From 606a62d4c740eec470b19e57322645de20866abf Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Tue, 26 May 2020 00:57:38 -0300 Subject: gl_rasterizer: Port front face flip check from Vulkan While Vulkan was assuming we had no negative viewports, OpenGL code was assuming we had them. Port the old code from Vulkan to OpenGL, checking if the first viewport is negative before flipping faces. This is not a complete implementation since we only check for the first viewport to be negative. That said, unless a game is using Vulkan, OpenGL and NVN games should be fine here, and we can always compare with our Vulkan backend to see if there's a difference. --- src/video_core/renderer_opengl/gl_rasterizer.cpp | 25 +++++++++++++++++++----- 1 file changed, 20 insertions(+), 5 deletions(-) (limited to 'src/video_core/renderer_opengl/gl_rasterizer.cpp') diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 8116a5daa..efbbf11f9 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -1031,6 +1031,26 @@ void RasterizerOpenGL::SyncViewport() { const auto& regs = gpu.regs; const bool dirty_viewport = flags[Dirty::Viewports]; + const bool dirty_clip_control = flags[Dirty::ClipControl]; + + if (dirty_clip_control || flags[Dirty::FrontFace]) { + flags[Dirty::FrontFace] = false; + + GLenum mode = MaxwellToGL::FrontFace(regs.front_face); + if (regs.screen_y_control.triangle_rast_flip != 0 && + regs.viewport_transform[0].scale_y < 0.0f) { + switch (mode) { + case GL_CW: + mode = GL_CCW; + break; + case GL_CCW: + mode = GL_CW; + break; + } + } + glFrontFace(mode); + } + if (dirty_viewport || flags[Dirty::ClipControl]) { flags[Dirty::ClipControl] = false; @@ -1128,11 +1148,6 @@ void RasterizerOpenGL::SyncCullMode() { glDisable(GL_CULL_FACE); } } - - if (flags[Dirty::FrontFace]) { - flags[Dirty::FrontFace] = false; - glFrontFace(MaxwellToGL::FrontFace(regs.front_face)); - } } void RasterizerOpenGL::SyncPrimitiveRestart() { -- cgit v1.2.3 From bb8ef381526cd9cc0a52f00383c755913a20aa43 Mon Sep 17 00:00:00 2001 From: Morph <39850852+Morph1984@users.noreply.github.com> Date: Sun, 31 May 2020 03:12:21 -0400 Subject: gl_device: Enable compute shaders for Intel proprietary drivers Previously we were disabling compute shaders on Intel's proprietary driver due to broken compute. This has been fixed in the latest Intel drivers. Re-enable compute for Intel proprietary drivers and remove the check for broken compute. --- src/video_core/renderer_opengl/gl_rasterizer.cpp | 4 ---- 1 file changed, 4 deletions(-) (limited to 'src/video_core/renderer_opengl/gl_rasterizer.cpp') diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 716d43e65..61cf99b9d 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -655,10 +655,6 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { } void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) { - if (device.HasBrokenCompute()) { - return; - } - buffer_cache.Acquire(); current_cbuf = 0; -- cgit v1.2.3 From ee21e4ecd372d6a191244dffbc5ac5c7b80150c1 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Thu, 28 May 2020 17:06:22 -0300 Subject: glsl: Squash constant buffers into a single SSBO when we hit the limit Avoids compilation errors at the cost of shader build times and runtime performance when a game hits the limit of uniform buffers we can use. --- src/video_core/renderer_opengl/gl_rasterizer.cpp | 78 ++++++++++++++++++------ 1 file changed, 59 insertions(+), 19 deletions(-) (limited to 'src/video_core/renderer_opengl/gl_rasterizer.cpp') diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 61cf99b9d..b0f303a3c 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -54,6 +54,12 @@ MICROPROFILE_DEFINE(OpenGL_PrimitiveAssembly, "OpenGL", "Prim Asmbl", MP_RGB(255 namespace { +constexpr std::size_t NUM_CONST_BUFFERS_PER_STAGE = 18; +constexpr std::size_t NUM_CONST_BUFFERS_BYTES_PER_STAGE = + NUM_CONST_BUFFERS_PER_STAGE * Maxwell::MaxConstBufferSize; +constexpr std::size_t TOTAL_CONST_BUFFER_BYTES = + NUM_CONST_BUFFERS_BYTES_PER_STAGE * Maxwell::MaxShaderStage; + constexpr std::size_t NumSupportedVertexAttributes = 16; template @@ -104,6 +110,9 @@ RasterizerOpenGL::RasterizerOpenGL(Core::System& system, Core::Frontend::EmuWind screen_info{info}, program_manager{program_manager}, state_tracker{state_tracker} { CheckExtensions(); + unified_uniform_buffer.Create(); + glNamedBufferStorage(unified_uniform_buffer.handle, TOTAL_CONST_BUFFER_BYTES, nullptr, 0); + if (device.UseAssemblyShaders()) { glCreateBuffers(static_cast(staging_cbufs.size()), staging_cbufs.data()); for (const GLuint cbuf : staging_cbufs) { @@ -842,34 +851,56 @@ void RasterizerOpenGL::SetupDrawConstBuffers(std::size_t stage_index, const Shad MICROPROFILE_SCOPE(OpenGL_UBO); const auto& stages = system.GPU().Maxwell3D().state.shader_stages; const auto& shader_stage = stages[stage_index]; + const auto& entries = shader->GetEntries(); + const bool use_unified = entries.use_unified_uniforms; + const std::size_t base_unified_offset = stage_index * NUM_CONST_BUFFERS_BYTES_PER_STAGE; - u32 binding = - device.UseAssemblyShaders() ? 0 : device.GetBaseBindings(stage_index).uniform_buffer; - for (const auto& entry : shader->GetEntries().const_buffers) { - const auto& buffer = shader_stage.const_buffers[entry.GetIndex()]; - SetupConstBuffer(PARAMETER_LUT[stage_index], binding++, buffer, entry); + const auto base_bindings = device.GetBaseBindings(stage_index); + u32 binding = device.UseAssemblyShaders() ? 0 : base_bindings.uniform_buffer; + for (const auto& entry : entries.const_buffers) { + const u32 index = entry.GetIndex(); + const auto& buffer = shader_stage.const_buffers[index]; + SetupConstBuffer(PARAMETER_LUT[stage_index], binding, buffer, entry, use_unified, + base_unified_offset + index * Maxwell::MaxConstBufferSize); + ++binding; + } + if (use_unified) { + const u32 index = static_cast(base_bindings.shader_storage_buffer + + entries.global_memory_entries.size()); + glBindBufferRange(GL_SHADER_STORAGE_BUFFER, index, unified_uniform_buffer.handle, + base_unified_offset, NUM_CONST_BUFFERS_BYTES_PER_STAGE); } } void RasterizerOpenGL::SetupComputeConstBuffers(const Shader& kernel) { MICROPROFILE_SCOPE(OpenGL_UBO); const auto& launch_desc = system.GPU().KeplerCompute().launch_description; + const auto& entries = kernel->GetEntries(); + const bool use_unified = entries.use_unified_uniforms; u32 binding = 0; - for (const auto& entry : kernel->GetEntries().const_buffers) { + for (const auto& entry : entries.const_buffers) { const auto& config = launch_desc.const_buffer_config[entry.GetIndex()]; const std::bitset<8> mask = launch_desc.const_buffer_enable_mask.Value(); Tegra::Engines::ConstBufferInfo buffer; buffer.address = config.Address(); buffer.size = config.size; buffer.enabled = mask[entry.GetIndex()]; - SetupConstBuffer(GL_COMPUTE_PROGRAM_PARAMETER_BUFFER_NV, binding++, buffer, entry); + SetupConstBuffer(GL_COMPUTE_PROGRAM_PARAMETER_BUFFER_NV, binding, buffer, entry, + use_unified, entry.GetIndex() * Maxwell::MaxConstBufferSize); + ++binding; + } + if (use_unified) { + const GLuint index = static_cast(entries.global_memory_entries.size()); + glBindBufferRange(GL_SHADER_STORAGE_BUFFER, index, unified_uniform_buffer.handle, 0, + NUM_CONST_BUFFERS_BYTES_PER_STAGE); } } void RasterizerOpenGL::SetupConstBuffer(GLenum stage, u32 binding, const Tegra::Engines::ConstBufferInfo& buffer, - const ConstBufferEntry& entry) { + const ConstBufferEntry& entry, bool use_unified, + std::size_t unified_offset) { if (!buffer.enabled) { // Set values to zero to unbind buffers if (device.UseAssemblyShaders()) { @@ -885,20 +916,29 @@ void RasterizerOpenGL::SetupConstBuffer(GLenum stage, u32 binding, // UBO alignment requirements. const std::size_t size = Common::AlignUp(GetConstBufferSize(buffer, entry), sizeof(GLvec4)); - const auto alignment = device.GetUniformBufferAlignment(); - auto [cbuf, offset] = buffer_cache.UploadMemory(buffer.address, size, alignment, false, - device.HasFastBufferSubData()); - if (!device.UseAssemblyShaders()) { - glBindBufferRange(GL_UNIFORM_BUFFER, binding, cbuf, offset, size); + const bool fast_upload = !use_unified && device.HasFastBufferSubData(); + + const std::size_t alignment = use_unified ? 