From c6590ad07b384762fd90ee8852796ec681a69286 Mon Sep 17 00:00:00 2001 From: ameerj <52414509+ameerj@users.noreply.github.com> Date: Mon, 19 Dec 2022 22:40:50 -0500 Subject: [PATCH] scratch_buffer: Explicitly defing resize and resize_destructive functions resize keeps previous data intact when the buffer grows resize_destructive destroys the previous data when the buffer grows --- src/common/scratch_buffer.h | 17 +++++ src/tests/common/scratch_buffer.cpp | 78 +++++++++++++++++++++- src/video_core/buffer_cache/buffer_cache.h | 2 +- src/video_core/dma_pusher.cpp | 2 +- src/video_core/engines/engine_upload.cpp | 4 +- src/video_core/engines/maxwell_dma.cpp | 18 ++--- src/video_core/host1x/vic.cpp | 6 +- 7 files changed, 108 insertions(+), 19 deletions(-) diff --git a/src/common/scratch_buffer.h b/src/common/scratch_buffer.h index 59bb8a9ea9..1245a5086f 100644 --- a/src/common/scratch_buffer.h +++ b/src/common/scratch_buffer.h @@ -25,7 +25,20 @@ public: ~ScratchBuffer() = default; /// This will only grow the buffer's capacity if size is greater than the current capacity. + /// The previously held data will remain intact. void resize(size_t size) { + if (size > buffer_capacity) { + auto new_buffer = Common::make_unique_for_overwrite(size); + std::move(buffer.get(), buffer.get() + buffer_capacity, new_buffer.get()); + buffer = std::move(new_buffer); + buffer_capacity = size; + } + last_requested_size = size; + } + + /// This will only grow the buffer's capacity if size is greater than the current capacity. + /// The previously held data will be destroyed if a reallocation occurs. + void resize_destructive(size_t size) { if (size > buffer_capacity) { buffer_capacity = size; buffer = Common::make_unique_for_overwrite(buffer_capacity); @@ -61,6 +74,10 @@ public: return buffer[i]; } + [[nodiscard]] const T& operator[](size_t i) const { + return buffer[i]; + } + [[nodiscard]] size_t size() const noexcept { return last_requested_size; } diff --git a/src/tests/common/scratch_buffer.cpp b/src/tests/common/scratch_buffer.cpp index a59490f559..b602c8d0a6 100644 --- a/src/tests/common/scratch_buffer.cpp +++ b/src/tests/common/scratch_buffer.cpp @@ -29,7 +29,7 @@ TEST_CASE("ScratchBuffer: Basic Test", "[common]") { } } -TEST_CASE("ScratchBuffer: Resize Grow", "[common]") { +TEST_CASE("ScratchBuffer: resize_destructive Grow", "[common]") { std::array payload; payload.fill(66); @@ -38,14 +38,86 @@ TEST_CASE("ScratchBuffer: Resize Grow", "[common]") { REQUIRE(buf.capacity() == payload.size()); // Increasing the size should reallocate the buffer - buf.resize(payload.size() * 2); + buf.resize_destructive(payload.size() * 2); REQUIRE(buf.size() == payload.size() * 2); REQUIRE(buf.capacity() == payload.size() * 2); // Since the buffer is not value initialized, reading its data will be garbage } -TEST_CASE("ScratchBuffer: Resize Shrink", "[common]") { +TEST_CASE("ScratchBuffer: resize_destructive Shrink", "[common]") { + std::array payload; + payload.fill(66); + + ScratchBuffer buf(payload.size()); + REQUIRE(buf.size() == payload.size()); + REQUIRE(buf.capacity() == payload.size()); + + std::memcpy(buf.data(), payload.data(), payload.size()); + for (size_t i = 0; i < payload.size(); ++i) { + REQUIRE(buf[i] == payload[i]); + } + + // Decreasing the size should not cause a buffer reallocation + // This can be tested by ensuring the buffer capacity and data has not changed, + buf.resize_destructive(1U); + REQUIRE(buf.size() == 1U); + REQUIRE(buf.capacity() == payload.size()); + + for (size_t i = 0; i < payload.size(); ++i) { + REQUIRE(buf[i] == payload[i]); + } +} + +TEST_CASE("ScratchBuffer: resize Grow u8", "[common]") { + std::array payload; + payload.fill(66); + + ScratchBuffer buf(payload.size()); + REQUIRE(buf.size() == payload.size()); + REQUIRE(buf.capacity() == payload.size()); + + std::memcpy(buf.data(), payload.data(), payload.size()); + for (size_t i = 0; i < payload.size(); ++i) { + REQUIRE(buf[i] == payload[i]); + } + + // Increasing the size should reallocate the buffer + buf.resize(payload.size() * 2); + REQUIRE(buf.size() == payload.size() * 2); + REQUIRE(buf.capacity() == payload.size() * 2); + + // resize() keeps the previous data intact + for (size_t i = 0; i < payload.size(); ++i) { + REQUIRE(buf[i] == payload[i]); + } +} + +TEST_CASE("ScratchBuffer: resize Grow u64", "[common]") { + std::array payload; + payload.fill(6666); + + ScratchBuffer buf(payload.size()); + REQUIRE(buf.size() == payload.size()); + REQUIRE(buf.capacity() == payload.size()); + + std::memcpy(buf.data(), payload.data(), payload.size() * sizeof(u64)); + for (size_t i = 0; i < payload.size(); ++i) { + REQUIRE(buf[i] == payload[i]); + } + + // Increasing the size should reallocate the buffer + buf.resize(payload.size() * 2); + REQUIRE(buf.size() == payload.size() * 2); + REQUIRE(buf.capacity() == payload.size() * 2); + + // resize() keeps the previous data intact + for (size_t i = 0; i < payload.size(); ++i) { + REQUIRE(buf[i] == payload[i]); + } +} + +TEST_CASE("ScratchBuffer: resize Shrink", "[common]") { std::array payload; payload.fill(66); diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index a8bd5585b0..6c8d98946d 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -1926,7 +1926,7 @@ std::span BufferCache

