aboutsummaryrefslogtreecommitdiff
path: root/src/video_core
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core')
-rw-r--r--src/video_core/engines/maxwell_3d.cpp109
-rw-r--r--src/video_core/engines/maxwell_3d.h50
-rw-r--r--src/video_core/engines/maxwell_dma.cpp20
-rw-r--r--src/video_core/engines/maxwell_dma.h2
-rw-r--r--src/video_core/renderer_opengl/gl_fence_manager.cpp3
-rw-r--r--src/video_core/renderer_opengl/gl_graphics_pipeline.cpp46
-rw-r--r--src/video_core/renderer_opengl/gl_graphics_pipeline.h7
-rw-r--r--src/video_core/renderer_vulkan/vk_compute_pass.cpp3
-rw-r--r--src/video_core/renderer_vulkan/vk_texture_cache.cpp3
-rw-r--r--src/video_core/video_core.cpp1
10 files changed, 146 insertions, 98 deletions
diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp
index 5d6d217bb..54a902f56 100644
--- a/src/video_core/engines/maxwell_3d.cpp
+++ b/src/video_core/engines/maxwell_3d.cpp
@@ -7,6 +7,7 @@
#include "common/assert.h"
#include "core/core.h"
#include "core/core_timing.h"
+#include "video_core/dirty_flags.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/gpu.h"
#include "video_core/memory_manager.h"
@@ -195,7 +196,7 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume
case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 13:
case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 14:
case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 15:
- return StartCBData(method);
+ return ProcessCBData(argument);
case MAXWELL3D_REG_INDEX(cb_bind[0]):
return ProcessCBBind(0);
case MAXWELL3D_REG_INDEX(cb_bind[1]):
@@ -208,6 +209,14 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume
return ProcessCBBind(4);
case MAXWELL3D_REG_INDEX(draw.vertex_end_gl):
return DrawArrays();
+ case MAXWELL3D_REG_INDEX(small_index):
+ regs.index_array.count = regs.small_index.count;
+ regs.index_array.first = regs.small_index.first;
+ dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
+ return DrawArrays();
+ case MAXWELL3D_REG_INDEX(topology_override):
+ use_topology_override = true;
+ return;
case MAXWELL3D_REG_INDEX(clear_buffers):
return ProcessClearBuffers();
case MAXWELL3D_REG_INDEX(query.query_get):
@@ -248,14 +257,6 @@ void Maxwell3D::CallMacroMethod(u32 method, const std::vector<u32>& parameters)
}
void Maxwell3D::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
- if (method == cb_data_state.current) {
- regs.reg_array[method] = method_argument;
- ProcessCBData(method_argument);
- return;
- } else if (cb_data_state.current != null_cb_data) {
- FinishCBData();
- }
-
// It is an error to write to a register other than the current macro's ARG register before it
// has finished execution.
if (executing_macro != 0) {
@@ -302,7 +303,7 @@ void Maxwell3D::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 13:
case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 14:
case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 15:
- ProcessCBMultiData(method, base_start, amount);
+ ProcessCBMultiData(base_start, amount);
break;
default:
for (std::size_t i = 0; i < amount; i++) {
@@ -360,6 +361,35 @@ void Maxwell3D::CallMethodFromMME(u32 method, u32 method_argument) {
}
}
+void Maxwell3D::ProcessTopologyOverride() {
+ using PrimitiveTopology = Maxwell3D::Regs::PrimitiveTopology;
+ using PrimitiveTopologyOverride = Maxwell3D::Regs::PrimitiveTopologyOverride;
+
+ PrimitiveTopology topology{};
+
+ switch (regs.topology_override) {
+ case PrimitiveTopologyOverride::None:
+ topology = regs.draw.topology;
+ break;
+ case PrimitiveTopologyOverride::Points:
+ topology = PrimitiveTopology::Points;
+ break;
+ case PrimitiveTopologyOverride::Lines:
+ topology = PrimitiveTopology::Lines;
+ break;
+ case PrimitiveTopologyOverride::LineStrip:
+ topology = PrimitiveTopology::LineStrip;
+ break;
+ default:
+ topology = static_cast<PrimitiveTopology>(regs.topology_override);
+ break;
+ }
+
+ if (use_topology_override) {
+ regs.draw.topology.Assign(topology);
+ }
+}
+
void Maxwell3D::FlushMMEInlineDraw() {
LOG_TRACE(HW_GPU, "called, topology={}, count={}", regs.draw.topology.Value(),
regs.vertex_buffer.count);
@@ -370,6 +400,8 @@ void Maxwell3D::FlushMMEInlineDraw() {
ASSERT_MSG(!regs.draw.instance_next || !regs.draw.instance_cont,
"Illegal combination of instancing parameters");
+ ProcessTopologyOverride();
+
const bool is_indexed = mme_draw.current_mode == MMEDrawMode::Indexed;
if (ShouldExecute()) {
rasterizer->Draw(is_indexed, true);
@@ -529,6 +561,8 @@ void Maxwell3D::DrawArrays() {
ASSERT_MSG(!regs.draw.instance_next || !regs.draw.instance_cont,
"Illegal combination of instancing parameters");
+ ProcessTopologyOverride();
+
if (regs.draw.instance_next) {
// Increment the current instance *before* drawing.