4 : device.GetUniformBufferAlignment(); + const GPUVAddr gpu_addr = buffer.address; + auto [cbuf, offset] = buffer_cache.UploadMemory(gpu_addr, size, alignment, false, fast_upload); + + if (device.UseAssemblyShaders()) { + UNIMPLEMENTED_IF(use_unified); + if (offset != 0) { + const GLuint staging_cbuf = staging_cbufs[current_cbuf++]; + glCopyNamedBufferSubData(cbuf, staging_cbuf, offset, 0, size); + cbuf = staging_cbuf; + offset = 0; + } + glBindBufferRangeNV(stage, binding, cbuf, offset, size); return; } - if (offset != 0) { - const GLuint staging_cbuf = staging_cbufs[current_cbuf++]; - glCopyNamedBufferSubData(cbuf, staging_cbuf, offset, 0, size); - cbuf = staging_cbuf; - offset = 0; + + if (use_unified) { + glCopyNamedBufferSubData(cbuf, unified_uniform_buffer.handle, offset, unified_offset, size); + } else { + glBindBufferRange(GL_UNIFORM_BUFFER, binding, cbuf, offset, size); } - glBindBufferRangeNV(stage, binding, cbuf, offset, size); } void RasterizerOpenGL::SetupDrawGlobalMemory(std::size_t stage_index, const Shader& shader) { -- cgit v1.2.3 From 3d99b449d3c285a50ffb68ae077882cc2998d62f Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Wed, 3 Jun 2020 02:48:41 -0300 Subject: gl_rasterizer: Use NV_transform_feedback for XFB on assembly shaders NV_transform_feedback, NV_transform_feedback2 and ARB_transform_feedback3 with NV_transform_feedback interactions allows implementing transform feedbacks as dynamic state. Maxwell implements transform feedbacks as dynamic state, so using these extensions with TransformFeedbackStreamAttribsNV allows us to properly emulate transform feedbacks without having to recompile shaders when the state changes. --- src/video_core/renderer_opengl/gl_rasterizer.cpp | 90 ++++++++++++++++++++++++ 1 file changed, 90 insertions(+) (limited to 'src/video_core/renderer_opengl/gl_rasterizer.cpp') diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 55e79aaf6..f802fd384 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -93,6 +93,34 @@ std::size_t GetConstBufferSize(const Tegra::Engines::ConstBufferInfo& buffer, return buffer.size; } +/// Translates hardware transform feedback indices +/// @param location Hardware location +/// @return Pair of ARB_transform_feedback3 token stream first and third arguments +/// @note Read https://www.khronos.org/registry/OpenGL/extensions/ARB/ARB_transform_feedback3.txt +std::pair TransformFeedbackEnum(u8 location) { + const u8 index = location / 4; + if (index >= 8 && index <= 39) { + return {GL_GENERIC_ATTRIB_NV, index - 8}; + } + if (index >= 48 && index <= 55) { + return {GL_TEXTURE_COORD_NV, index - 48}; + } + switch (index) { + case 7: + return {GL_POSITION, 0}; + case 40: + return {GL_PRIMARY_COLOR_NV, 0}; + case 41: + return {GL_SECONDARY_COLOR_NV, 0}; + case 42: + return {GL_BACK_PRIMARY_COLOR_NV, 0}; + case 43: + return {GL_BACK_SECONDARY_COLOR_NV, 0}; + } + UNIMPLEMENTED_MSG("index={}", static_cast(index)); + return {GL_POSITION, 0}; +} + void oglEnable(GLenum cap, bool state) { (state ? glEnable : glDisable)(cap); } @@ -1547,12 +1575,70 @@ void RasterizerOpenGL::SyncFramebufferSRGB() { oglEnable(GL_FRAMEBUFFER_SRGB, gpu.regs.framebuffer_srgb); } +void RasterizerOpenGL::SyncTransformFeedback() { + // TODO(Rodrigo): Inject SKIP_COMPONENTS*_NV when required. An unimplemented message will signal + // when this is required. + const auto& regs = system.GPU().Maxwell3D().regs; + + static constexpr std::size_t STRIDE = 3; + std::array attribs; + std::array streams; + + GLint* cursor = attribs.data(); + GLint* current_stream = streams.data(); + + for (std::size_t feedback = 0; feedback < Maxwell::NumTransformFeedbackBuffers; ++feedback) { + const auto& layout = regs.tfb_layouts[feedback]; + UNIMPLEMENTED_IF_MSG(layout.stride != layout.varying_count * 4, "Stride padding"); + if (layout.varying_count == 0) { + continue; + } + + *current_stream = static_cast(feedback); + if (current_stream != streams.data()) { + // When stepping one stream, push the expected token + cursor[0] = GL_NEXT_BUFFER_NV; + cursor[1] = 0; + cursor[2] = 0; + cursor += STRIDE; + } + ++current_stream; + + const auto& locations = regs.tfb_varying_locs[feedback]; + std::optional current_index; + for (u32 offset = 0; offset < layout.varying_count; ++offset) { + const u8 location = locations[offset]; + const u8 index = location / 4; + + if (current_index == index) { + // Increase number of components of the previous attachment + ++cursor[-2]; + continue; + } + current_index = index; + + std::tie(cursor[0], cursor[2]) = TransformFeedbackEnum(location); + cursor[1] = 1; + cursor += STRIDE; + } + } + + const GLsizei num_attribs = static_cast((cursor - attribs.data()) / STRIDE); + const GLsizei num_strides = static_cast(current_stream - streams.data()); + glTransformFeedbackStreamAttribsNV(num_attribs, attribs.data(), num_strides, streams.data(), + GL_INTERLEAVED_ATTRIBS); +} + void RasterizerOpenGL::BeginTransformFeedback(GLenum primitive_mode) { const auto& regs = system.GPU().Maxwell3D().regs; if (regs.tfb_enabled == 0) { return; } + if (device.UseAssemblyShaders()) { + SyncTransformFeedback(); + } + UNIMPLEMENTED_IF(regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationControl) || regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationEval) || regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::Geometry)); @@ -1579,6 +1665,10 @@ void RasterizerOpenGL::BeginTransformFeedback(GLenum primitive_mode) { static_cast(size)); } + // We may have to call BeginTransformFeedbackNV here since they seem to call different + // implementations on Nvidia's driver (the pointer is different) but we are using + // ARB_transform_feedback3 features with NV_transform_feedback interactions and the ARB + // extension doesn't define BeginTransformFeedback (without NV) interactions. It just works. glBeginTransformFeedback(GL_POINTS); } -- cgit v1.2.3 From 5b2b6d594c6cfa77c3fb92faee63ad524bfe7204 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Thu, 4 Jun 2020 23:03:49 -0300 Subject: shader/texture: Join separate image and sampler pairs offline Games using D3D idioms can join images and samplers when a shader executes, instead of baking them into a combined sampler image. This is also possible on Vulkan. One approach to this solution would be to use separate samplers on Vulkan and leave this unimplemented on OpenGL, but we can't do this because there's no consistent way of determining which constant buffer holds a sampler and which one an image. We could in theory find the first bit and if it's in the TIC area, it's an image; but this falls apart when an image or sampler handle use an index of zero. The used approach is to track for a LOP.OR operation (this is done at an IR level, not at an ISA level), track again the constant buffers used as source and store this pair. Then, outside of shader execution, join the sample and image pair with a bitwise or operation. This approach won't work on games that truly use separate samplers in a meaningful way. For example, pooling textures in a 2D array and determining at runtime what sampler to use. This invalidates OpenGL's disk shader cache :) - Used mostly by D3D ports to Switch --- src/video_core/renderer_opengl/gl_rasterizer.cpp | 16 ++++++++++++++-- 1 file changed, 14 insertions(+), 2 deletions(-) (limited to 'src/video_core/renderer_opengl/gl_rasterizer.cpp') diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 55e79aaf6..aedcdcb78 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -65,10 +65,22 @@ constexpr std::size_t NumSupportedVertexAttributes = 16; template Tegra::Texture::FullTextureInfo GetTextureInfo(const Engine& engine, const Entry& entry, ShaderType shader_type, std::size_t index = 0) { + if constexpr (std::is_same_v) { + if (entry.is_separated) { + const u32 buffer_1 = entry.buffer; + const u32 buffer_2 = entry.secondary_buffer; + const u32 offset_1 = entry.offset; + const u32 offset_2 = entry.secondary_offset; + const u32 handle_1 = engine.AccessConstBuffer32(shader_type, buffer_1, offset_1); + const u32 handle_2 = engine.AccessConstBuffer32(shader_type, buffer_2, offset_2); + return engine.GetTextureInfo(handle_1 | handle_2); + } + } if (entry.is_bindless) { - const auto tex_handle = engine.AccessConstBuffer32(shader_type, entry.buffer, entry.offset); - return engine.GetTextureInfo(tex_handle); + const u32 handle = engine.AccessConstBuffer32(shader_type, entry.buffer, entry.offset); + return engine.GetTextureInfo(handle); } + const auto& gpu_profile = engine.AccessGuestDriverProfile(); const u32 offset = entry.offset + static_cast(index * gpu_profile.GetTextureHandlerSize()); if constexpr (std::is_same_v) { -- cgit v1.2.3 From b96f65b62b143544716f66a65e93d2dcb141ee2b Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Fri, 22 May 2020 20:55:38 -0300 Subject: gl_shader_cache: Use generic shader cache Trivially port the generic shader cache to OpenGL. --- src/video_core/renderer_opengl/gl_rasterizer.cpp | 19 ++++++++++--------- 1 file changed, 10 insertions(+), 9 deletions(-) (limited to 'src/video_core/renderer_opengl/gl_rasterizer.cpp') diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 55e79aaf6..909ca9e0e 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -30,6 +30,7 @@ #include "video_core/renderer_opengl/gl_shader_cache.h" #include "video_core/renderer_opengl/maxwell_to_gl.h" #include "video_core/renderer_opengl/renderer_opengl.h" +#include "video_core/shader_cache.h" namespace OpenGL { @@ -282,7 +283,7 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) { continue; } - Shader shader{shader_cache.GetStageProgram(program)}; + Shader* const shader = shader_cache.GetStageProgram(program); if (device.UseAssemblyShaders()) { // Check for ARB limitation. We only have 16 SSBOs per context state. To workaround this @@ -842,7 +843,7 @@ bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config, return true; } -void RasterizerOpenGL::SetupDrawConstBuffers(std::size_t stage_index, const Shader& shader) { +void RasterizerOpenGL::SetupDrawConstBuffers(std::size_t stage_index, Shader* shader) { static constexpr std::array PARAMETER_LUT = { GL_VERTEX_PROGRAM_PARAMETER_BUFFER_NV, GL_TESS_CONTROL_PROGRAM_PARAMETER_BUFFER_NV, GL_TESS_EVALUATION_PROGRAM_PARAMETER_BUFFER_NV, GL_GEOMETRY_PROGRAM_PARAMETER_BUFFER_NV, @@ -872,7 +873,7 @@ void RasterizerOpenGL::SetupDrawConstBuffers(std::size_t stage_index, const Shad } } -void RasterizerOpenGL::SetupComputeConstBuffers(const Shader& kernel) { +void RasterizerOpenGL::SetupComputeConstBuffers(Shader* kernel) { MICROPROFILE_SCOPE(OpenGL_UBO); const auto& launch_desc = system.GPU().KeplerCompute().launch_description; const auto& entries = kernel->GetEntries(); @@ -941,7 +942,7 @@ void RasterizerOpenGL::SetupConstBuffer(GLenum stage, u32 binding, } } -void RasterizerOpenGL::SetupDrawGlobalMemory(std::size_t stage_index, const Shader& shader) { +void RasterizerOpenGL::SetupDrawGlobalMemory(std::size_t stage_index, Shader* shader) { auto& gpu{system.GPU()}; auto& memory_manager{gpu.MemoryManager()}; const auto cbufs{gpu.Maxwell3D().state.shader_stages[stage_index]}; @@ -956,7 +957,7 @@ void RasterizerOpenGL::SetupDrawGlobalMemory(std::size_t stage_index, const Shad } } -void RasterizerOpenGL::SetupComputeGlobalMemory(const Shader& kernel) { +void RasterizerOpenGL::SetupComputeGlobalMemory(Shader* kernel) { auto& gpu{system.GPU()}; auto& memory_manager{gpu.MemoryManager()}; const auto cbufs{gpu.KeplerCompute().launch_description.const_buffer_config}; @@ -979,7 +980,7 @@ void RasterizerOpenGL::SetupGlobalMemory(u32 binding, const GlobalMemoryEntry& e static_cast(size)); } -void RasterizerOpenGL::SetupDrawTextures(std::size_t stage_index, const Shader& shader) { +void RasterizerOpenGL::SetupDrawTextures(std::size_t stage_index, Shader* shader) { MICROPROFILE_SCOPE(OpenGL_Texture); const auto& maxwell3d = system.GPU().Maxwell3D(); u32 binding = device.GetBaseBindings(stage_index).sampler; @@ -992,7 +993,7 @@ void RasterizerOpenGL::SetupDrawTextures(std::size_t stage_index, const Shader& } } -void RasterizerOpenGL::SetupComputeTextures(const Shader& kernel) { +void