7 files changed, 103 insertions, 50 deletions

diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_atomic.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_atomic.cpp
index 4b3043b65..0ce73f289 100644
--- a/src/shader_recompiler/backend/spirv/emit_spirv_atomic.cpp
+++ b/src/shader_recompiler/backend/spirv/emit_spirv_atomic.cpp
@@ -69,6 +69,11 @@ Id StorageAtomicU32(EmitContext& ctx, const IR::Value& binding, const IR::Value&
 Id StorageAtomicU64(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset, Id value,
                     Id (Sirit::Module::*atomic_func)(Id, Id, Id, Id, Id),
                     Id (Sirit::Module::*non_atomic_func)(Id, Id, Id)) {
+    if (!ctx.profile.support_descriptor_aliasing) {
+        LOG_WARNING(Shader_SPIRV, "Descriptor aliasing not supported, this cannot be atomic.");
+        return ctx.ConstantNull(ctx.U64);
+    }
+
     if (ctx.profile.support_int64_atomics) {
         const Id pointer{StoragePointer(ctx, ctx.storage_types.U64, &StorageDefinitions::U64,
                                         binding, offset, sizeof(u64))};
@@ -86,6 +91,11 @@ Id StorageAtomicU64(EmitContext& ctx, const IR::Value& binding, const IR::Value&
 
 Id StorageAtomicU32x2(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset, Id value,
                       Id (Sirit::Module::*non_atomic_func)(Id, Id, Id)) {
+    if (!ctx.profile.support_descriptor_aliasing) {
+        LOG_WARNING(Shader_SPIRV, "Descriptor aliasing not supported, this cannot be atomic.");
+        return ctx.ConstantNull(ctx.U32[2]);
+    }
+
     LOG_WARNING(Shader_SPIRV, "Int64 atomics not supported, fallback to non-atomic");
     const Id pointer{StoragePointer(ctx, ctx.storage_types.U32x2, &StorageDefinitions::U32x2,
                                     binding, offset, sizeof(u32[2]))};
diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp
index 07c2b7b8a..2868fc57d 100644
--- a/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp
+++ b/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp
@@ -10,27 +10,6 @@
 
 namespace Shader::Backend::SPIRV {
 namespace {
-struct AttrInfo {
-    Id pointer;
-    Id id;
-    bool needs_cast;
-};
-
-std::optional<AttrInfo> AttrTypes(EmitContext& ctx, u32 index) {
-    const AttributeType type{ctx.runtime_info.generic_input_types.at(index)};
-    switch (type) {
-    case AttributeType::Float:
-        return AttrInfo{ctx.input_f32, ctx.F32[1], false};
-    case AttributeType::UnsignedInt:
-        return AttrInfo{ctx.input_u32, ctx.U32[1], true};
-    case AttributeType::SignedInt:
-        return AttrInfo{ctx.input_s32, ctx.TypeInt(32, true), true};
-    case AttributeType::Disabled:
-        return std::nullopt;
-    }
-    throw InvalidArgument("Invalid attribute type {}", type);
-}
-
 template <typename... Args>
 Id AttrPointer(EmitContext& ctx, Id pointer_type, Id vertex, Id base, Args&&... args) {
     switch (ctx.stage) {
@@ -302,15 +281,26 @@ Id EmitGetAttribute(EmitContext& ctx, IR::Attribute attr, Id vertex) {
     const u32 element{static_cast<u32>(attr) % 4};
     if (IR::IsGeneric(attr)) {
         const u32 index{IR::GenericAttributeIndex(attr)};
-        const std::optional<AttrInfo> type{AttrTypes(ctx, index)};
-        if (!type || !ctx.runtime_info.previous_stage_stores.Generic(index, element)) {
+        const auto& generic{ctx.input_generics.at(index)};
+        if (!ValidId(generic.id)) {
             // Attribute is disabled or varying component is not written
             return ctx.Const(element == 3 ? 1.0f : 0.0f);
         }
-        const Id generic_id{ctx.input_generics.at(index)};
-        const Id pointer{AttrPointer(ctx, type->pointer, vertex, generic_id, ctx.Const(element))};
-        const Id value{ctx.OpLoad(type->id, pointer)};
-        return type->needs_cast ? ctx.OpBitcast(ctx.F32[1], value) : value;
+        const Id pointer{
+            AttrPointer(ctx, generic.pointer_type, vertex, generic.id, ctx.Const(element))};
+        const Id value{ctx.OpLoad(generic.component_type, pointer)};
+        return [&ctx, generic, value]() {
+            switch (generic.load_op) {
+            case InputGenericLoadOp::Bitcast:
+                return ctx.OpBitcast(ctx.F32[1], value);
+            case InputGenericLoadOp::SToF:
+                return ctx.OpConvertSToF(ctx.F32[1], value);
+            case InputGenericLoadOp::UToF:
+                return ctx.OpConvertUToF(ctx.F32[1], value);
+            default:
+                return value;
+            };
+        }();
     }
     switch (attr) {
     case IR::Attribute::PrimitiveId:
diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp
index c5db19d09..77ff8c573 100644
--- a/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp
+++ b/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp
@@ -17,7 +17,22 @@ Id GetThreadId(EmitContext& ctx) {
 Id WarpExtract(EmitContext& ctx, Id value) {
     const Id thread_id{GetThreadId(ctx)};
     const Id local_index{ctx.OpShiftRightArithmetic(ctx.U32[1], thread_id, ctx.Const(5U))};
-    return ctx.OpVectorExtractDynamic(ctx.U32[1], value, local_index);
+    if (ctx.profile.has_broken_spirv_subgroup_mask_vector_extract_dynamic) {
+        const Id c0_sel{ctx.OpSelect(ctx.U32[1], ctx.OpIEqual(ctx.U1, local_index, ctx.Const(0U)),
+                                     ctx.OpCompositeExtract(ctx.U32[1], value, 0U), ctx.Const(0U))};
+        const Id c1_sel{ctx.OpSelect(ctx.U32[1], ctx.OpIEqual(ctx.U1, local_index, ctx.Const(1U)),
+                                     ctx.OpCompositeExtract(ctx.U32[1], value, 1U), ctx.Const(0U))};
+        const Id c2_sel{ctx.OpSelect(ctx.U32[1], ctx.OpIEqual(ctx.U1, local_index, ctx.Const(2U)),
+                                     ctx.OpCompositeExtract(ctx.U32[1], value, 2U), ctx.Const(0U))};
+        const Id c3_sel{ctx.OpSelect(ctx.U32[1], ctx.OpIEqual(ctx.U1, local_index, ctx.Const(3U)),
+                                     ctx.OpCompositeExtract(ctx.U32[1], value, 3U), ctx.Const(0U))};
+        const Id c0_or_c1{ctx.OpBitwiseOr(ctx.U32[1], c0_sel, c1_sel)};
+        const Id c2_or_c3{ctx.OpBitwiseOr(ctx.U32[1], c2_sel, c3_sel)};
+        const Id c0_or_c1_or_c2_or_c3{ctx.OpBitwiseOr(ctx.U32[1], c0_or_c1, c2_or_c3)};
+        return c0_or_c1_or_c2_or_c3;
+    } else {
+        return ctx.OpVectorExtractDynamic(ctx.U32[1], value, local_index);
+    }
 }
 
