diff --git a/README.md b/README.md index eb09f1d15..03e5221d5 100755 --- a/README.md +++ b/README.md @@ -1,7 +1,7 @@ yuzu emulator early access ============= -This is the source code for early-access 1871. +This is the source code for early-access 1873. ## Legal Notice diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/common_funcs.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/common_funcs.cpp index 10bb01d99..20458d2ad 100755 --- a/src/shader_recompiler/frontend/maxwell/translate/impl/common_funcs.cpp +++ b/src/shader_recompiler/frontend/maxwell/translate/impl/common_funcs.cpp @@ -29,6 +29,49 @@ IR::U1 IntegerCompare(IR::IREmitter& ir, const IR::U32& operand_1, const IR::U32 } } +IR::U1 ExtendedIntegerCompare(IR::IREmitter& ir, const IR::U32& operand_1, const IR::U32& operand_2, + CompareOp compare_op, bool is_signed) { + const IR::U32 zero{ir.Imm32(0)}; + const IR::U32 carry{ir.Select(ir.GetCFlag(), ir.Imm32(1), zero)}; + const IR::U1 z_flag{ir.GetZFlag()}; + const IR::U32 intermediate{ir.IAdd(ir.IAdd(operand_1, ir.BitwiseNot(operand_2)), carry)}; + const IR::U1 flip_logic{is_signed ? ir.Imm1(false) + : ir.LogicalXor(ir.ILessThan(operand_1, zero, true), + ir.ILessThan(operand_2, zero, true))}; + switch (compare_op) { + case CompareOp::False: + return ir.Imm1(false); + case CompareOp::LessThan: + return IR::U1{ir.Select(flip_logic, ir.IGreaterThanEqual(intermediate, zero, true), + ir.ILessThan(intermediate, zero, true))}; + case CompareOp::Equal: + return ir.LogicalAnd(ir.IEqual(intermediate, zero), z_flag); + case CompareOp::LessThanEqual: { + const IR::U1 base_cmp{ir.Select(flip_logic, ir.IGreaterThanEqual(intermediate, zero, true), + ir.ILessThan(intermediate, zero, true))}; + return ir.LogicalOr(base_cmp, ir.LogicalAnd(ir.IEqual(intermediate, zero), z_flag)); + } + case CompareOp::GreaterThan: { + const IR::U1 base_cmp{ir.Select(flip_logic, ir.ILessThanEqual(intermediate, zero, true), + ir.IGreaterThan(intermediate, zero, true))}; + const IR::U1 not_z{ir.LogicalNot(z_flag)}; + return ir.LogicalOr(base_cmp, ir.LogicalAnd(ir.IEqual(intermediate, zero), not_z)); + } + case CompareOp::NotEqual: + return ir.LogicalOr(ir.INotEqual(intermediate, zero), + ir.LogicalAnd(ir.IEqual(intermediate, zero), ir.LogicalNot(z_flag))); + case CompareOp::GreaterThanEqual: { + const IR::U1 base_cmp{ir.Select(flip_logic, ir.ILessThan(intermediate, zero, true), + ir.IGreaterThanEqual(intermediate, zero, true))}; + return ir.LogicalOr(base_cmp, ir.LogicalAnd(ir.IEqual(intermediate, zero), z_flag)); + } + case CompareOp::True: + return ir.Imm1(true); + default: + throw NotImplementedException("Invalid compare op {}", compare_op); + } +} + IR::U1 PredicateCombine(IR::IREmitter& ir, const IR::U1& predicate_1, const IR::U1& predicate_2, BooleanOp bop) { switch (bop) { diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/common_funcs.h b/src/shader_recompiler/frontend/maxwell/translate/impl/common_funcs.h index f584060b3..214d0af3c 100755 --- a/src/shader_recompiler/frontend/maxwell/translate/impl/common_funcs.h +++ b/src/shader_recompiler/frontend/maxwell/translate/impl/common_funcs.h @@ -11,6 +11,10 @@ namespace Shader::Maxwell { [[nodiscard]] IR::U1 IntegerCompare(IR::IREmitter& ir, const IR::U32& operand_1, const IR::U32& operand_2, CompareOp compare_op, bool is_signed); +[[nodiscard]] IR::U1 ExtendedIntegerCompare(IR::IREmitter& ir, const IR::U32& operand_1, + const IR::U32& operand_2, CompareOp compare_op, + bool is_signed); + [[nodiscard]] IR::U1 PredicateCombine(IR::IREmitter& ir, const IR::U1& predicate_1, const IR::U1& predicate_2, BooleanOp bop); diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/integer_compare_and_set.