceb3cd10a5548ab71519771640 Mon Sep 17 00:00:00 2001
From: ameerj <aj662@drexel.edu>
Date: Mon, 23 Nov 2020 13:25:01 -0500
Subject: nvdec cleanup

---
 src/video_core/command_classes/codecs/codec.cpp |  7 ++++---
 src/video_core/command_classes/vic.cpp          | 21 +++++++++++++--------
 src/video_core/command_classes/vic.h            |  7 +++++++
 3 files changed, 24 insertions(+), 11 deletions(-)

(limited to 'src/video_core/command_classes')

diff --git a/src/video_core/command_classes/codecs/codec.cpp b/src/video_core/command_classes/codecs/codec.cpp
index 39bc923a5..d02dc6260 100644
--- a/src/video_core/command_classes/codecs/codec.cpp
+++ b/src/video_core/command_classes/codecs/codec.cpp
@@ -44,8 +44,10 @@ Codec::~Codec() {
 }
 
 void Codec::SetTargetCodec(NvdecCommon::VideoCodec codec) {
-    LOG_INFO(Service_NVDRV, "NVDEC video codec initialized to {}", codec);
-    current_codec = codec;
+    if (current_codec != codec) {
+        LOG_INFO(Service_NVDRV, "NVDEC video codec initialized to {}", static_cast<u32>(codec));
+        current_codec = codec;
+    }
 }
 
 void Codec::StateWrite(u32 offset, u64 arguments) {
@@ -55,7 +57,6 @@ void Codec::StateWrite(u32 offset, u64 arguments) {
 
 void Codec::Decode() {
     bool is_first_frame = false;
-
     if (!initialized) {
         if (current_codec == NvdecCommon::VideoCodec::H264) {
             av_codec = avcodec_find_decoder(AV_CODEC_ID_H264);
diff --git a/src/video_core/command_classes/vic.cpp b/src/video_core/command_classes/vic.cpp
index 2b7569335..73680d057 100644
--- a/src/video_core/command_classes/vic.cpp
+++ b/src/video_core/command_classes/vic.cpp
@@ -18,7 +18,10 @@ extern "C" {
 namespace Tegra {
 
 Vic::Vic(GPU& gpu_, std::shared_ptr<Nvdec> nvdec_processor_)
-    : gpu(gpu_), nvdec_processor(std::move(nvdec_processor_)) {}
+    : gpu(gpu_),
+      nvdec_processor(std::move(nvdec_processor_)), converted_frame_buffer{nullptr, av_free}
+
+{}
 Vic::~Vic() = default;
 
 void Vic::VicStateWrite(u32 offset, u32 arguments) {
@@ -89,8 +92,10 @@ void Vic::Execute() {
         // Get Converted frame
         const std::size_t linear_size = frame->width * frame->height * 4;
 
-        using AVMallocPtr = std::unique_ptr<u8, decltype(&av_free)>;
-        AVMallocPtr converted_frame_buffer{static_cast<u8*>(av_malloc(linear_size)), av_free};
+        // Only allocate frame_buffer once per stream, as the size is not expected to change
+        if (!converted_frame_buffer) {
+            converted_frame_buffer = AVMallocPtr{static_cast<u8*>(av_malloc(linear_size)), av_free};
+        }
 
         const int converted_stride{frame->width * 4};
         u8* const converted_frame_buf_addr{converted_frame_buffer.get()};
@@ -104,12 +109,12 @@ void Vic::Execute() {
             const u32 block_height = static_cast<u32>(config.block_linear_height_log2);
             const auto size = Tegra::Texture::CalculateSize(true, 4, frame->width, frame->height, 1,
                                                             block_height, 0);
-            std::vector<u8> swizzled_data(size);
+            luma_buffer.resize(size);
             Tegra::Texture::SwizzleSubrect(frame->width, frame->height, frame->width * 4,
-                                           frame->width, 4, swizzled_data.data(),
+                                           frame->width, 4, luma_buffer.data(),
                                            converted_frame_buffer.get(), block_height, 0, 0);
 
-            gpu.MemoryManager().WriteBlock(output_surface_luma_address, swizzled_data.data(), size);
+            gpu.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(), size);
         } else {
             // send pitch linear frame
             gpu.MemoryManager().WriteBlock(output_surface_luma_address, converted_frame_buf_addr,
@@ -132,8 +137,8 @@ void Vic::Execute() {
         const auto stride = frame->linesize[0];
         const auto half_stride = frame->linesize[1];
 
-        std::vector<u8> luma_buffer(aligned_width * surface_height);
-        std::vector<u8> chroma_buffer(aligned_width * half_height);
+        luma_buffer.resize(aligned_width * surface_height);
+        chroma_buffer.resize(aligned_width * half_height);
 
         // Populate luma buffer
         for (std::size_t y = 0; y < surface_height - 1; ++y) {
diff --git a/src/video_core/command_classes/vic.h b/src/video_core/command_classes/vic.h
index 8c4e284a1..6eaf72f21 100644
--- a/src/video_core/command_classes/vic.h
+++ b/src/video_core/command_classes/vic.h
@@ -97,6 +97,13 @@ private:
     GPU& gpu;
     std::shared_ptr<Tegra::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_u_address{};
-- 
cgit v1.2.3