8 files changed, 127 insertions, 125 deletions

diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h
index 78d6886fb..9fafed4a2 100644
--- a/src/video_core/engines/shader_bytecode.h
+++ b/src/video_core/engines/shader_bytecode.h
@@ -616,6 +616,14 @@ union Instruction {
     } shfl;
 
     union {
+        BitField<44, 1, u64> ftz;
+        BitField<39, 2, u64> tab5cb8_2;
+        BitField<38, 1, u64> ndv;
+        BitField<47, 1, u64> cc;
+        BitField<28, 8, u64> swizzle;
+    } fswzadd;
+
+    union {
         BitField<8, 8, Register> gpr;
         BitField<20, 24, s64> offset;
     } gmem;
@@ -1592,6 +1600,7 @@ public:
         DEPBAR,
         VOTE,
         SHFL,
+        FSWZADD,
         BFE_C,
         BFE_R,
         BFE_IMM,
@@ -1890,6 +1899,7 @@ private:
             INST("1111000011110---", Id::DEPBAR, Type::Synch, "DEPBAR"),
             INST("0101000011011---", Id::VOTE, Type::Warp, "VOTE"),
             INST("1110111100010---", Id::SHFL, Type::Warp, "SHFL"),
+            INST("0101000011111---", Id::FSWZADD, Type::Warp, "FSWZADD"),
             INST("1110111111011---", Id::LD_A, Type::Memory, "LD_A"),
             INST("1110111101001---", Id::LD_S, Type::Memory, "LD_S"),
             INST("1110111101000---", Id::LD_L, Type::Memory, "LD_L"),
diff --git a/src/video_core/renderer_opengl/gl_device.cpp b/src/video_core/renderer_opengl/gl_device.cpp
index c65b24c69..b30d5be74 100644
--- a/src/video_core/renderer_opengl/gl_device.cpp
+++ b/src/video_core/renderer_opengl/gl_device.cpp
@@ -62,6 +62,7 @@ Device::Device() {
     max_varyings = GetInteger<u32>(GL_MAX_VARYING_VECTORS);
     has_warp_intrinsics = GLAD_GL_NV_gpu_shader5 && GLAD_GL_NV_shader_thread_group &&
                           GLAD_GL_NV_shader_thread_shuffle;
+    has_shader_ballot = GLAD_GL_ARB_shader_ballot;
     has_vertex_viewport_layer = GLAD_GL_ARB_shader_viewport_layer_array;
     has_image_load_formatted = HasExtension(extensions, "GL_EXT_shader_image_load_formatted");
     has_variable_aoffi = TestVariableAoffi();
@@ -79,6 +80,7 @@ Device::Device(std::nullptr_t) {
     max_vertex_attributes = 16;
     max_varyings = 15;
     has_warp_intrinsics = true;
+    has_shader_ballot = true;
     has_vertex_viewport_layer = true;
     has_image_load_formatted = true;
     has_variable_aoffi = true;
diff --git a/src/video_core/renderer_opengl/gl_device.h b/src/video_core/renderer_opengl/gl_device.h
index bf35bd0b6..6c86fe207 100644
--- a/src/video_core/renderer_opengl/gl_device.h
+++ b/src/video_core/renderer_opengl/gl_device.h
@@ -34,6 +34,10 @@ public:
         return has_warp_intrinsics;
     }
 
+    bool HasShaderBallot() const {
+        return has_shader_ballot;
+    }
+
     bool HasVertexViewportLayer() const {
         return has_vertex_viewport_layer;
     }
@@ -68,6 +72,7 @@ private:
     u32 max_vertex_attributes{};
     u32 max_varyings{};
     bool has_warp_intrinsics{};
+    bool has_shader_ballot{};
     bool has_vertex_viewport_layer{};
     bool has_image_load_formatted{};
     bool has_variable_aoffi{};
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp
index 35e5214a5..04a239a39 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp
@@ -275,16 +275,25 @@ CachedProgram BuildShader(const Device& device, u64 unique_identifier, ProgramTy
     std::string source = fmt::format(R"(// {}
 #version 430 core
 #extension GL_ARB_separate_shader_objects : enable
-#extension GL_ARB_shader_viewport_layer_array : enable
-#extension GL_EXT_shader_image_load_formatted : enable
-#extension GL_NV_gpu_shader5 : enable
-#extension GL_NV_shader_thread_group : enable
-#extension GL_NV_shader_thread_shuffle : enable
 )",
                                      GetShaderId(unique_identifier, program_type));
     if (is_compute) {
         source += "#extension GL_ARB_compute_variable_group_size : require\n";
     }
+    if (device.HasShaderBallot()) {
+        source += "#extension GL_ARB_shader_ballot : require\n";
+    }
+    if (device.HasVertexViewportLayer()) {
+        source += "#extension GL_ARB_shader_viewport_layer_array : require\n";
+    }
+    if (device.HasImageLoadFormatted()) {
+        source += "#extension GL_EXT_shader_image_load_formatted : require\n";
+    }
+    if (device.HasWarpIntrinsics()) {
+        source += "#extension GL_NV_gpu_shader5 : require\n"
+                  "#extension GL_NV_shader_thread_group : require\n"
+                  "#extension GL_NV_shader_thread_shuffle : require\n";
+    }
     source += '\n';
 