 Id LoadMask(EmitContext& ctx, Id mask) {
diff --git a/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp b/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp
index 47739794f..fd15f47ea 100644
--- a/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp
+++ b/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp
@@ -25,12 +25,6 @@ enum class Operation {
     FPMax,
 };
 
-struct AttrInfo {
-    Id pointer;
-    Id id;
-    bool needs_cast;
-};
-
 Id ImageType(EmitContext& ctx, const TextureDescriptor& desc) {
     const spv::ImageFormat format{spv::ImageFormat::Unknown};
     const Id type{ctx.F32[1]};
@@ -206,23 +200,37 @@ Id GetAttributeType(EmitContext& ctx, AttributeType type) {
         return ctx.TypeVector(ctx.TypeInt(32, true), 4);
     case AttributeType::UnsignedInt:
         return ctx.U32[4];
+    case AttributeType::SignedScaled:
+        return ctx.profile.support_scaled_attributes ? ctx.F32[4]
+                                                     : ctx.TypeVector(ctx.TypeInt(32, true), 4);
+    case AttributeType::UnsignedScaled:
+        return ctx.profile.support_scaled_attributes ? ctx.F32[4] : ctx.U32[4];
     case AttributeType::Disabled:
         break;
     }
     throw InvalidArgument("Invalid attribute type {}", type);
 }
 
