shader: Implement SHFL

16 files changed, 284 insertions, 69 deletions

diff --git a/src/shader_recompiler/CMakeLists.txt b/src/shader_recompiler/CMakeLists.txt
index 028e8b2d2..4161783c8 100644
--- a/src/shader_recompiler/CMakeLists.txt
+++ b/src/shader_recompiler/CMakeLists.txt
@@ -16,7 +16,7 @@ add_library(shader_recompiler STATIC
     backend/spirv/emit_spirv_select.cpp
     backend/spirv/emit_spirv_special.cpp
     backend/spirv/emit_spirv_undefined.cpp
-    backend/spirv/emit_spirv_vote.cpp
+    backend/spirv/emit_spirv_warp.cpp
     environment.h
     exception.h
     file_environment.cpp
@@ -125,6 +125,7 @@ add_library(shader_recompiler STATIC
     frontend/maxwell/translate/impl/texture_fetch.cpp
     frontend/maxwell/translate/impl/texture_fetch_swizzled.cpp
     frontend/maxwell/translate/impl/vote.cpp
+    frontend/maxwell/translate/impl/warp_shuffle.cpp
     frontend/maxwell/translate/translate.cpp
     frontend/maxwell/translate/translate.h
     ir_opt/collect_shader_info_pass.cpp
diff --git a/src/shader_recompiler/backend/spirv/emit_context.cpp b/src/shader_recompiler/backend/spirv/emit_context.cpp
index ea46af244..5db4a9082 100644
--- a/src/shader_recompiler/backend/spirv/emit_context.cpp
+++ b/src/shader_recompiler/backend/spirv/emit_context.cpp
@@ -274,7 +274,8 @@ void EmitContext::DefineInputs(const Info& info) {
     if (info.uses_local_invocation_id) {
         local_invocation_id = DefineInput(*this, U32[3], spv::BuiltIn::LocalInvocationId);
     }
-    if (profile.warp_size_potentially_larger_than_guest && info.uses_subgroup_vote) {
+    if (info.uses_subgroup_invocation_id ||
+        (profile.warp_size_potentially_larger_than_guest && info.uses_subgroup_vote)) {
         subgroup_local_invocation_id =
             DefineInput(*this, U32[1], spv::BuiltIn::SubgroupLocalInvocationId);
     }
diff --git a/src/shader_recompiler/backend/spirv/emit_spirv.cpp b/src/shader_recompiler/backend/spirv/emit_spirv.cpp
index 107403912..cee72f50d 100644
--- a/src/shader_recompiler/backend/spirv/emit_spirv.cpp
+++ b/src/shader_recompiler/backend/spirv/emit_spirv.cpp
@@ -224,7 +224,7 @@ void SetupCapabilities(const Profile& profile, const Info& info, EmitContext& ct
         ctx.AddExtension("SPV_KHR_shader_draw_parameters");
         ctx.AddCapability(spv::Capability::DrawParameters);
     }
-    if (info.uses_subgroup_vote && profile.support_vote) {
+    if ((info.uses_subgroup_vote || info.uses_subgroup_invocation_id) && profile.support_vote) {
         ctx.AddExtension("SPV_KHR_shader_ballot");
         ctx.AddCapability(spv::Capability::SubgroupBallotKHR);
         if (!profile.warp_size_potentially_larger_than_guest) {
@@ -315,4 +315,8 @@ void EmitGetSparseFromOp(EmitContext&) {
     throw LogicError("Unreachable instruction");
 }
 
+void EmitGetInBoundsFromOp(EmitContext&) {
+    throw LogicError("Unreachable instruction");
+}
+
 } // namespace Shader::Backend::SPIRV
diff --git a/src/shader_recompiler/backend/spirv/emit_spirv.h b/src/shader_recompiler/backend/spirv/emit_spirv.h
index 6d4adafc7..a233a4817 100644
--- a/src/shader_recompiler/backend/spirv/emit_spirv.h
+++ b/src/shader_recompiler/backend/spirv/emit_spirv.h
@@ -158,6 +158,7 @@ void EmitGetSignFromOp(EmitContext& ctx);
 void EmitGetCarryFromOp(EmitContext& ctx);
 void EmitGetOverflowFromOp(EmitContext& ctx);
 void EmitGetSparseFromOp(EmitContext& ctx);
+void EmitGetInBoundsFromOp(EmitContext& ctx);
 Id EmitFPAbs16(EmitContext& ctx, Id value);
 Id EmitFPAbs32(EmitContext& ctx, Id value);
 Id EmitFPAbs64(EmitContext& ctx, Id value);
@@ -355,5 +356,13 @@ Id EmitVoteAll(EmitContext& ctx, Id pred);
 Id EmitVoteAny(EmitContext& ctx, Id pred);
 Id EmitVoteEqual(EmitContext& ctx, Id pred);
 Id EmitSubgroupBallot(EmitContext& ctx, Id pred);
+Id EmitShuffleIndex(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp,
+                    Id segmentation_mask);
+Id EmitShuffleUp(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp,
+                 Id segmentation_mask);
+Id EmitShuffleDown(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp,
+                   Id segmentation_mask);
+Id EmitShuffleButterfly(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp,
+                        Id segmentation_mask);
 
