summaryrefslogtreecommitdiffstats
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.h3
-rw-r--r--src/video_core/engines/shader_bytecode.h3
-rw-r--r--src/video_core/memory_manager.cpp18
-rw-r--r--src/video_core/renderer_opengl/gl_rasterizer.cpp12
-rw-r--r--src/video_core/renderer_vulkan/fixed_pipeline_state.h2
-rw-r--r--src/video_core/renderer_vulkan/vk_rasterizer.cpp42
-rw-r--r--src/video_core/renderer_vulkan/vk_rasterizer.h6
-rw-r--r--src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp2
-rw-r--r--src/video_core/renderer_vulkan/wrapper.cpp21
-rw-r--r--src/video_core/shader/decode/arithmetic_half.cpp51
-rw-r--r--src/video_core/shader/decode/arithmetic_integer.cpp4
11 files changed, 119 insertions, 45 deletions
diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h
index cfcda4f53..3dfba8197 100644
--- a/src/video_core/engines/maxwell_3d.h
+++ b/src/video_core/engines/maxwell_3d.h
@@ -1259,7 +1259,8 @@ public:
GPUVAddr LimitAddress() const {
return static_cast<GPUVAddr>((static_cast<GPUVAddr>(limit_high) << 32) |
- limit_low);
+ limit_low) +
+ 1;
}
} vertex_array_limit[NumVertexArrays];
diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h
index 7231597d4..cde3a26b9 100644
--- a/src/video_core/engines/shader_bytecode.h
+++ b/src/video_core/engines/shader_bytecode.h
@@ -655,6 +655,7 @@ union Instruction {
}
constexpr Instruction(u64 value) : value{value} {}
+ constexpr Instruction(const Instruction& instr) : value(instr.value) {}
BitField<0, 8, Register> gpr0;
BitField<8, 8, Register> gpr8;
@@ -817,11 +818,9 @@ union Instruction {
BitField<32, 1, u64> saturate;
BitField<49, 2, HalfMerge> merge;
- BitField<43, 1, u64> negate_a;
BitField<44, 1, u64> abs_a;
BitField<47, 2, HalfType> type_a;
- BitField<31, 1, u64> negate_b;
BitField<30, 1, u64> abs_b;
BitField<28, 2, HalfType> type_b;
diff --git a/src/video_core/memory_manager.cpp b/src/video_core/memory_manager.cpp
index fd49bc2a9..dbee9f634 100644
--- a/src/video_core/memory_manager.cpp
+++ b/src/video_core/memory_manager.cpp
@@ -51,11 +51,8 @@ GPUVAddr MemoryManager::MapBufferEx(VAddr cpu_addr, u64 size) {
const GPUVAddr gpu_addr{FindFreeRegion(address_space_base, aligned_size)};
MapBackingMemory(gpu_addr, system.Memory().GetPointer(cpu_addr), aligned_size, cpu_addr);
- ASSERT(system.CurrentProcess()
- ->PageTable()
- .SetMemoryAttribute(cpu_addr, size, Kernel::Memory::MemoryAttribute::DeviceShared,
- Kernel::Memory::MemoryAttribute::DeviceShared)
- .IsSuccess());
+ ASSERT(
+ system.CurrentProcess()->PageTable().LockForDeviceAddressSpace(cpu_addr, size).IsSuccess());
return gpu_addr;
}
@@ -66,11 +63,8 @@ GPUVAddr MemoryManager::MapBufferEx(VAddr cpu_addr, GPUVAddr gpu_addr, u64 size)
const u64 aligned_size{Common::AlignUp(size, page_size)};
MapBackingMemory(gpu_addr, system.Memory().GetPointer(cpu_addr), aligned_size, cpu_addr);
- ASSERT(system.CurrentProcess()
- ->PageTable()
- .SetMemoryAttribute(cpu_addr, size, Kernel::Memory::MemoryAttribute::DeviceShared,
- Kernel::Memory::MemoryAttribute::DeviceShared)
- .IsSuccess());
+ ASSERT(
+ system.CurrentProcess()->PageTable().LockForDeviceAddressSpace(cpu_addr, size).IsSuccess());
return gpu_addr;
}
@@ -87,9 +81,7 @@ GPUVAddr MemoryManager::UnmapBuffer(GPUVAddr gpu_addr, u64 size) {