     if (!is_compute) {
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
index 0ce59a852..e56ed51de 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
@@ -1379,6 +1379,26 @@ private:
         return GenerateUnary(operation, "float", Type::Float, type);
     }
 
+    Expression FSwizzleAdd(Operation operation) {
+        const std::string op_a = VisitOperand(operation, 0).AsFloat();
+        const std::string op_b = VisitOperand(operation, 1).AsFloat();
+
+        if (!device.HasShaderBallot()) {
+            LOG_ERROR(Render_OpenGL, "Shader ballot is unavailable but required by the shader");
+            return {fmt::format("{} + {}", op_a, op_b), Type::Float};
+        }
+
+        const std::string instr_mask = VisitOperand(operation, 2).AsUint();
+        const std::string mask = code.GenerateTemporary();
+        code.AddLine("uint {} = ({} >> ((gl_SubGroupInvocationARB & 3) << 1)) & 3;", mask,
+                     instr_mask);
+
+        const std::string modifier_a = fmt::format("fswzadd_modifiers_a[{}]", mask);
+        const std::string modifier_b = fmt::format("fswzadd_modifiers_b[{}]", mask);
+        return {fmt::format("(({} * {}) + ({} * {}))", op_a, modifier_a, op_b, modifier_b),
+                Type::Float};
+    }
+
     Expression ICastFloat(Operation operation) {
         return GenerateUnary(operation, "int", Type::Int, Type::Float);
     }
@@ -1942,34 +1962,24 @@ private:
         return Vote(operation, "allThreadsEqualNV");
     }
 
-    template <const std::string_view& func>
-    Expression Shuffle(Operation operation) {
-        const std::string value = VisitOperand(operation, 0).AsFloat();
-        if (!device.HasWarpIntrinsics()) {
-            LOG_ERROR(Render_OpenGL, "Nvidia shuffle intrinsics are required by this shader");
-            // On a "single-thread" device we are either on the same thread or out of bounds. Both
-            // cases return the passed value.
-            return {value, Type::Float};
+    Expression ThreadId(Operation operation) {
+        if (!device.HasShaderBallot()) {
+            LOG_ERROR(Render_OpenGL, "Shader ballot is unavailable but required by the shader");
+            return {"0U", Type::Uint};
         }
-
-        const std::string index = VisitOperand(operation, 1).AsUint();
-        const std::string width = VisitOperand(operation, 2).AsUint();
-        return {fmt::format("{}({}, {}, {})", func, value, index, width), Type::Float};
+        return {"gl_SubGroupInvocationARB", Type::Uint};
     }
 
-    template <const std::string_view& func>
-    Expression InRangeShuffle(Operation operation) {
-        const std::string index = VisitOperand(operation, 0).AsUint();
-        const std::string width = VisitOperand(operation, 1).AsUint();
-        if (!device.HasWarpIntrinsics()) {
-            // On a "single-thread" device we are only in bounds when the requested index is 0.
-            return {fmt::format("({} == 0U)", index), Type::Bool};
+    Expression ShuffleIndexed(Operation operation) {
+        std::string value = VisitOperand(operation, 0).AsFloat();
+
+        if (!device.HasShaderBallot()) {
+            LOG_ERROR(Render_OpenGL, "Shader ballot is unavailable but required by the shader");
+            return {std::move(value), Type::Float};
         }
 