-std::optional<AttrInfo> AttrTypes(EmitContext& ctx, u32 index) {
-    const AttributeType type{ctx.runtime_info.generic_input_types.at(index)};
+InputGenericInfo GetAttributeInfo(EmitContext& ctx, AttributeType type, Id id) {
     switch (type) {
     case AttributeType::Float:
-        return AttrInfo{ctx.input_f32, ctx.F32[1], false};
+        return InputGenericInfo{id, ctx.input_f32, ctx.F32[1], InputGenericLoadOp::None};
     case AttributeType::UnsignedInt:
-        return AttrInfo{ctx.input_u32, ctx.U32[1], true};
+        return InputGenericInfo{id, ctx.input_u32, ctx.U32[1], InputGenericLoadOp::Bitcast};
     case AttributeType::SignedInt:
-        return AttrInfo{ctx.input_s32, ctx.TypeInt(32, true), true};
+        return InputGenericInfo{id, ctx.input_s32, ctx.TypeInt(32, true),
+                                InputGenericLoadOp::Bitcast};
+    case AttributeType::SignedScaled:
+        return ctx.profile.support_scaled_attributes
+                   ? InputGenericInfo{id, ctx.input_f32, ctx.F32[1], InputGenericLoadOp::None}
+                   : InputGenericInfo{id, ctx.input_s32, ctx.TypeInt(32, true),
+                                      InputGenericLoadOp::SToF};
+    case AttributeType::UnsignedScaled:
+        return ctx.profile.support_scaled_attributes
+                   ? InputGenericInfo{id, ctx.input_f32, ctx.F32[1], InputGenericLoadOp::None}
+                   : InputGenericInfo{id, ctx.input_u32, ctx.U32[1], InputGenericLoadOp::UToF};
     case AttributeType::Disabled:
-        return std::nullopt;
+        return InputGenericInfo{};
     }
     throw InvalidArgument("Invalid attribute type {}", type);
 }
@@ -746,18 +754,29 @@ void EmitContext::DefineAttributeMemAccess(const Info& info) {
                 continue;
             }
             AddLabel(labels[label_index]);
-            const auto type{AttrTypes(*this, static_cast<u32>(index))};
-            if (!type) {
+            const auto& generic{input_generics.at(index)};
+            const Id generic_id{generic.id};
+            if (!ValidId(generic_id)) {
                 OpReturnValue(Const(0.0f));
                 ++label_index;
                 continue;
             }
-            const Id generic_id{input_generics.at(index)};
-            const Id pointer{is_array
-                                 ? OpAccessChain(type->pointer, generic_id, vertex, masked_index)
-                                 : OpAccessChain(type->pointer, generic_id, masked_index)};
-            const Id value{OpLoad(type->id, pointer)};
-            const Id result{type->needs_cast ? OpBitcast(F32[1], value) : value};
+            const Id pointer{
+                is_array ? OpAccessChain(generic.pointer_type, generic_id, vertex, masked_index)
+                         : OpAccessChain(generic.pointer_type, generic_id, masked_index)};
+            const Id value{OpLoad(generic.component_type, pointer)};
+            const Id result{[this, generic, value]() {
+                switch (generic.load_op) {
+                case InputGenericLoadOp::Bitcast:
+                    return OpBitcast(F32[1], value);
+                case InputGenericLoadOp::SToF:
+                    return OpConvertSToF(F32[1], value);
+                case InputGenericLoadOp::UToF:
+                    return OpConvertUToF(F32[1], value);
+                default:
+                    return value;
+                };
+            }()};
             OpReturnValue(result);
             ++label_index;
         }
@@ -1457,7 +1476,7 @@ void EmitContext::DefineInputs(const IR::Program& program) {
         const Id id{DefineInput(*this, type, true)};
         Decorate(id, spv::Decoration::Location, static_cast<u32>(index));
         Name(id, fmt::format("in_attr{}", index));
-        input_generics[index] = id;
+        input_generics[index] = GetAttributeInfo(*this, input_type, id);
 
         if (info.passthrough.Generic(index) && profile.support_geometry_shader_passthrough) {
             Decorate(id, spv::Decoration::PassthroughNV);
diff --git a/src/shader_recompiler/backend/spirv/spirv_emit_context.h b/src/shader_recompiler/backend/spirv/spirv_emit_context.h
index 768a4fbb5..e63330f11 100644
--- a/src/shader_recompiler/backend/spirv/spirv_emit_context.h
+++ b/src/shader_recompiler/backend/spirv/spirv_emit_context.h
@@ -95,6 +95,20 @@ struct StorageDefinitions {
     Id U32x4{};
 };
 
+enum class InputGenericLoadOp {
+    None,
+    Bitcast,
+    SToF,
+    UToF,
+};
+
+struct InputGenericInfo {
+    Id id;
+    Id pointer_type;
+    Id component_type;
+    InputGenericLoadOp load_op;
+};
+
 struct GenericElementInfo {
     Id id{};
     u32 first_element{};
@@ -283,7 +297,7 @@ public:
 
     bool need_input_position_indirect{};
     Id input_position{};
-    std::array<Id, 32> input_generics{};
+    std::array<InputGenericInfo, 32> input_generics{};
 
     Id output_point_size{};
     Id output_position{};
diff --git a/src/shader_recompiler/profile.h b/src/shader_recompiler/profile.h
index 9f88fb440..9ca97f6a4 100644
--- a/src/shader_recompiler/profile.h
+++ b/src/shader_recompiler/profile.h
@@ -43,6 +43,7 @@ struct Profile {
     bool support_gl_variable_aoffi{};
     bool support_gl_sparse_textures{};
     bool support_gl_derivative_control{};
+    bool support_scaled_attributes{};
 
     bool warp_size_potentially_larger_than_guest{};
 
@@ -77,6 +78,8 @@ struct Profile {
     bool has_gl_bool_ref_bug{};
     /// Ignores SPIR-V ordered vs unordered using GLSL semantics
     bool ignore_nan_fp_comparisons{};
+    /// Some drivers have broken support for OpVectorExtractDynamic on subgroup mask inputs
+    bool has_broken_spirv_subgroup_mask_vector_extract_dynamic{};
 
     u32 gl_max_compute_smem_size{};
 };
diff --git a/src/shader_recompiler/runtime_info.h b/src/shader_recompiler/runtime_info.h
index 549b81ef7..3b63c249f 100644
--- a/src/shader_recompiler/runtime_info.h
+++ b/src/shader_recompiler/runtime_info.h
@@ -17,6 +17,8 @@ enum class AttributeType : u8 {
     Float,
     SignedInt,
     UnsignedInt,
+    SignedScaled,
+    UnsignedScaled,
     Disabled,
 };