UnmapRange(gpu_addr, aligned_size);
ASSERT(system.CurrentProcess()
->PageTable()
- .SetMemoryAttribute(cpu_addr.value(), size,
- Kernel::Memory::MemoryAttribute::DeviceShared,
- Kernel::Memory::MemoryAttribute::None)
+ .UnlockForDeviceAddressSpace(cpu_addr.value(), size)
.IsSuccess());
return gpu_addr;
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index 4c16c89d2..6fe155bcc 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -186,8 +186,12 @@ void RasterizerOpenGL::SetupVertexBuffer() {
const GPUVAddr start = vertex_array.StartAddress();
const GPUVAddr end = regs.vertex_array_limit[index].LimitAddress();
- ASSERT(end > start);
- const u64 size = end - start + 1;
+ ASSERT(end >= start);
+ const u64 size = end - start;
+ if (size == 0) {
+ glBindVertexBuffer(static_cast<GLuint>(index), 0, 0, vertex_array.stride);
+ continue;
+ }
const auto [vertex_buffer, vertex_buffer_offset] = buffer_cache.UploadMemory(start, size);
glBindVertexBuffer(static_cast<GLuint>(index), vertex_buffer, vertex_buffer_offset,
vertex_array.stride);
@@ -311,8 +315,8 @@ std::size_t RasterizerOpenGL::CalculateVertexArraysSize() const {
const GPUVAddr start = regs.vertex_array[index].StartAddress();
const GPUVAddr end = regs.vertex_array_limit[index].LimitAddress();
- ASSERT(end > start);
- size += end - start + 1;
+ size += end - start;
+ ASSERT(end >= start);
}
return size;
diff --git a/src/video_core/renderer_vulkan/fixed_pipeline_state.h b/src/video_core/renderer_vulkan/fixed_pipeline_state.h
index 9fe6bdbf9..9a950f4de 100644
--- a/src/video_core/renderer_vulkan/fixed_pipeline_state.h
+++ b/src/video_core/renderer_vulkan/fixed_pipeline_state.h
@@ -129,7 +129,7 @@ struct alignas(32) FixedPipelineState {
auto& binding = bindings[index];
binding.raw = 0;
binding.enabled.Assign(enabled ? 1 : 0);
- binding.stride.Assign(stride);
+ binding.stride.Assign(static_cast<u16>(stride));
binding_divisors[index] = divisor;
}
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
index 8a1f57891..68464e637 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
@@ -877,8 +877,12 @@ void RasterizerVulkan::SetupVertexArrays(FixedPipelineState::VertexInput& vertex
const GPUVAddr start{vertex_array.StartAddress()};
const GPUVAddr end{regs.vertex_array_limit[index].LimitAddress()};
- ASSERT(end > start);
- const std::size_t size{end - start + 1};
+ ASSERT(end >= start);
+ const std::size_t size{end - start};
+ if (size == 0) {
+ buffer_bindings.AddVertexBinding(DefaultBuffer(), 0);
+ continue;
+ }
const auto [buffer, offset] = buffer_cache.UploadMemory(start, size);
buffer_bindings.AddVertexBinding(buffer, offset);
}
@@ -1033,8 +1037,7 @@ void RasterizerVulkan::SetupConstBuffer(const ConstBufferEntry& entry,
const Tegra::Engines::ConstBufferInfo& buffer) {
if (!buffer.enabled) {
// Set values to zero to unbind buffers
- update_descriptor_queue.AddBuffer(buffer_cache.GetEmptyBuffer(sizeof(float)), 0,
- sizeof(float));
+ update_descriptor_queue.AddBuffer(DefaultBuffer(), 0, DEFAULT_BUFFER_SIZE);
return;
}
@@ -1057,7 +1060,9 @@ void RasterizerVulkan::SetupGlobalBuffer(const GlobalBufferEntry& entry, GPUVAdd
if (size == 0) {
// Sometimes global memory pointers don't have a proper size. Upload a dummy entry
// because Vulkan doesn't like empty buffers.