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/integer_compare_and_set.cpp index 34fa7345c..8ce1aee04 100755 --- a/src/shader_recompiler/frontend/maxwell/translate/impl/integer_compare_and_set.cpp +++ b/src/shader_recompiler/frontend/maxwell/translate/impl/integer_compare_and_set.cpp @@ -9,49 +9,6 @@ namespace Shader::Maxwell { namespace { -IR::U1 ExtendedIntegerCompare(IR::IREmitter& ir, const IR::U32& operand_1, const IR::U32& operand_2, - CompareOp compare_op, bool is_signed) { - const IR::U32 zero{ir.Imm32(0)}; - const IR::U32 carry{ir.Select(ir.GetCFlag(), ir.Imm32(1), zero)}; - const IR::U1 z_flag{ir.GetZFlag()}; - const IR::U32 intermediate{ir.IAdd(ir.IAdd(operand_1, ir.BitwiseNot(operand_2)), carry)}; - const IR::U1 flip_logic{is_signed ? ir.Imm1(false) - : ir.LogicalXor(ir.ILessThan(operand_1, zero, true), - ir.ILessThan(operand_2, zero, true))}; - switch (compare_op) { - case CompareOp::False: - return ir.Imm1(false); - case CompareOp::LessThan: - return IR::U1{ir.Select(flip_logic, ir.IGreaterThanEqual(intermediate, zero, true), - ir.ILessThan(intermediate, zero, true))}; - case CompareOp::Equal: - return ir.LogicalAnd(ir.IEqual(intermediate, zero), z_flag); - case CompareOp::LessThanEqual: { - const IR::U1 base_cmp{ir.Select(flip_logic, ir.IGreaterThanEqual(intermediate, zero, true), - ir.ILessThan(intermediate, zero, true))}; - return ir.LogicalOr(base_cmp, ir.LogicalAnd(ir.IEqual(intermediate, zero), z_flag)); - } - case CompareOp::GreaterThan: { - const IR::U1 base_cmp{ir.Select(flip_logic, ir.ILessThanEqual(intermediate, zero, true), - ir.IGreaterThan(intermediate, zero, true))}; - const IR::U1 not_z{ir.LogicalNot(z_flag)}; - return ir.LogicalOr(base_cmp, ir.LogicalAnd(ir.IEqual(intermediate, zero), not_z)); - } - case CompareOp::NotEqual: - return ir.LogicalOr(ir.INotEqual(intermediate, zero), - ir.LogicalAnd(ir.IEqual(intermediate, zero), ir.LogicalNot(z_flag))); - case CompareOp::GreaterThanEqual: { - const IR::U1 base_cmp{ir.Select(flip_logic, ir.ILessThan(intermediate, zero, true), - ir.IGreaterThanEqual(intermediate, zero, true))}; - return ir.LogicalOr(base_cmp, ir.LogicalAnd(ir.IEqual(intermediate, zero), z_flag)); - } - case CompareOp::True: - return ir.Imm1(true); - default: - throw NotImplementedException("Invalid compare op {}", compare_op); - } -} - IR::U1 IsetCompare(IR::IREmitter& ir, const IR::U32& operand_1, const IR::U32& operand_2, CompareOp compare_op, bool is_signed, bool x) { return x ? ExtendedIntegerCompare(ir, operand_1, operand_2, compare_op, is_signed) diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/integer_set_predicate.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/integer_set_predicate.cpp index 7743701d0..bee10e5b9 100755 --- a/src/shader_recompiler/frontend/maxwell/translate/impl/integer_set_predicate.cpp +++ b/src/shader_recompiler/frontend/maxwell/translate/impl/integer_set_predicate.cpp @@ -9,6 +9,12 @@ namespace Shader::Maxwell { namespace { +IR::U1 IsetpCompare(IR::IREmitter& ir, const IR::U32& operand_1, const IR::U32& operand_2, + CompareOp compare_op, bool is_signed, bool x) { + return x ? ExtendedIntegerCompare(ir, operand_1, operand_2, compare_op, is_signed) + : IntegerCompare(ir, operand_1, operand_2, compare_op, is_signed); +} + void ISETP(TranslatorVisitor& v, u64 insn, const IR::U32& op_b) { union { u64 raw; @@ -17,15 +23,18 @@ void ISETP(TranslatorVisitor& v, u64 insn, const IR::U32& op_b) { BitField<8, 8, IR::Reg> src_reg_a; BitField<39, 3, IR::Pred> bop_pred; BitField<42, 