state.current_instance += 1;
@@ -587,46 +621,7 @@ void Maxwell3D::ProcessCBBind(size_t stage_index) {
rasterizer->BindGraphicsUniformBuffer(stage_index, bind_data.index, gpu_addr, size);
}
-void Maxwell3D::ProcessCBData(u32 value) {
- const u32 id = cb_data_state.id;
- cb_data_state.buffer[id][cb_data_state.counter] = value;
- // Increment the current buffer position.
- regs.const_buffer.cb_pos = regs.const_buffer.cb_pos + 4;
- cb_data_state.counter++;
-}
-
-void Maxwell3D::StartCBData(u32 method) {
- constexpr u32 first_cb_data = MAXWELL3D_REG_INDEX(const_buffer.cb_data);
- cb_data_state.start_pos = regs.const_buffer.cb_pos;
- cb_data_state.id = method - first_cb_data;
- cb_data_state.current = method;
- cb_data_state.counter = 0;
- ProcessCBData(regs.const_buffer.cb_data[cb_data_state.id]);
-}
-
-void Maxwell3D::ProcessCBMultiData(u32 method, const u32* start_base, u32 amount) {
- if (cb_data_state.current != method) {
- if (cb_data_state.current != null_cb_data) {
- FinishCBData();
- }
- constexpr u32 first_cb_data = MAXWELL3D_REG_INDEX(const_buffer.cb_data);
- cb_data_state.start_pos = regs.const_buffer.cb_pos;
- cb_data_state.id = method - first_cb_data;
- cb_data_state.current = method;
- cb_data_state.counter = 0;
- }
- const std::size_t id = cb_data_state.id;
- const std::size_t size = amount;
- std::size_t i = 0;
- for (; i < size; i++) {
- cb_data_state.buffer[id][cb_data_state.counter] = start_base[i];
- cb_data_state.counter++;
- }
- // Increment the current buffer position.
- regs.const_buffer.cb_pos = regs.const_buffer.cb_pos + 4 * amount;
-}
-
-void Maxwell3D::FinishCBData() {
+void Maxwell3D::ProcessCBMultiData(const u32* start_base, u32 amount) {
// Write the input value to the current const buffer at the current position.
const GPUVAddr buffer_address = regs.const_buffer.BufferAddress();
ASSERT(buffer_address != 0);
@@ -634,14 +629,16 @@ void Maxwell3D::FinishCBData() {
// Don't allow writing past the end of the buffer.
ASSERT(regs.const_buffer.cb_pos <= regs.const_buffer.cb_size);
- const GPUVAddr address{buffer_address + cb_data_state.start_pos};
- const std::size_t size = regs.const_buffer.cb_pos - cb_data_state.start_pos;
+ const GPUVAddr address{buffer_address + regs.const_buffer.cb_pos};
+ const size_t copy_size = amount * sizeof(u32);
+ memory_manager.WriteBlock(address, start_base, copy_size);
- const u32 id = cb_data_state.id;
- memory_manager.WriteBlock(address, cb_data_state.buffer[id].data(), size);
+ // Increment the current buffer position.