- constexpr std::size_t dummy_size = 4;
+ // Note: Do *not* use DefaultBuffer() here, storage buffers can be written breaking the
+ // default buffer.
+ static constexpr std::size_t dummy_size = 4;
const auto buffer = buffer_cache.GetEmptyBuffer(dummy_size);
update_descriptor_queue.AddBuffer(buffer, 0, dummy_size);
return;
@@ -1222,7 +1227,7 @@ std::size_t RasterizerVulkan::CalculateVertexArraysSize() const {
const GPUVAddr end{regs.vertex_array_limit[index].LimitAddress()};
DEBUG_ASSERT(end >= start);
- size += (end - start + 1) * regs.vertex_array[index].enable;
+ size += (end - start) * regs.vertex_array[index].enable;
}
return size;
}
@@ -1269,4 +1274,29 @@ RenderPassParams RasterizerVulkan::GetRenderPassParams(Texceptions texceptions)
return renderpass_params;
}
+VkBuffer RasterizerVulkan::DefaultBuffer() {
+ if (default_buffer) {
+ return *default_buffer;
+ }
+
+ VkBufferCreateInfo ci;
+ ci.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
+ ci.pNext = nullptr;
+ ci.flags = 0;
+ ci.size = DEFAULT_BUFFER_SIZE;
+ ci.usage = VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_VERTEX_BUFFER_BIT |
+ VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT;
+ ci.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
+ ci.queueFamilyIndexCount = 0;
+ ci.pQueueFamilyIndices = nullptr;
+ default_buffer = device.GetLogical().CreateBuffer(ci);
+ default_buffer_commit = memory_manager.Commit(default_buffer, false);
+
+ scheduler.RequestOutsideRenderPassOperationContext();
+ scheduler.Record([buffer = *default_buffer](vk::CommandBuffer cmdbuf) {
+ cmdbuf.FillBuffer(buffer, 0, DEFAULT_BUFFER_SIZE, 0);
+ });
+ return *default_buffer;
+}
+
} // namespace Vulkan
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h
index 2fa46b0cc..d41a7929e 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.h
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.h
@@ -155,6 +155,7 @@ private:
using Texceptions = std::bitset<Maxwell::NumRenderTargets + 1>;
static constexpr std::size_t ZETA_TEXCEPTION_INDEX = 8;
+ static constexpr VkDeviceSize DEFAULT_BUFFER_SIZE = 4 * sizeof(float);
void FlushWork();
@@ -247,6 +248,8 @@ private:
RenderPassParams GetRenderPassParams(Texceptions texceptions) const;
+ VkBuffer DefaultBuffer();
+
Core::System& system;
Core::Frontend::EmuWindow& render_window;
VKScreenInfo& screen_info;
@@ -271,6 +274,9 @@ private:
VKFenceManager fence_manager;
VKQueryCache query_cache;
+ vk::Buffer default_buffer;
+ VKMemoryCommit default_buffer_commit;
+
std::array<View, Maxwell::NumRenderTargets> color_attachments;
View zeta_attachment;
diff --git a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp
index 94d954d7a..c76ab5c2d 100644
--- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp
+++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp
@@ -81,7 +81,7 @@ VKBuffer& VKStagingBufferPool::CreateStagingBuffer(std::size_t size, bool host_v
ci.size = 1ULL << log2;
ci.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT |
VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT |
- VK_BUFFER_USAGE_INDEX_BUFFER_BIT;
+ VK_BUFFER_USAGE_INDEX_BUFFER_BIT | VK_BUFFER_USAGE_VERTEX_BUFFER_BIT;
ci.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
ci.queueFamilyIndexCount = 0;
ci.pQueueFamilyIndices = nullptr;
diff --git a/src/video_core/renderer_vulkan/wrapper.cpp b/src/video_core/renderer_vulkan/wrapper.cpp
index 539f3c974..7f5bc1404 100644
--- a/src/video_core/renderer_vulkan/wrapper.cpp
+++ b/src/video_core/renderer_vulkan/wrapper.cpp
@@ -2,6 +2,7 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
+#include <algorithm>
#include <exception>
#include <memory>
#include <optional>
@@ -16,6 +17,23 @@ namespace Vulkan::vk {
namespace {
+void SortPhysicalDevices(std::vector<VkPhysicalDevice>& devices, const InstanceDispatch& dld) {
+ std::stable_sort(devices.begin(), devices.end(), [&](auto lhs, auto rhs) {
+ // This will call Vulkan more than needed, but these calls are cheap.
