shader_recompiler: Replace buffer pulling with attribute divisor for instance step rates (#3238)

* shader_recompiler: Replace buffer pulling with attribute divisor for instance step rates

* flatten_extended_userdata: Remove special step rate buffer handling

* Review comments

* spirv_emit_context: Name all instance rate attribs properly

* spirv: Merge ReadConstBuffer again

template function only has 1 user now

* attribute: Add missing attributes

* translate: Reimplement step rate instance id

* Resolve validation warnings

* shader_recompiler: Separate vertex inputs from LS stage, cleanup tess

src/shader_recompiler/ir/attribute.cpp

@@ -100,22 +100,36 @@ std::string NameOf(Attribute attribute) {
         return "Param30";
     case Attribute::Param31:
         return "Param31";
+    case Attribute::ClipDistance:
+        return "ClipDistanace";
+    case Attribute::CullDistance:
+        return "CullDistance";
+    case Attribute::RenderTargetId:
+        return "RenderTargetId";
+    case Attribute::ViewportId:
+        return "ViewportId";
     case Attribute::VertexId:
         return "VertexId";
-    case Attribute::InstanceId:
-        return "InstanceId";
     case Attribute::PrimitiveId:
         return "PrimitiveId";
-    case Attribute::FragCoord:
-        return "FragCoord";
+    case Attribute::InstanceId:
+        return "InstanceId";
     case Attribute::IsFrontFace:
         return "IsFrontFace";
+    case Attribute::SampleIndex:
+        return "SampleIndex";
+    case Attribute::GlobalInvocationId:
+        return "GlobalInvocationId";
     case Attribute::WorkgroupId:
         return "WorkgroupId";
+    case Attribute::WorkgroupIndex:
+        return "WorkgroupIndex";
     case Attribute::LocalInvocationId:
         return "LocalInvocationId";
     case Attribute::LocalInvocationIndex:
         return "LocalInvocationIndex";
+    case Attribute::FragCoord:
+        return "FragCoord";
     case Attribute::InvocationId:
         return "InvocationId";
     case Attribute::PatchVertices:

src/shader_recompiler/ir/attribute.h

@@ -73,8 +73,6 @@ enum class Attribute : u64 {
     LocalInvocationId = 76,
     LocalInvocationIndex = 77,
     FragCoord = 78,
-    InstanceId0 = 79,  // step rate 0
-    InstanceId1 = 80,  // step rate 1
     InvocationId = 81, // TCS id in output patch and instanced geometry shader id
     PatchVertices = 82,
     TessellationEvaluationPointU = 83,

src/shader_recompiler/ir/ir_emitter.cpp

@@ -255,8 +255,8 @@ void IREmitter::SetM0(const U32& value) {
     Inst(Opcode::SetM0, value);
 }
 
-F32 IREmitter::GetAttribute(IR::Attribute attribute, u32 comp, IR::Value index) {
-    return Inst<F32>(Opcode::GetAttribute, attribute, Imm32(comp), index);
+F32 IREmitter::GetAttribute(IR::Attribute attribute, u32 comp, u32 index) {
+    return Inst<F32>(Opcode::GetAttribute, attribute, Imm32(comp), Imm32(index));
 }
 
 U32 IREmitter::GetAttributeU32(IR::Attribute attribute, u32 comp) {

src/shader_recompiler/ir/ir_emitter.h

@@ -81,8 +81,7 @@ public:
 
     [[nodiscard]] U1 Condition(IR::Condition cond);
 
-    [[nodiscard]] F32 GetAttribute(Attribute attribute, u32 comp = 0,
-                                   IR::Value index = IR::Value(u32(0u)));
+    [[nodiscard]] F32 GetAttribute(Attribute attribute, u32 comp = 0, u32 index = 0);
     [[nodiscard]] U32 GetAttributeU32(Attribute attribute, u32 comp = 0);
     void SetAttribute(Attribute attribute, const F32& value, u32 comp = 0);
 

src/shader_recompiler/ir/passes/flatten_extended_userdata_pass.cpp

@@ -191,7 +191,7 @@ static void VisitPointer(u32 off_dw, IR::Inst* subtree, PassInfo& pass_info,
 static void GenerateSrtProgram(Info& info, PassInfo& pass_info) {
     Xbyak::CodeGenerator& c = g_srt_codegen;
 