RasterizerOpenGL::SetupComputeTextures(Shader* kernel) { MICROPROFILE_SCOPE(OpenGL_Texture); const auto& compute = system.GPU().KeplerCompute(); u32 binding = 0; @@ -1021,7 +1022,7 @@ void RasterizerOpenGL::SetupTexture(u32 binding, const Tegra::Texture::FullTextu } } -void RasterizerOpenGL::SetupDrawImages(std::size_t stage_index, const Shader& shader) { +void RasterizerOpenGL::SetupDrawImages(std::size_t stage_index, Shader* shader) { const auto& maxwell3d = system.GPU().Maxwell3D(); u32 binding = device.GetBaseBindings(stage_index).image; for (const auto& entry : shader->GetEntries().images) { @@ -1031,7 +1032,7 @@ void RasterizerOpenGL::SetupDrawImages(std::size_t stage_index, const Shader& sh } } -void RasterizerOpenGL::SetupComputeImages(const Shader& shader) { +void RasterizerOpenGL::SetupComputeImages(Shader* shader) { const auto& compute = system.GPU().KeplerCompute(); u32 binding = 0; for (const auto& entry : shader->GetEntries().images) { -- cgit v1.2.3 From 7646f2c21dcd3e36073142da84d56cd7640476a9 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Mon, 8 Jun 2020 20:24:16 -0300 Subject: gl_rasterizer: Mark vertex buffers as dirty after buffer cache invalidation Vertex buffers bindings become invalid after the stream buffer is invalidated. We were originally doing this, but it got lost at some point. - Fixes Animal Crossing: New Horizons, but it affects everything. --- src/video_core/renderer_opengl/gl_rasterizer.cpp | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-) (limited to 'src/video_core/renderer_opengl/gl_rasterizer.cpp') diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 55e79aaf6..5673e30b3 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -576,7 +576,16 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { (Maxwell::MaxConstBufferSize + device.GetUniformBufferAlignment()); // Prepare the vertex array. - buffer_cache.Map(buffer_size); + const bool invalidated = buffer_cache.Map(buffer_size); + + if (invalidated) { + // When the stream buffer has been invalidated, we have to consider vertex buffers as dirty + auto& dirty = gpu.dirty.flags; + dirty[Dirty::VertexBuffers] = true; + for (int index = Dirty::VertexBuffer0; index <= Dirty::VertexBuffer31; ++index) { + dirty[index] = true; + } + } // Prepare vertex array format. SetupVertexFormat(); -- cgit v1.2.3 From 32485917ba7cb7b2f0cad766c0897365294650a7 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Mon, 11 May 2020 16:35:04 -0300 Subject: gl_buffer_cache: Mark buffers as resident Make stream buffer and cached buffers as resident and query their address. This allows us to use GPU addresses for several proprietary Nvidia extensions. --- src/video_core/renderer_opengl/gl_rasterizer.cpp | 44 ++++++++++++------------ 1 file changed, 22 insertions(+), 22 deletions(-) (limited to 'src/video_core/renderer_opengl/gl_rasterizer.cpp') diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 2d6c11320..7cb378a71 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -253,8 +253,8 @@ void RasterizerOpenGL::SetupVertexBuffer() { glBindVertexBuffer(static_cast(index), 0, 0, vertex_array.stride); continue; } - const auto [vertex_buffer, vertex_buffer_offset] = buffer_cache.UploadMemory(start, size); - glBindVertexBuffer(static_cast(index), vertex_buffer, vertex_buffer_offset, + const auto info = buffer_cache.UploadMemory(start, size); + glBindVertexBuffer(static_cast(index), info.handle, info.offset, vertex_array.stride); } } @@ -285,9 +285,9 @@ GLintptr RasterizerOpenGL::SetupIndexBuffer() { MICROPROFILE_SCOPE(OpenGL_Index); const auto& regs = system.GPU().Maxwell3D().regs; const std::size_t size = CalculateIndexBufferSize(); - const auto [buffer, offset] = buffer_cache.UploadMemory(regs.index_array.IndexStart(), size); - glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, buffer); - return offset; + const auto info = buffer_cache.UploadMemory(regs.index_array.IndexStart(), size); + glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, info.handle); + return info.offset; } void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) { @@ -643,9 +643,9 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { if (!device.UseAssemblyShaders()) { MaxwellUniformData ubo; ubo.SetFromRegs(gpu); - const auto [buffer, offset] = + const auto info = buffer_cache.UploadHostMemory(&ubo, sizeof(ubo), device.GetUniformBufferAlignment()); - glBindBufferRange(GL_UNIFORM_BUFFER, EmulationUniformBlockBinding, buffer, offset, + glBindBufferRange(GL_UNIFORM_BUFFER, EmulationUniformBlockBinding, info.handle, info.offset, static_cast(sizeof(ubo))); } @@ -956,8 +956,7 @@ void RasterizerOpenGL::SetupConstBuffer(GLenum stage, u32 binding, if (device.UseAssemblyShaders()) { glBindBufferRangeNV(stage, entry.GetIndex(), 0, 0, 0); } else { - glBindBufferRange(GL_UNIFORM_BUFFER, binding, - buffer_cache.GetEmptyBuffer(sizeof(float)), 0, sizeof(float)); + glBindBufferRange(GL_UNIFORM_BUFFER, binding, 0, 0, sizeof(float)); } return; } @@ -970,24 +969,25 @@ void RasterizerOpenGL::SetupConstBuffer(GLenum stage, u32 binding, const std::size_t alignment = use_unified ? 4 : device.GetUniformBufferAlignment(); const GPUVAddr gpu_addr = buffer.address; - auto [cbuf, offset] = buffer_cache.UploadMemory(gpu_addr, size, alignment, false, fast_upload); + auto info = buffer_cache.UploadMemory(gpu_addr, size, alignment, false, fast_upload); if (device.UseAssemblyShaders()) { UNIMPLEMENTED_IF(use_unified); - if (offset != 0) { + if (info.offset != 0) { const GLuint staging_cbuf = staging_cbufs[current_cbuf++]; - glCopyNamedBufferSubData(cbuf, staging_cbuf, offset, 0, size); - cbuf = staging_cbuf; - offset = 0; + glCopyNamedBufferSubData(info.handle, staging_cbuf, info.offset, 0, size); + info.handle = staging_cbuf; + info.offset = 0; } - glBindBufferRangeNV(stage, binding, cbuf, offset, size); + glBindBufferRangeNV(stage, binding, info.handle, info.offset, size); return; } if (use_unified) { - glCopyNamedBufferSubData(cbuf, unified_uniform_buffer.handle, offset, unified_offset, size); + glCopyNamedBufferSubData(info.handle, unified_uniform_buffer.handle, info.offset, + unified_offset, size); } else { - glBindBufferRange(GL_UNIFORM_BUFFER, binding, cbuf, offset, size); + glBindBufferRange(GL_UNIFORM_BUFFER, binding, info.handle, info.offset, size); } } @@ -1023,9 +1023,8 @@ void RasterizerOpenGL::SetupComputeGlobalMemory(Shader* kernel) { void RasterizerOpenGL::SetupGlobalMemory(u32 binding, const GlobalMemoryEntry& entry, GPUVAddr gpu_addr, std::size_t size) { const auto alignment{device.GetShaderStorageBufferAlignment()}; - const auto [ssbo, buffer_offset] = - buffer_cache.UploadMemory(gpu_addr, size, alignment, entry.is_written); - glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding, ssbo, buffer_offset, + const auto info = buffer_cache.UploadMemory(gpu_addr, size, alignment, entry.is_written); + glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding, info.handle, info.offset, static_cast(size)); } @@ -1712,8 +1711,9 @@ void RasterizerOpenGL::EndTransformFeedback() { const GLuint handle = transform_feedback_buffers[index].handle; const GPUVAddr gpu_addr = binding.Address(); const std::size_t size = binding.buffer_size; - const auto [dest_buffer, offset] = buffer_cache.UploadMemory(gpu_addr, size, 4, true); - glCopyNamedBufferSubData(handle, dest_buffer, 0, offset, static_cast(size)); + const auto info = buffer_cache.UploadMemory(gpu_addr, size, 4, true); + glCopyNamedBufferSubData(handle, info.handle, 0, info.offset, + static_cast(size)); } } -- cgit v1.2.3 From 41a4090320ee52e914e8b4c789dfe14210794fed Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Thu, 18 Jun 2020 03:56:31 -0300 Subject: gl_rasterizer: Use NV_vertex_buffer_unified_memory for vertex buffer robustness Switch games are allowed to bind less data than what they use in a vertex buffer, the expected behavior here is that these values are read as zero. At the moment of writing this only D3D12, OpenGL and NVN through NV_vertex_buffer_unified_memory support vertex buffer with a size limit. In theory this could be emulated on Vulkan creating a new VkBuffer for each (handle, offset, length) tuple