1, u64> neg_bop_pred; + BitField<43, 1, u64> x; BitField<45, 2, BooleanOp> bop; BitField<48, 1, u64> is_signed; BitField<49, 3, CompareOp> compare_op; } const isetp{insn}; + const bool is_signed{isetp.is_signed != 0}; + const bool x{isetp.x != 0}; const BooleanOp bop{isetp.bop}; const CompareOp compare_op{isetp.compare_op}; const IR::U32 op_a{v.X(isetp.src_reg_a)}; - const IR::U1 comparison{IntegerCompare(v.ir, op_a, op_b, compare_op, isetp.is_signed != 0)}; + const IR::U1 comparison{IsetpCompare(v.ir, op_a, op_b, compare_op, is_signed, x)}; const IR::U1 bop_pred{v.ir.GetPred(isetp.bop_pred, isetp.neg_bop_pred != 0)}; const IR::U1 result_a{PredicateCombine(v.ir, comparison, bop_pred, bop)}; const IR::U1 result_b{PredicateCombine(v.ir, v.ir.LogicalNot(comparison), bop_pred, bop)}; diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index 798b327d2..79c0bea0e 100755 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -186,11 +186,16 @@ public: /// Pop asynchronous downloads void PopAsyncFlushes(); - [[nodiscard]] bool DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount); + bool DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount); + + bool DMAClear(GPUVAddr src_address, u64 amount, u32 value); /// Return true when a CPU region is modified from the GPU [[nodiscard]] bool IsRegionGpuModified(VAddr addr, size_t size); + /// Return true when a region is registered on the cache + [[nodiscard]] bool IsRegionRegistered(VAddr addr, size_t size); + /// Return true when a CPU region is modified from the CPU [[nodiscard]] bool IsRegionCpuModified(VAddr addr, size_t size); @@ -516,8 +521,8 @@ bool BufferCache

::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am if (!cpu_src_address || !cpu_dest_address) { return false; } - const bool source_dirty = IsRegionGpuModified(*cpu_src_address, amount); - const bool dest_dirty = IsRegionGpuModified(*cpu_dest_address, amount); + const bool source_dirty = IsRegionRegistered(*cpu_src_address, amount); + const bool dest_dirty = IsRegionRegistered(*cpu_dest_address, amount); if (!source_dirty && !dest_dirty) { return false; } @@ -562,7 +567,7 @@ bool BufferCache

::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am } runtime.CopyBuffer(dest_buffer, src_buffer, copies); - if (source_dirty) { + if (IsRegionGpuModified(*cpu_src_address, amount)) { dest_buffer.MarkRegionAsGpuModified(*cpu_dest_address, amount); } std::vector tmp_buffer(amount); @@ -571,6 +576,37 @@ bool BufferCache

::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am return true; } +template +bool BufferCache

::DMAClear(GPUVAddr dst_address, u64 amount, u32 value) { + const std::optional cpu_dst_address = gpu_memory.GpuToCpuAddress(dst_address); + if (!cpu_dst_address) { + return false; + } + const bool dest_dirty = IsRegionRegistered(*cpu_dst_address, amount); + if (!dest_dirty) { + return false; + } + + const IntervalType subtract_interval{*cpu_dst_address, *cpu_dst_address + amount * sizeof(u32)}; + uncommitted_ranges.subtract(subtract_interval); + for (auto& interval_set : committed_ranges) { + interval_set.subtract(subtract_interval); + } + common_ranges.subtract(subtract_interval); + + const size_t size = amount * sizeof(u32); + BufferId buffer; + do { + has_deleted_buffers = false; + buffer = FindBuffer(*cpu_dst_address, static_cast(size)); + } while (has_deleted_buffers); + + auto& dest_buffer = slot_buffers[buffer]; + const u32 offset = static_cast(*cpu_dst_address - dest_buffer.CpuAddr()); + runtime.ClearBuffer(dest_buffer, offset, size, value); + return true; +} + template void BufferCache

::BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size) { @@ -876,6 +912,27 @@ bool BufferCache

::IsRegionGpuModified(VAddr addr, size_t size) { return false; } +template +bool BufferCache

::IsRegionRegistered(VAddr addr, size_t size) { + const VAddr end_addr = addr + size; + const u64 page_end = Common::DivCeil(end_addr, PAGE_SIZE); + for (u64 page = addr >> PAGE_BITS; page < page_end;) { + const BufferId buffer_id = page_table[page]; + if (!buffer_id) { + ++page; + continue; + } + Buffer& buffer = slot_buffers[buffer_id]; + const VAddr buf_start_addr = buffer.CpuAddr(); + const VAddr buf_end_addr = buf_start_addr + buffer.SizeBytes(); + if (buf_start_addr < end_addr && addr < buf_end_addr) { + return true; + } + page = Common::DivCeil(end_addr, PAGE_SIZE); + } + return false; +} + template bool BufferCache

::IsRegionCpuModified(VAddr addr, size_t size) { const u64 page_end = Common::DivCeil(addr + size, PAGE_SIZE); diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp index 9c657e32d..7a06531b4 100755 --- a/src/video_core/engines/maxwell_dma.cpp +++ b/src/video_core/engines/maxwell_dma.cpp @@ -87,9 +87,11 @@ void MaxwellDMA::CopyPitchToPitch() { // TODO: allow multisized components. if (is_buffer_clear) { ASSERT(regs.remap_const.component_size_minus_one == 3); + accelerate.BufferClear(regs.offset_out, regs.line_length_in, regs.remap_consta_value); std::vector tmp_buffer(regs.line_length_in, regs.remap_consta_value); - memory_manager.WriteBlock(regs.offset_out, reinterpret_cast(tmp_buffer.data()), - regs.line_length_in * sizeof(u32)); + memory_manager.WriteBlockUnsafe(regs.offset_out, + reinterpret_cast(tmp_buffer.data()), + regs.line_length_in * sizeof(u32)); return; } UNIMPLEMENTED_IF(regs.launch_dma.remap_enable != 0); diff --git a/src/video_core/engines/maxwell_dma.h b/src/video_core/engines/maxwell_dma.h index 4ed0d0996..d3329b0f8 100755 --- a/src/video_core/engines/maxwell_dma.h +++ b/src/video_core/engines/maxwell_dma.h @@ -31,6 +31,8 @@ class AccelerateDMAInterface { public: /// Write the value to the register identified by method. virtual bool BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) = 0; + + virtual bool BufferClear(GPUVAddr src_address, u64 amount, u32 value) = 0; }; /** diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.cpp b/src/video_core/renderer_opengl/gl_buffer_cache.cpp index 91c45c9c0..0703614de 100755 --- a/src/video_core/renderer_opengl/gl_buffer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_buffer_cache.cpp @@ -145,6 +145,12 @@ void BufferCacheRuntime::CopyBuffer(Buffer& dst_buffer, Buffer& src_buffer, } } +void BufferCacheRuntime::ClearBuffer(Buffer& dest_buffer, u32 offset, size_t size, u32 value) { + glClearNamedBufferSubData(dest_buffer.Handle(), GL_R32UI, static_cast(offset), + static_cast(size / sizeof(u32)), GL_RGBA, GL_UNSIGNED_INT, + &value); +} + void BufferCacheRuntime::BindIndexBuffer(Buffer& buffer, u32 offset, u32 size) { if (has_unified_vertex_buffers) { buffer.MakeResident(GL_READ_ONLY); diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.h b/src/video_core/renderer_opengl/gl_buffer_cache.h index af5720afd..060d36427 100755 --- a/src/video_core/renderer_opengl/gl_buffer_cache.h +++ b/src/video_core/renderer_opengl/gl_buffer_cache.h @@ -67,6 +67,8 @@ public: void CopyBuffer(Buffer& dst_buffer, Buffer& src_buffer, std::span copies); + void ClearBuffer(Buffer& dest_buffer, u32 offset, size_t size, u32 value); + void BindIndexBuffer(Buffer& buffer, u32 offset, u32 size); void BindVertexBuffer(u32 index, Buffer& buffer, u32 offset, u32 size, u32 stride); diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index