+ regs.const_buffer.cb_pos += static_cast<u32>(copy_size);
+}
- cb_data_state.id = null_cb_data;
- cb_data_state.current = null_cb_data;
+void Maxwell3D::ProcessCBData(u32 value) {
+ ProcessCBMultiData(&value, 1);
}
Texture::TICEntry Maxwell3D::GetTICEntry(u32 tic_index) const {
diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h
index dc9df6c8b..357a74c70 100644
--- a/src/video_core/engines/maxwell_3d.h
+++ b/src/video_core/engines/maxwell_3d.h
@@ -367,6 +367,22 @@ public:
Patches = 0xe,
};
+ // Constants as from NVC0_3D_UNK1970_D3D
+ // https://gitlab.freedesktop.org/mesa/mesa/-/blob/main/src/gallium/drivers/nouveau/nvc0/nvc0_3d.xml.h#L1598
+ enum class PrimitiveTopologyOverride : u32 {
+ None = 0x0,
+ Points = 0x1,
+ Lines = 0x2,
+ LineStrip = 0x3,
+ Triangles = 0x4,
+ TriangleStrip = 0x5,
+ LinesAdjacency = 0xa,
+ LineStripAdjacency = 0xb,
+ TrianglesAdjacency = 0xc,
+ TriangleStripAdjacency = 0xd,
+ Patches = 0xe,
+ };
+
enum class IndexFormat : u32 {
UnsignedByte = 0x0,
UnsignedShort = 0x1,
@@ -1200,7 +1216,12 @@ public:
}
} index_array;
- INSERT_PADDING_WORDS_NOINIT(0x7);
+ union {
+ BitField<0, 16, u32> first;
+ BitField<16, 16, u32> count;
+ } small_index;
+
+ INSERT_PADDING_WORDS_NOINIT(0x6);
INSERT_PADDING_WORDS_NOINIT(0x1F);
@@ -1244,7 +1265,11 @@ public:
BitField<11, 1, u32> depth_clamp_disabled;
} view_volume_clip_control;
- INSERT_PADDING_WORDS_NOINIT(0x1F);
+ INSERT_PADDING_WORDS_NOINIT(0xC);
+
+ PrimitiveTopologyOverride topology_override;
+
+ INSERT_PADDING_WORDS_NOINIT(0x12);
u32 depth_bounds_enable;
@@ -1520,10 +1545,8 @@ private:
void ProcessSyncPoint();
/// Handles a write to the CB_DATA[i] register.
- void StartCBData(u32 method);
void ProcessCBData(u32 value);
- void ProcessCBMultiData(u32 method, const u32* start_base, u32 amount);
- void FinishCBData();
+ void ProcessCBMultiData(const u32* start_base, u32 amount);
/// Handles a write to the CB_BIND register.
void ProcessCBBind(size_t stage_index);
@@ -1531,6 +1554,9 @@ private:
/// Handles a write to the VERTEX_END_GL register, triggering a draw.
void DrawArrays();
+ /// Handles use of topology overrides (e.g., to avoid using a topology assigned from a macro)
+ void ProcessTopologyOverride();
+
// Handles a instance drawcall from MME
void StepInstance(MMEDrawMode expected_mode, u32 count);
@@ -1555,20 +1581,10 @@ private:
/// Interpreter for the macro codes uploaded to the GPU.
std::unique_ptr<MacroEngine> macro_engine;
- static constexpr u32 null_cb_data = 0xFFFFFFFF;
- struct CBDataState {
- static constexpr size_t inline_size = 0x4000;
- std::array<std::array<u32, inline_size>, 16> buffer;
- u32 current{null_cb_data};
- u32 id{null_cb_data};
- u32 start_pos{};
- u32 counter{};
- };
- CBDataState cb_data_state;
-
Upload::State upload_state;
bool execute_on{true};
+ bool use_topology_override{false};
};
#define ASSERT_REG_POSITION(field_name, position) \
@@ -1685,6 +1701,7 @@ ASSERT_REG_POSITION(draw, 0x585);
ASSERT_REG_POSITION(primitive_restart, 0x591);
ASSERT_REG_POSITION(provoking_vertex_last, 0x5A1);
ASSERT_REG_POSITION(index_array, 0x5F2);
+ASSERT_REG_POSITION(small_index, 0x5F9);
ASSERT_REG_POSITION(polygon_offset_clamp, 0x61F);
ASSERT_REG_POSITION(instanced_arrays, 0x620);
ASSERT_REG_POSITION(vp_point_size, 0x644);
@@ -1694,6 +1711,7 @@ ASSERT_REG_POSITION(cull_face, 0x648);
ASSERT_REG_POSITION(pixel_center_integer, 0x649);
ASSERT_REG_POSITION(viewport_transform_enabled, 0x64B);
ASSERT_REG_POSITION(view_volume_clip_control, 0x64F);
+ASSERT_REG_POSITION(topology_override, 0x65C);
ASSERT_REG_POSITION(depth_bounds_enable, 0x66F);
ASSERT_REG_POSITION(logic_op, 0x671);
ASSERT_REG_POSITION(clear_buffers, 0x674);
diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp
index 67388d980..1fc1358bc 100644
--- a/src/video_core/engines/maxwell_dma.cpp
+++ b/src/video_core/engines/maxwell_dma.cpp
@@ -53,7 +53,6 @@ void MaxwellDMA::Launch() {
// TODO(Subv): Perform more research and implement all features of this engine.