-    if (info.srt_info.srt_reservations.empty() && pass_info.srt_roots.empty()) {
+    if (pass_info.srt_roots.empty()) {
         return;
     }
 
@@ -205,29 +205,7 @@ static void GenerateSrtProgram(Info& info, PassInfo& pass_info) {
     }
 
     info.srt_info.walker_func = c.getCurr<PFN_SrtWalker>();
-
     pass_info.dst_off_dw = NumUserDataRegs;
-
-    // Special case for V# step rate buffers in fetch shader
-    for (const auto [sgpr_base, dword_offset, num_dwords] : info.srt_info.srt_reservations) {
-        // get pointer to V#
-        if (sgpr_base != IR::NumScalarRegs) {
-            PushPtr(c, sgpr_base);
-        }
-        u32 src_off = dword_offset << 2;
-
-        for (auto j = 0; j < num_dwords; j++) {
-            c.mov(r11d, ptr[rdi + src_off]);
-            c.mov(ptr[rsi + (pass_info.dst_off_dw << 2)], r11d);
-
-            src_off += 4;
-            ++pass_info.dst_off_dw;
-        }
-        if (sgpr_base != IR::NumScalarRegs) {
-            PopPtr(c);
-        }
-    }
-
     ASSERT(pass_info.dst_off_dw == info.srt_info.flattened_bufsize_dw);
 
     for (const auto& [sgpr_base, root] : pass_info.srt_roots) {

src/shader_recompiler/ir/passes/ring_access_elimination.cpp

@@ -33,12 +33,9 @@ void RingAccessElimination(const IR::Program& program, const RuntimeInfo& runtim
                 bool is_composite = opcode == IR::Opcode::WriteSharedU64;
                 u32 num_components = opcode == IR::Opcode::WriteSharedU32 ? 1 : 2;
 
-                u32 offset = 0;
-                const auto* addr = inst.Arg(0).InstRecursive();
-                if (addr->GetOpcode() == IR::Opcode::IAdd32) {
-                    ASSERT(addr->Arg(1).IsImmediate());
-                    offset = addr->Arg(1).U32();
-                }
+                ASSERT(inst.Arg(0).IsImmediate());
+
+                u32 offset = inst.Arg(0).U32();
                 IR::Value data = is_composite ? ir.UnpackUint2x32(IR::U64{inst.Arg(1).Resolve()})
                                               : inst.Arg(1).Resolve();
                 for (s32 i = 0; i < num_components; i++) {
@@ -116,7 +113,7 @@ void RingAccessElimination(const IR::Program& program, const RuntimeInfo& runtim
                 }
 
                 const auto shl_inst = inst.Arg(1).TryInstRecursive();
-                const auto vertex_id = ir.Imm32(shl_inst->Arg(0).Resolve().U32() >> 2);
+                const auto vertex_id = shl_inst->Arg(0).Resolve().U32() >> 2;
                 const auto offset = inst.Arg(1).TryInstRecursive()->Arg(1);
                 const auto bucket = offset.Resolve().U32() / 256u;
                 const auto attrib = bucket < 4 ? IR::Attribute::Position0

src/shader_recompiler/ir/passes/srt.h

@@ -20,18 +20,7 @@ struct PersistentSrtInfo {
     };
 
     PFN_SrtWalker walker_func{};
-    boost::container::small_vector<SrtSharpReservation, 2> srt_reservations;
     u32 flattened_bufsize_dw = 16; // NumUserDataRegs
-
-    // Special case for fetch shaders because we don't generate IR to read from step rate buffers,
-    // so we won't see usage with GetUserData/ReadConst.
-    // Reserve space in the flattened buffer for a sharp ahead of time
-    u32 ReserveSharp(u32 sgpr_base, u32 dword_offset, u32 num_dwords) {
-        u32 rv = flattened_bufsize_dw;
-        srt_reservations.emplace_back(sgpr_base, dword_offset, num_dwords);
-        flattened_bufsize_dw += num_dwords;
-        return rv;
-    }
 };
 
-} // namespace Shader
+} // namespace Shader