Merge pull request #3981 from ReinUsesLisp/bar

shader/other: Implement BAR.SYNC 0x0

5 files changed, 33 insertions, 0 deletions

diff --git a/externals/sirit b/externals/sirit
-Subproject 414fc4dbd28d8fe48f735a0c389db8a234f733c
+Subproject a62c5bbc100a5e5a31ea0ccc4a78d8fa6a4167c
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
index 84ca830f4..253484968 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
@@ -2335,6 +2335,15 @@ private:
         return {fmt::format("readInvocationARB({}, {})", value, index), Type::Float};
     }
 
+    Expression Barrier(Operation) {
+        if (!ir.IsDecompiled()) {
+            LOG_ERROR(Render_OpenGL, "barrier() used but shader is not decompiled");
+            return {};
+        }
+        code.AddLine("barrier();");
+        return {};
+    }
+
     Expression MemoryBarrierGL(Operation) {
         code.AddLine("memoryBarrier();");
         return {};
@@ -2581,6 +2590,7 @@ private:
         &GLSLDecompiler::ThreadMask<Func::Lt>,
         &GLSLDecompiler::ShuffleIndexed,
 
+        &GLSLDecompiler::Barrier,
         &GLSLDecompiler::MemoryBarrierGL,
     };
     static_assert(operation_decompilers.size() == static_cast<std::size_t>(OperationCode::Amount));
diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
index b8169d832..890f34a2c 100644
--- a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
+++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
@@ -2199,6 +2199,22 @@ private:
         return {OpSubgroupReadInvocationKHR(t_float, value, index), Type::Float};
     }
 
+    Expression Barrier(Operation) {
+        if (!ir.IsDecompiled()) {
+            LOG_ERROR(Render_Vulkan, "OpBarrier used by shader is not decompiled");
+            return {};
+        }
+
+        const auto scope = spv::Scope::Workgroup;
+        const auto memory = spv::Scope::Workgroup;
+        const auto semantics =
+            spv::MemorySemanticsMask::WorkgroupMemory | spv::MemorySemanticsMask::AcquireRelease;
+        OpControlBarrier(Constant(t_uint, static_cast<u32>(scope)),
+                         Constant(t_uint, static_cast<u32>(memory)),
+                         Constant(t_uint, static_cast<u32>(semantics)));
+        return {};
+    }
+
     Expression MemoryBarrierGL(Operation) {
         const auto scope = spv::Scope::Device;
         const auto semantics =
@@ -2664,6 +2680,7 @@ private:
         &SPIRVDecompiler::ThreadMask<4>, // Lt
         &SPIRVDecompiler::ShuffleIndexed,
 
+        &SPIRVDecompiler::Barrier,
         &SPIRVDecompiler::MemoryBarrierGL,
     };
     static_assert(operation_decompilers.size() == static_cast<std::size_t>(OperationCode::Amount));
diff --git a/src/video_core/shader/decode/other.cpp b/src/video_core/shader/decode/other.cpp
index 399a455c4..694b325e1 100644
--- a/src/video_core/shader/decode/other.cpp
+++ b/src/video_core/shader/decode/other.cpp
@@ -293,6 +293,11 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
         SetRegister(bb, instr.gpr0, GetRegister(instr.gpr8));
         break;
     }
+    case OpCode::Id::BAR: {
+        UNIMPLEMENTED_IF_MSG(instr.value != 0xF0A81B8000070000ULL, "BAR is not BAR.SYNC 0x0");
+        bb.push_back(Operation(OperationCode::Barrier));
+        break;
+    }
     case OpCode::Id::MEMBAR: {
         UNIMPLEMENTED_IF(instr.membar.type != Tegra::Shader::MembarType::GL);
         UNIMPLEMENTED_IF(instr.membar.unknown != Tegra::Shader::MembarUnknown::Default);
diff --git a/src/video_core/shader/node.h b/src/video_core/shader/node.h
index cce8aeebe..c06512413 100644
--- a/src/video_core/shader/node.h
+++ b/src/video_core/shader/node.h
@@ -233,6 +233,7 @@ enum class OperationCode {
     ThreadLtMask,   /// () -> uint
     ShuffleIndexed, /// (uint value, uint index) -> uint
 
+    Barrier,         /// () -> void
     MemoryBarrierGL, /// () -> void
 
     Amount,