+ const auto lhs_properties = vk::PhysicalDevice(lhs, dld).GetProperties();
+ const auto rhs_properties = vk::PhysicalDevice(rhs, dld).GetProperties();
+
+ // Prefer discrete GPUs, Nvidia over AMD, AMD over Intel, Intel over the rest.
+ const bool preferred =
+ (lhs_properties.deviceType == VK_PHYSICAL_DEVICE_TYPE_DISCRETE_GPU &&
+ rhs_properties.deviceType != VK_PHYSICAL_DEVICE_TYPE_DISCRETE_GPU) ||
+ (lhs_properties.vendorID == 0x10DE && rhs_properties.vendorID != 0x10DE) ||
+ (lhs_properties.vendorID == 0x1002 && rhs_properties.vendorID != 0x1002) ||
+ (lhs_properties.vendorID == 0x8086 && rhs_properties.vendorID != 0x8086);
+ return !preferred;
+ });
+}
+
template <typename T>
bool Proc(T& result, const InstanceDispatch& dld, const char* proc_name,
VkInstance instance = nullptr) noexcept {
@@ -389,7 +407,8 @@ std::optional<std::vector<VkPhysicalDevice>> Instance::EnumeratePhysicalDevices(
if (dld->vkEnumeratePhysicalDevices(handle, &num, physical_devices.data()) != VK_SUCCESS) {
return std::nullopt;
}
- return physical_devices;
+ SortPhysicalDevices(physical_devices, *dld);
+ return std::make_optional(std::move(physical_devices));
}
DebugCallback Instance::TryCreateDebugCallback(
diff --git a/src/video_core/shader/decode/arithmetic_half.cpp b/src/video_core/shader/decode/arithmetic_half.cpp
index ee7d9a29d..a276aee44 100644
--- a/src/video_core/shader/decode/arithmetic_half.cpp
+++ b/src/video_core/shader/decode/arithmetic_half.cpp
@@ -19,22 +19,46 @@ u32 ShaderIR::DecodeArithmeticHalf(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
- if (opcode->get().GetId() == OpCode::Id::HADD2_C ||
- opcode->get().GetId() == OpCode::Id::HADD2_R) {
+ bool negate_a = false;
+ bool negate_b = false;
+ bool absolute_a = false;
+ bool absolute_b = false;
+
+ switch (opcode->get().GetId()) {
+ case OpCode::Id::HADD2_R:
if (instr.alu_half.ftz == 0) {
LOG_DEBUG(HW_GPU, "{} without FTZ is not implemented", opcode->get().GetName());
}
+ negate_a = ((instr.value >> 43) & 1) != 0;
+ negate_b = ((instr.value >> 31) & 1) != 0;
+ absolute_a = ((instr.value >> 44) & 1) != 0;
+ absolute_b = ((instr.value >> 30) & 1) != 0;
+ break;
+ case OpCode::Id::HADD2_C:
+ if (instr.alu_half.ftz == 0) {
+ LOG_DEBUG(HW_GPU, "{} without FTZ is not implemented", opcode->get().GetName());
+ }
+ negate_a = ((instr.value >> 43) & 1) != 0;
+ negate_b = ((instr.value >> 56) & 1) != 0;
+ absolute_a = ((instr.value >> 44) & 1) != 0;
+ absolute_b = ((instr.value >> 54) & 1) != 0;
+ break;
+ case OpCode::Id::HMUL2_R:
+ negate_a = ((instr.value >> 43) & 1) != 0;
+ absolute_a = ((instr.value >> 44) & 1) != 0;