af6156d90..d5169e9c4 100755 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -1051,4 +1051,9 @@ bool AccelerateDMA::BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 return buffer_cache.DMACopy(src_address, dest_address, amount); } +bool AccelerateDMA::BufferClear(GPUVAddr src_address, u64 amount, u32 value) { + std::scoped_lock lock{buffer_cache.mutex}; + return buffer_cache.DMAClear(src_address, amount, value); +} + } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index b7ad2252f..2618e095a 100755 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -63,6 +63,8 @@ public: bool BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) override; + bool BufferClear(GPUVAddr src_address, u64 amount, u32 value) override; + private: BufferCache& buffer_cache; }; diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index 6738106e1..e8f8ac5f3 100755 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -172,6 +172,30 @@ void BufferCacheRuntime::CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer, }); } +void BufferCacheRuntime::ClearBuffer(VkBuffer dest_buffer, u32 offset, size_t size, u32 value) { + static constexpr VkMemoryBarrier READ_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, + .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + }; + static constexpr VkMemoryBarrier WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, + }; + + scheduler.RequestOutsideRenderPassOperationContext(); + scheduler.Record([dest_buffer, offset, size, value](vk::CommandBuffer cmdbuf) { + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, + 0, READ_BARRIER); + cmdbuf.FillBuffer(dest_buffer, offset, size, value); + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, + 0, WRITE_BARRIER); + }); +} + void BufferCacheRuntime::BindIndexBuffer(PrimitiveTopology topology, IndexFormat index_format, u32 base_vertex, u32 num_indices, VkBuffer buffer, u32 offset, [[maybe_unused]] u32 size) { diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.h b/src/video_core/renderer_vulkan/vk_buffer_cache.h index d3fa2fba8..d29aab907 100755 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.h +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.h @@ -72,6 +72,8 @@ public: void CopyBuffer(VkBuffer src_buffer, VkBuffer dst_buffer, std::span copies); + void ClearBuffer(VkBuffer dest_buffer, u32 offset, size_t size, u32 value); + void BindIndexBuffer(PrimitiveTopology topology, IndexFormat index_format, u32 num_indices, u32 base_vertex, VkBuffer buffer, u32 offset, u32 size); diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index 96d56473a..8ca5fca61 100755 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -541,6 +541,11 @@ void RasterizerVulkan::FlushWork() { AccelerateDMA::AccelerateDMA(BufferCache& buffer_cache_) : buffer_cache{buffer_cache_} {} +bool AccelerateDMA::BufferClear(GPUVAddr src_address, u64 amount, u32 value) { + std::scoped_lock lock{buffer_cache.mutex}; + return buffer_cache.DMAClear(src_address, amount, value); +} + bool AccelerateDMA::BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) { std::scoped_lock lock{buffer_cache.mutex}; return buffer_cache.DMACopy(src_address, dest_address, amount); diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h index 0b6a0da94..2fc249563 100755 --- a/src/video_core/renderer_vulkan/vk_rasterizer.h +++ b/src/video_core/renderer_vulkan/vk_rasterizer.h @@ -55,6 +55,8 @@ public: bool BufferCopy(GPUVAddr start_address, GPUVAddr end_address, u64 amount) override; + bool BufferClear(GPUVAddr src_address, u64 amount, u32 value) override; + private: BufferCache& buffer_cache; };