-        const std::string in_range = code.GenerateTemporary();
-        code.AddLine("bool {};", in_range);
-        code.AddLine("{}(0U, {}, {}, {});", func, index, width, in_range);
-        return {in_range, Type::Bool};
+        const std::string index = VisitOperand(operation, 1).AsUint();
+        return {fmt::format("readInvocationARB({}, {})", value, index), Type::Float};
     }
 
     struct Func final {
@@ -1981,11 +1991,6 @@ private:
         static constexpr std::string_view Or = "Or";
         static constexpr std::string_view Xor = "Xor";
         static constexpr std::string_view Exchange = "Exchange";
-
-        static constexpr std::string_view ShuffleIndexed = "shuffleNV";
-        static constexpr std::string_view ShuffleUp = "shuffleUpNV";
-        static constexpr std::string_view ShuffleDown = "shuffleDownNV";
-        static constexpr std::string_view ShuffleButterfly = "shuffleXorNV";
     };
 
     static constexpr std::array operation_decompilers = {
@@ -2016,6 +2021,7 @@ private:
         &GLSLDecompiler::FTrunc,
         &GLSLDecompiler::FCastInteger<Type::Int>,
         &GLSLDecompiler::FCastInteger<Type::Uint>,
+        &GLSLDecompiler::FSwizzleAdd,
 
         &GLSLDecompiler::Add<Type::Int>,
         &GLSLDecompiler::Mul<Type::Int>,
@@ -2151,15 +2157,8 @@ private:
         &GLSLDecompiler::VoteAny,
         &GLSLDecompiler::VoteEqual,
 
-        &GLSLDecompiler::Shuffle<Func::ShuffleIndexed>,
-        &GLSLDecompiler::Shuffle<Func::ShuffleUp>,
-        &GLSLDecompiler::Shuffle<Func::ShuffleDown>,
-        &GLSLDecompiler::Shuffle<Func::ShuffleButterfly>,
-
-        &GLSLDecompiler::InRangeShuffle<Func::ShuffleIndexed>,
-        &GLSLDecompiler::InRangeShuffle<Func::ShuffleUp>,
-        &GLSLDecompiler::InRangeShuffle<Func::ShuffleDown>,
-        &GLSLDecompiler::InRangeShuffle<Func::ShuffleButterfly>,
+        &GLSLDecompiler::ThreadId,
+        &GLSLDecompiler::ShuffleIndexed,
     };
     static_assert(operation_decompilers.size() == static_cast<std::size_t>(OperationCode::Amount));
 
@@ -2492,6 +2491,9 @@ bvec2 HalfFloatNanComparison(bvec2 comparison, vec2 pair1, vec2 pair2) {
     bvec2 is_nan2 = isnan(pair2);
     return bvec2(comparison.x || is_nan1.x || is_nan2.x, comparison.y || is_nan1.y || is_nan2.y);
 }
+
+const float fswzadd_modifiers_a[] = float[4](-1.0f,  1.0f, -1.0f,  0.0f );
+const float fswzadd_modifiers_b[] = float[4](-1.0f, -1.0f,  1.0f, -1.0f );
 )";
 }
 
diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
index 42cf068b6..2850d5b59 100644
--- a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
+++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
@@ -783,6 +783,11 @@ private:
         return {};
     }
 
+    Id FSwizzleAdd(Operation operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
     Id HNegate(Operation operation) {
         UNIMPLEMENTED();
         return {};
@@ -1195,42 +1200,12 @@ private:
         return {};
     }
 
-    Id ShuffleIndexed(Operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
-
-    Id ShuffleUp(Operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
-
-    Id ShuffleDown(Operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
-
-    Id ShuffleButterfly(Operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
-
-    Id InRangeShuffleIndexed(Operation) {
+    Id ThreadId(Operation) {
         UNIMPLEMENTED();
         return {};
     }
 
-    Id InRangeShuffleUp(Operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
-
-    Id InRangeShuffleDown(Operation) {
-        UNIMPLEMENTED();
-        return {};
-    }
-
-    Id InRangeShuffleButterfly(Operation) {
+    Id ShuffleIndexed(Operation) {
         UNIMPLEMENTED();
         return {};
     }
@@ -1393,6 +1368,7 @@ private:
         &SPIRVDecompiler::Unary<&Module::OpTrunc, Type::Float>,
         &SPIRVDecompiler::Unary<&Module::OpConvertSToF, Type::Float, Type::Int>,
         &SPIRVDecompiler::Unary<&Module::OpConvertUToF, Type::Float, Type::Uint>,
+        &SPIRVDecompiler::FSwizzleAdd,
 