+ absolute_b = ((instr.value >> 30) & 1) != 0;
+ break;
+ case OpCode::Id::HMUL2_C:
+ negate_b = ((instr.value >> 31) & 1) != 0;
+ absolute_a = ((instr.value >> 44) & 1) != 0;
+ absolute_b = ((instr.value >> 54) & 1) != 0;
+ break;
}
- const bool negate_a =
- opcode->get().GetId() != OpCode::Id::HMUL2_R && instr.alu_half.negate_a != 0;
- const bool negate_b =
- opcode->get().GetId() != OpCode::Id::HMUL2_C && instr.alu_half.negate_b != 0;
-
Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.alu_half.type_a);
- op_a = GetOperandAbsNegHalf(op_a, instr.alu_half.abs_a, negate_a);
+ op_a = GetOperandAbsNegHalf(op_a, absolute_a, negate_a);
- auto [type_b, op_b] = [&]() -> std::tuple<HalfType, Node> {
+ auto [type_b, op_b] = [this, instr, opcode]() -> std::pair<HalfType, Node> {
switch (opcode->get().GetId()) {
case OpCode::Id::HADD2_C:
case OpCode::Id::HMUL2_C:
@@ -48,17 +72,16 @@ u32 ShaderIR::DecodeArithmeticHalf(NodeBlock& bb, u32 pc) {
}
}();
op_b = UnpackHalfFloat(op_b, type_b);
- // redeclaration to avoid a bug in clang with reusing local bindings in lambdas
- Node op_b_alt = GetOperandAbsNegHalf(op_b, instr.alu_half.abs_b, negate_b);
+ op_b = GetOperandAbsNegHalf(op_b, absolute_b, negate_b);
- Node value = [&]() {
+ Node value = [this, opcode, op_a, op_b = op_b] {
switch (opcode->get().GetId()) {
case OpCode::Id::HADD2_C:
case OpCode::Id::HADD2_R:
- return Operation(OperationCode::HAdd, PRECISE, op_a, op_b_alt);
+ return Operation(OperationCode::HAdd, PRECISE, op_a, op_b);
case OpCode::Id::HMUL2_C:
case OpCode::Id::HMUL2_R:
- return Operation(OperationCode::HMul, PRECISE, op_a, op_b_alt);
+ return Operation(OperationCode::HMul, PRECISE, op_a, op_b);
default:
UNIMPLEMENTED_MSG("Unhandled half float instruction: {}", opcode->get().GetName());
return Immediate(0);
diff --git a/src/video_core/shader/decode/arithmetic_integer.cpp b/src/video_core/shader/decode/arithmetic_integer.cpp
index 0f4c3103a..9af8c606d 100644
--- a/src/video_core/shader/decode/arithmetic_integer.cpp
+++ b/src/video_core/shader/decode/arithmetic_integer.cpp
@@ -249,8 +249,8 @@ u32 ShaderIR::DecodeArithmeticInteger(NodeBlock& bb, u32 pc) {
}
case OpCode::Id::LEA_IMM: {
const bool neg = instr.lea.imm.neg != 0;
- return {Immediate(static_cast<u32>(instr.lea.imm.entry_a)),
- GetOperandAbsNegInteger(GetRegister(instr.gpr8), false, neg, true),
+ return {GetOperandAbsNegInteger(GetRegister(instr.gpr8), false, neg, true),
+ Immediate(static_cast<u32>(instr.lea.imm.entry_a)),
Immediate(static_cast<u32>(instr.lea.imm.entry_b))};
}
case OpCode::Id::LEA_RZ: {
generated by cgit v1.2.3 (git 2.25.1) at 2024-07-16 17:17:27 +0200