and binding the expected data to it. This is likely going to be slow and memory expensive when used on the vertex buffer and we have to do it on all draws because we can't know without analyzing indices when a game is going to read vertex data out of bounds. This is not a problem on OpenGL's BufferAddressRangeNV because it takes a length parameter, unlike Vulkan's CmdBindVertexBuffers that only takes buffers and offsets (the length is implicit in VkBuffer). It isn't a problem on D3D12 either, because D3D12_VERTEX_BUFFER_VIEW on IASetVertexBuffers takes SizeInBytes as a parameter (although I am not familiar with robustness on D3D12). Currently this only implements buffer ranges for vertex buffers, although indices can also be affected. A KHR_robustness profile is not created, but Nvidia's driver reads out of bound vertex data as zero anyway, this might have to be changed in the future. - Fixes SMO random triangles when capturing an enemy, getting hit, or looking at the environment on certain maps. --- src/video_core/renderer_opengl/gl_rasterizer.cpp | 28 +++++++++++++++++------- 1 file changed, 20 insertions(+), 8 deletions(-) (limited to 'src/video_core/renderer_opengl/gl_rasterizer.cpp') diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 7cb378a71..362457ffe 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -61,7 +61,8 @@ constexpr std::size_t NUM_CONST_BUFFERS_BYTES_PER_STAGE = constexpr std::size_t TOTAL_CONST_BUFFER_BYTES = NUM_CONST_BUFFERS_BYTES_PER_STAGE * Maxwell::MaxShaderStage; -constexpr std::size_t NumSupportedVertexAttributes = 16; +constexpr std::size_t NUM_SUPPORTED_VERTEX_ATTRIBUTES = 16; +constexpr std::size_t NUM_SUPPORTED_VERTEX_BINDINGS = 16; template Tegra::Texture::FullTextureInfo GetTextureInfo(const Engine& engine, const Entry& entry, @@ -193,7 +194,7 @@ void RasterizerOpenGL::SetupVertexFormat() { // avoid OpenGL errors. // TODO(Subv): Analyze the shader to identify which attributes are actually used and don't // assume every shader uses them all. - for (std::size_t index = 0; index < NumSupportedVertexAttributes; ++index) { + for (std::size_t index = 0; index < NUM_SUPPORTED_VERTEX_ATTRIBUTES; ++index) { if (!flags[Dirty::VertexFormat0 + index]) { continue; } @@ -231,9 +232,11 @@ void RasterizerOpenGL::SetupVertexBuffer() { MICROPROFILE_SCOPE(OpenGL_VB); + const bool use_unified_memory = device.HasVertexBufferUnifiedMemory(); + // Upload all guest vertex arrays sequentially to our buffer const auto& regs = gpu.regs; - for (std::size_t index = 0; index < Maxwell::NumVertexArrays; ++index) { + for (std::size_t index = 0; index < NUM_SUPPORTED_VERTEX_BINDINGS; ++index) { if (!flags[Dirty::VertexBuffer0 + index]) { continue; } @@ -246,16 +249,25 @@ void RasterizerOpenGL::SetupVertexBuffer() { const GPUVAddr start = vertex_array.StartAddress(); const GPUVAddr end = regs.vertex_array_limit[index].LimitAddress(); - ASSERT(end >= start); + + const GLuint gl_index = static_cast(index); const u64 size = end - start; if (size == 0) { - glBindVertexBuffer(static_cast(index), 0, 0, vertex_array.stride); + glBindVertexBuffer(gl_index, 0, 0, vertex_array.stride); + if (use_unified_memory) { + glBufferAddressRangeNV(GL_VERTEX_ATTRIB_ARRAY_ADDRESS_NV, gl_index, 0, 0); + } continue; } const auto info = buffer_cache.UploadMemory(start, size); - glBindVertexBuffer(static_cast(index), info.handle, info.offset, - vertex_array.stride); + if (use_unified_memory) { + glBindVertexBuffer(gl_index, 0, 0, vertex_array.stride); + glBufferAddressRangeNV(GL_VERTEX_ATTRIB_ARRAY_ADDRESS_NV, gl_index, + info.address + info.offset, size); + } else { + glBindVertexBuffer(gl_index, info.handle, info.offset, vertex_array.stride); + } } } @@ -268,7 +280,7 @@ void RasterizerOpenGL::SetupVertexInstances() { flags[Dirty::VertexInstances] = false; const auto& regs = gpu.regs; - for (std::size_t index = 0; index < NumSupportedVertexAttributes; ++index) { + for (std::size_t index = 0; index < NUM_SUPPORTED_VERTEX_ATTRIBUTES; ++index) { if (!flags[Dirty::VertexInstance0 + index]) { continue; } -- cgit v1.2.3