::ImmediateBufferWithData(VAddr cpu_addr, size template std::span BufferCache

::ImmediateBuffer(size_t wanted_capacity) { - immediate_buffer_alloc.resize(wanted_capacity); + immediate_buffer_alloc.resize_destructive(wanted_capacity); return std::span(immediate_buffer_alloc.data(), wanted_capacity); } diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp index d1f541bf5e..322de2606b 100644 --- a/src/video_core/dma_pusher.cpp +++ b/src/video_core/dma_pusher.cpp @@ -74,7 +74,7 @@ bool DmaPusher::Step() { } // Push buffer non-empty, read a word - command_headers.resize(command_list_header.size); + command_headers.resize_destructive(command_list_header.size); if (Settings::IsGPULevelHigh()) { memory_manager.ReadBlock(dma_get, command_headers.data(), command_list_header.size * sizeof(u32)); diff --git a/src/video_core/engines/engine_upload.cpp b/src/video_core/engines/engine_upload.cpp index e4f8331ab8..cea1dd8b0f 100644 --- a/src/video_core/engines/engine_upload.cpp +++ b/src/video_core/engines/engine_upload.cpp @@ -24,7 +24,7 @@ void State::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) { void State::ProcessExec(const bool is_linear_) { write_offset = 0; copy_size = regs.line_length_in * regs.line_count; - inner_buffer.resize(copy_size); + inner_buffer.resize_destructive(copy_size); is_linear = is_linear_; } @@ -70,7 +70,7 @@ void State::ProcessData(std::span read_buffer) { const std::size_t dst_size = Tegra::Texture::CalculateSize( true, bytes_per_pixel, width, regs.dest.height, regs.dest.depth, regs.dest.BlockHeight(), regs.dest.BlockDepth()); - tmp_buffer.resize(dst_size); + tmp_buffer.resize_destructive(dst_size); memory_manager.ReadBlock(address, tmp_buffer.data(), dst_size); Tegra::Texture::SwizzleSubrect(tmp_buffer, read_buffer, bytes_per_pixel, width, regs.dest.height, regs.dest.depth, x_offset, regs.dest.y, diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp index dc873732eb..f73d7bf0f7 100644 --- a/src/video_core/engines/maxwell_dma.cpp +++ b/src/video_core/engines/maxwell_dma.cpp @@ -184,8 +184,8 @@ void MaxwellDMA::CopyBlockLinearToPitch() { const size_t src_size = CalculateSize(true, bytes_per_pixel, width, height, depth, block_height, block_depth); - read_buffer.resize(src_size); - write_buffer.resize(dst_size); + read_buffer.resize_destructive(src_size); + write_buffer.resize_destructive(dst_size); memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size); memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size); @@ -231,8 +231,8 @@ void MaxwellDMA::CopyPitchToBlockLinear() { CalculateSize(true, bytes_per_pixel, width, height, depth, block_height, block_depth); const size_t src_size = static_cast(regs.pitch_in) * regs.line_count; - read_buffer.resize(src_size); - write_buffer.resize(dst_size); + read_buffer.resize_destructive(src_size); + write_buffer.resize_destructive(dst_size); memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size); if (Settings::IsGPULevelExtreme()) { @@ -261,8 +261,8 @@ void MaxwellDMA::FastCopyBlockLinearToPitch() { pos_x = pos_x % x_in_gob; pos_y = pos_y % 8; - read_buffer.resize(src_size); - write_buffer.resize(dst_size); + read_buffer.resize_destructive(src_size); + write_buffer.resize_destructive(dst_size); if (Settings::IsGPULevelExtreme()) { memory_manager.ReadBlock(regs.offset_in + offset, read_buffer.data(), src_size); @@ -321,10 +321,10 @@ void MaxwellDMA::CopyBlockLinearToBlockLinear() { const u32 pitch = x_elements * bytes_per_pixel; const size_t mid_buffer_size = pitch * regs.line_count; - read_buffer.resize(src_size); - write_buffer.resize(dst_size); + read_buffer.resize_destructive(src_size); + write_buffer.resize_destructive(dst_size); - intermediate_buffer.resize(mid_buffer_size); + intermediate_buffer.resize_destructive(mid_buffer_size); memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size); memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size); diff --git a/src/video_core/host1x/vic.cpp b/src/video_core/host1x/vic.cpp index ac0b7d20e0..36a04e4e0b 100644 --- a/src/video_core/host1x/vic.cpp +++ b/src/video_core/host1x/vic.cpp @@ -155,7 +155,7 @@ void Vic::WriteRGBFrame(const AVFrame* frame, const VicConfig& config) { // swizzle pitch linear to block linear const u32 block_height = static_cast(config.block_linear_height_log2); const auto size = Texture::CalculateSize(true, 4, width, height, 1, block_height, 0); - luma_buffer.resize(size); + luma_buffer.resize_destructive(size); std::span 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); @@ -181,8 +181,8 @@ void Vic::WriteYUVFrame(const AVFrame* frame, const VicConfig& config) { const auto stride = static_cast(frame->linesize[0]); - luma_buffer.resize(aligned_width * surface_height); - chroma_buffer.resize(aligned_width * surface_height / 2); + luma_buffer.resize_destructive(aligned_width * surface_height); + chroma_buffer.resize_destructive(aligned_width * surface_height / 2); // Populate luma buffer const u8* luma_src = frame->data[0];