const LaunchDMA& launch = regs.launch_dma;
- ASSERT(launch.semaphore_type == LaunchDMA::SemaphoreType::NONE);
ASSERT(launch.interrupt_type == LaunchDMA::InterruptType::NONE);
ASSERT(launch.data_transfer_type == LaunchDMA::DataTransferType::NON_PIPELINED);
ASSERT(regs.dst_params.origin.x == 0);
@@ -79,6 +78,7 @@ void MaxwellDMA::Launch() {
CopyPitchToBlockLinear();
}
}
+ ReleaseSemaphore();
}
void MaxwellDMA::CopyPitchToPitch() {
@@ -244,4 +244,22 @@ void MaxwellDMA::FastCopyBlockLinearToPitch() {
memory_manager.WriteBlock(regs.offset_out, write_buffer.data(), dst_size);
}
+void MaxwellDMA::ReleaseSemaphore() {
+ const auto type = regs.launch_dma.semaphore_type;
+ const GPUVAddr address = regs.semaphore.address;
+ switch (type) {
+ case LaunchDMA::SemaphoreType::NONE:
+ break;
+ case LaunchDMA::SemaphoreType::RELEASE_ONE_WORD_SEMAPHORE:
+ memory_manager.Write<u32>(address, regs.semaphore.payload);
+ break;
+ case LaunchDMA::SemaphoreType::RELEASE_FOUR_WORD_SEMAPHORE:
+ memory_manager.Write<u64>(address, static_cast<u64>(regs.semaphore.payload));
+ memory_manager.Write<u64>(address + 8, system.GPU().GetTicks());
+ break;
+ default:
+ UNREACHABLE_MSG("Unknown semaphore type: {}", static_cast<u32>(type.Value()));
+ }
+}
+
} // namespace Tegra::Engines
diff --git a/src/video_core/engines/maxwell_dma.h b/src/video_core/engines/maxwell_dma.h
index a04514425..2692cac8a 100644
--- a/src/video_core/engines/maxwell_dma.h
+++ b/src/video_core/engines/maxwell_dma.h
@@ -224,6 +224,8 @@ private:
void FastCopyBlockLinearToPitch();
+ void ReleaseSemaphore();
+
Core::System& system;
MemoryManager& memory_manager;
diff --git a/src/video_core/renderer_opengl/gl_fence_manager.cpp b/src/video_core/renderer_opengl/gl_fence_manager.cpp
index 151290101..293ad7d59 100644
--- a/src/video_core/renderer_opengl/gl_fence_manager.cpp
+++ b/src/video_core/renderer_opengl/gl_fence_manager.cpp
@@ -31,9 +31,8 @@ bool GLInnerFence::IsSignaled() const {
return true;
}
ASSERT(sync_object.handle != 0);
- GLsizei length;
GLint sync_status;
- glGetSynciv(sync_object.handle, GL_SYNC_STATUS, sizeof(GLint), &length, &sync_status);
+ glGetSynciv(sync_object.handle, GL_SYNC_STATUS, 1, nullptr, &sync_status);
return sync_status == GL_SIGNALED;
}
diff --git a/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp b/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp
index f8495896c..9e6732abd 100644
--- a/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp
+++ b/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp
@@ -243,10 +243,6 @@ GraphicsPipeline::GraphicsPipeline(
case Settings::ShaderBackend::GLASM:
if (!sources[stage].empty()) {
assembly_programs[stage] = CompileProgram(sources[stage], AssemblyStage(stage));
- if (in_parallel) {
- // Make sure program is built before continuing when building in parallel
- glGetString(GL_PROGRAM_ERROR_STRING_NV);
- }
}
break;
case Settings::ShaderBackend::SPIRV:
@@ -256,20 +252,18 @@ GraphicsPipeline::GraphicsPipeline(
break;
}
}
- if (in_parallel && backend != Settings::ShaderBackend::GLASM) {
- // Make sure programs have built if we are building shaders in parallel
- for (OGLProgram& program : source_programs) {
- if (program.handle != 0) {
- GLint status{};
- glGetProgramiv(program.handle, GL_LINK_STATUS, &status);
- }
- }
+ if (in_parallel) {
+ std::lock_guard lock{built_mutex};
+ built_fence.Create();
+ // Flush this context to ensure compilation commands and fence are in the GPU pipe.