         &SPIRVDecompiler::Binary<&Module::OpIAdd, Type::Int>,
         &SPIRVDecompiler::Binary<&Module::OpIMul, Type::Int>,
@@ -1528,15 +1504,8 @@ private:
         &SPIRVDecompiler::VoteAny,
         &SPIRVDecompiler::VoteEqual,
 
+        &SPIRVDecompiler::ThreadId,
         &SPIRVDecompiler::ShuffleIndexed,
-        &SPIRVDecompiler::ShuffleUp,
-        &SPIRVDecompiler::ShuffleDown,
-        &SPIRVDecompiler::ShuffleButterfly,
-
-        &SPIRVDecompiler::InRangeShuffleIndexed,
-        &SPIRVDecompiler::InRangeShuffleUp,
-        &SPIRVDecompiler::InRangeShuffleDown,
-        &SPIRVDecompiler::InRangeShuffleButterfly,
     };
     static_assert(operation_decompilers.size() == static_cast<std::size_t>(OperationCode::Amount));
 
diff --git a/src/video_core/shader/decode/warp.cpp b/src/video_core/shader/decode/warp.cpp
index fa8a250cc..d98d0e1dd 100644
--- a/src/video_core/shader/decode/warp.cpp
+++ b/src/video_core/shader/decode/warp.cpp
@@ -17,6 +17,7 @@ using Tegra::Shader::ShuffleOperation;
 using Tegra::Shader::VoteOperation;
 
 namespace {
+
 OperationCode GetOperationCode(VoteOperation vote_op) {
     switch (vote_op) {
     case VoteOperation::All:
@@ -30,6 +31,7 @@ OperationCode GetOperationCode(VoteOperation vote_op) {
         return OperationCode::VoteAll;
     }
 }
+
 } // Anonymous namespace
 
 u32 ShaderIR::DecodeWarp(NodeBlock& bb, u32 pc) {
@@ -46,50 +48,59 @@ u32 ShaderIR::DecodeWarp(NodeBlock& bb, u32 pc) {
         break;
     }
     case OpCode::Id::SHFL: {
-        Node width = [this, instr] {
-            Node mask = instr.shfl.is_mask_imm ? Immediate(static_cast<u32>(instr.shfl.mask_imm))
-                                               : GetRegister(instr.gpr39);
-
-            // Convert the obscure SHFL mask back into GL_NV_shader_thread_shuffle's width. This has
-            // been done reversing Nvidia's math. It won't work on all cases due to SHFL having
-            // different parameters that don't properly map to GLSL's interface, but it should work
-            // for cases emitted by Nvidia's compiler.
-            if (instr.shfl.operation == ShuffleOperation::Up) {
-                return Operation(
-                    OperationCode::ILogicalShiftRight,
-                    Operation(OperationCode::IAdd, std::move(mask), Immediate(-0x2000)),
-                    Immediate(8));
-            } else {
-                return Operation(OperationCode::ILogicalShiftRight,
-                                 Operation(OperationCode::IAdd, Immediate(0x201F),
-                                           Operation(OperationCode::INegate, std::move(mask))),
-                                 Immediate(8));
-            }
-        }();
+        Node mask = instr.shfl.is_mask_imm ? Immediate(static_cast<u32>(instr.shfl.mask_imm))
+                                           : GetRegister(instr.gpr39);
+        Node index = instr.shfl.is_index_imm ? Immediate(static_cast<u32>(instr.shfl.index_imm))
+                                             : GetRegister(instr.gpr20);
+
+        Node thread_id = Operation(OperationCode::ThreadId);
+        Node clamp = Operation(OperationCode::IBitwiseAnd, mask, Immediate(0x1FU));
+        Node seg_mask = BitfieldExtract(mask, 8, 16);
 
-        const auto [operation, in_range] = [instr]() -> std::pair<OperationCode, OperationCode> {
+        Node neg_seg_mask = Operation(OperationCode::IBitwiseNot, seg_mask);
+        Node min_thread_id = Operation(OperationCode::IBitwiseAnd, thread_id, seg_mask);
+        Node max_thread_id = Operation(OperationCode::IBitwiseOr, min_thread_id,
+                                       Operation(OperationCode::IBitwiseAnd, clamp, neg_seg_mask));
+
+        Node src_thread_id = [instr, index, neg_seg_mask, min_thread_id, thread_id] {
             switch (instr.shfl.operation) {
             case ShuffleOperation::Idx:
-                return {OperationCode::ShuffleIndexed, OperationCode::InRangeShuffleIndexed};
-            case ShuffleOperation::Up:
-                return {OperationCode::ShuffleUp, OperationCode::InRangeShuffleUp};
+                return Operation(OperationCode::IBitwiseOr,
+                                 Operation(OperationCode::IBitwiseAnd, index, neg_seg_mask),
+                                 min_thread_id);
             case ShuffleOperation::Down:
-                return {OperationCode::ShuffleDown, OperationCode::InRangeShuffleDown};
+                return Operation(OperationCode::IAdd, thread_id, index);
+            case ShuffleOperation::Up:
+                return Operation(OperationCode::IAdd, thread_id,
+                                 Operation(OperationCode::INegate, index));
             case ShuffleOperation::Bfly:
-                return {OperationCode::ShuffleButterfly, OperationCode::InRangeShuffleButterfly};
+                return Operation(OperationCode::IBitwiseXor, thread_id, index);
             }
-            UNREACHABLE_MSG("Invalid SHFL operation: {}",
-                            static_cast<u64>(instr.shfl.operation.Value()));
-            return {};
+            UNREACHABLE();
+            return Immediate(0U);
         }();
 
-        // Setting the predicate before the register is intentional to avoid overwriting.
-        Node index = instr.shfl.is_index_imm ? Immediate(static_cast<u32>(instr.shfl.index_imm))
-                                             : GetRegister(instr.gpr20);
-        SetPredicate(bb, instr.shfl.pred48, Operation(in_range, index, width));
+        Node in_bounds = [instr, src_thread_id, min_thread_id, max_thread_id] {
+            if (instr.shfl.operation == ShuffleOperation::Up) {
+                return Operation(OperationCode::LogicalIGreaterEqual, src_thread_id, min_thread_id);
+            } else {
+                return Operation(OperationCode::LogicalILessEqual, src_thread_id, max_thread_id);
+            }
+        }();
+
+        SetPredicate(bb, instr.shfl.pred48, in_bounds);
         SetRegister(
             bb, instr.gpr0,
-            Operation(operation, GetRegister(instr.gpr8), std::move(index), std::move(width)));
+            Operation(OperationCode::ShuffleIndexed, GetRegister(instr.gpr8), src_thread_id));
+        break;
+    }
+    case OpCode::Id::FSWZADD: {
+        UNIMPLEMENTED_IF(instr.fswzadd.ndv);
+
+        Node op_a = GetRegister(instr.gpr8);
+        Node op_b = GetRegister(instr.gpr20);
+        Node mask = Immediate(static_cast<u32>(instr.fswzadd.swizzle));
+        SetRegister(bb, instr.gpr0, Operation(OperationCode::FSwizzleAdd, op_a, op_b, mask));
         break;
     }
     default:
diff --git a/src/video_core/shader/node.h b/src/video_core/shader/node.h
index 4300d9ff4..54217e6a4 100644
--- a/src/video_core/shader/node.h
+++ b/src/video_core/shader/node.h
@@ -47,6 +47,7 @@ enum class OperationCode {
     FTrunc,        /// (MetaArithmetic, float a) -> float
     FCastInteger,  /// (MetaArithmetic, int a) -> float
     FCastUInteger, /// (MetaArithmetic, uint a) -> float
+    FSwizzleAdd,   /// (float a, float b, uint mask) -> float
 
     IAdd,                  /// (MetaArithmetic, int a, int b) -> int
     IMul,                  /// (MetaArithmetic, int a, int b) -> int
@@ -181,15 +182,8 @@ enum class OperationCode {
     VoteAny,      /// (bool) -> bool
     VoteEqual,    /// (bool) -> bool
 
-    ShuffleIndexed,   /// (uint value, uint index, uint width) -> uint
-    ShuffleUp,        /// (uint value, uint index, uint width) -> uint
-    ShuffleDown,      /// (uint value, uint index, uint width) -> uint
-    ShuffleButterfly, /// (uint value, uint index, uint width) -> uint
-
-    InRangeShuffleIndexed,   /// (uint index, uint width) -> bool
-    InRangeShuffleUp,        /// (uint index, uint width) -> bool
-    InRangeShuffleDown,      /// (uint index, uint width) -> bool
-    InRangeShuffleButterfly, /// (uint index, uint width) -> bool
+    ThreadId,       /// () -> uint
+    ShuffleIndexed, /// (uint value, uint index) -> uint
 
     Amount,
 };