+ glFlush();
+ built_condvar.notify_one();
+ } else {
+ is_built = true;
}
if (shader_notify) {
shader_notify->MarkShaderComplete();
}
- is_built = true;
- built_condvar.notify_one();
}};
if (thread_worker) {
thread_worker->QueueWork(std::move(func));
@@ -440,7 +434,7 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) {
buffer_cache.UpdateGraphicsBuffers(is_indexed);
buffer_cache.BindHostGeometryBuffers(is_indexed);
- if (!is_built.load(std::memory_order::relaxed)) {
+ if (!IsBuilt()) {
WaitForBuild();
}
const bool use_assembly{assembly_programs[0].handle != 0};
@@ -585,8 +579,26 @@ void GraphicsPipeline::GenerateTransformFeedbackState() {
}
void GraphicsPipeline::WaitForBuild() {
- std::unique_lock lock{built_mutex};
- built_condvar.wait(lock, [this] { return is_built.load(std::memory_order::relaxed); });
+ if (built_fence.handle == 0) {
+ std::unique_lock lock{built_mutex};
+ built_condvar.wait(lock, [this] { return built_fence.handle != 0; });
+ }
+ ASSERT(glClientWaitSync(built_fence.handle, 0, GL_TIMEOUT_IGNORED) != GL_WAIT_FAILED);
+ is_built = true;
+}
+
+bool GraphicsPipeline::IsBuilt() noexcept {
+ if (is_built) {
+ return true;
+ }
+ if (built_fence.handle == 0) {
+ return false;
+ }
+ // Timeout of zero means this is non-blocking
+ const auto sync_status = glClientWaitSync(built_fence.handle, 0, 0);
+ ASSERT(sync_status != GL_WAIT_FAILED);
+ is_built = sync_status != GL_TIMEOUT_EXPIRED;
+ return is_built;
}
} // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_graphics_pipeline.h b/src/video_core/renderer_opengl/gl_graphics_pipeline.h
index 4e28d9a42..311d49f3f 100644
--- a/src/video_core/renderer_opengl/gl_graphics_pipeline.h
+++ b/src/video_core/renderer_opengl/gl_graphics_pipeline.h
@@ -100,9 +100,7 @@ public:
return writes_global_memory;
}
- [[nodiscard]] bool IsBuilt() const noexcept {
- return is_built.load(std::memory_order::relaxed);
- }
+ [[nodiscard]] bool IsBuilt() noexcept;
template <typename Spec>
static auto MakeConfigureSpecFunc() {
@@ -154,7 +152,8 @@ private:
std::mutex built_mutex;
std::condition_variable built_condvar;
- std::atomic_bool is_built{false};
+ OGLSync built_fence{};
+ bool is_built{false};
};
} // namespace OpenGL
diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp
index 3e96c0f60..4d73427b4 100644
--- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp
+++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp
@@ -2,6 +2,7 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
+#include <array>
#include <cstring>
#include <memory>
#include <optional>
@@ -292,7 +293,7 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble(
.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
.dstAccessMask = VK_ACCESS_INDEX_READ_BIT,
};
- const std::array push_constants{base_vertex, index_shift};
+ const std::array<u32, 2> push_constants{base_vertex, index_shift};
const VkDescriptorSet set = descriptor_allocator.Commit();
device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline);
diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
index 0f62779de..ca6019a3a 100644
--- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
@@ -1067,7 +1067,8 @@ void TextureCacheRuntime::ConvertImage(Framebuffer* dst, ImageView& dst_view, Im
}
break;
case PixelFormat::A8B8G8R8_UNORM:
- if (src_view.format == PixelFormat::S8_UINT_D24_UNORM) {
+ if (src_view.format == PixelFormat::S8_UINT_D24_UNORM ||
+ src_view.format == PixelFormat::D24_UNORM_S8_UINT) {
return blit_image_helper.ConvertD24S8ToABGR8(dst, src_view);
}
break;
diff --git a/src/video_core/video_core.cpp b/src/video_core/video_core.cpp
index 329bf4def..2f2594585 100644
--- a/src/video_core/video_core.cpp
+++ b/src/video_core/video_core.cpp
@@ -50,6 +50,7 @@ std::unique_ptr<Tegra::GPU> CreateGPU(Core::Frontend::EmuWindow& emu_window, Cor
gpu->BindRenderer(std::move(renderer));
return gpu;
} catch (const std::runtime_error& exception) {
+ scope.Cancel();
LOG_ERROR(HW_GPU, "Failed to initialize GPU: {}", exception.what());
return nullptr;
}
generated by cgit v1.2.3 (git 2.25.1) at 2025-12-29 20:52:45 +0000