 } // namespace Shader::Backend::SPIRV
diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_vote.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_vote.cpp
deleted file mode 100644
index a63677ef2..000000000
--- a/src/shader_recompiler/backend/spirv/emit_spirv_vote.cpp
+++ /dev/null
@@ -1,58 +0,0 @@
-// Copyright 2021 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include "shader_recompiler/backend/spirv/emit_spirv.h"
-
-namespace Shader::Backend::SPIRV {
-namespace {
-Id LargeWarpBallot(EmitContext& ctx, Id ballot) {
-    const Id shift{ctx.Constant(ctx.U32[1], 5)};
-    const Id local_index{ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id)};
-    return ctx.OpVectorExtractDynamic(ctx.U32[1], ballot, local_index);
-}
-} // Anonymous namespace
-
-Id EmitVoteAll(EmitContext& ctx, Id pred) {
-    if (!ctx.profile.warp_size_potentially_larger_than_guest) {
-        return ctx.OpSubgroupAllKHR(ctx.U1, pred);
-    }
-    const Id mask_ballot{ctx.OpSubgroupBallotKHR(ctx.U32[4], ctx.true_value)};
-    const Id active_mask{LargeWarpBallot(ctx, mask_ballot)};
-    const Id ballot{LargeWarpBallot(ctx, ctx.OpSubgroupBallotKHR(ctx.U32[4], pred))};
-    const Id lhs{ctx.OpBitwiseAnd(ctx.U32[1], ballot, active_mask)};
-    return ctx.OpIEqual(ctx.U1, lhs, active_mask);
-}
-
-Id EmitVoteAny(EmitContext& ctx, Id pred) {
-    if (!ctx.profile.warp_size_potentially_larger_than_guest) {
-        return ctx.OpSubgroupAnyKHR(ctx.U1, pred);
-    }
-    const Id mask_ballot{ctx.OpSubgroupBallotKHR(ctx.U32[4], ctx.true_value)};
-    const Id active_mask{LargeWarpBallot(ctx, mask_ballot)};
-    const Id ballot{LargeWarpBallot(ctx, ctx.OpSubgroupBallotKHR(ctx.U32[4], pred))};
-    const Id lhs{ctx.OpBitwiseAnd(ctx.U32[1], ballot, active_mask)};
-    return ctx.OpINotEqual(ctx.U1, lhs, ctx.u32_zero_value);
-}
-
-Id EmitVoteEqual(EmitContext& ctx, Id pred) {
-    if (!ctx.profile.warp_size_potentially_larger_than_guest) {
-        return ctx.OpSubgroupAllEqualKHR(ctx.U1, pred);
-    }
-    const Id mask_ballot{ctx.OpSubgroupBallotKHR(ctx.U32[4], ctx.true_value)};
-    const Id active_mask{LargeWarpBallot(ctx, mask_ballot)};
-    const Id ballot{LargeWarpBallot(ctx, ctx.OpSubgroupBallotKHR(ctx.U32[4], pred))};
-    const Id lhs{ctx.OpBitwiseXor(ctx.U32[1], ballot, active_mask)};
-    return ctx.OpLogicalOr(ctx.U1, ctx.OpIEqual(ctx.U1, lhs, ctx.u32_zero_value),
-                           ctx.OpIEqual(ctx.U1, lhs, active_mask));
-}
-
-Id EmitSubgroupBallot(EmitContext& ctx, Id pred) {
-    const Id ballot{ctx.OpSubgroupBallotKHR(ctx.U32[4], pred)};
-    if (!ctx.profile.warp_size_potentially_larger_than_guest) {
-        return ctx.OpCompositeExtract(ctx.U32[1], ballot, 0U);
-    }
-    return LargeWarpBallot(ctx, ballot);
-}
-
-} // namespace Shader::Backend::SPIRV
diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp
new file mode 100644
index 000000000..44d8a347f
--- /dev/null
+++ b/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp
@@ -0,0 +1,135 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "shader_recompiler/backend/spirv/emit_spirv.h"
+
+namespace Shader::Backend::SPIRV {
+namespace {
+Id LargeWarpBallot(EmitContext& ctx, Id ballot) {
+    const Id shift{ctx.Constant(ctx.U32[1], 5)};
+    const Id local_index{ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id)};
+    return ctx.OpVectorExtractDynamic(ctx.U32[1], ballot, local_index);
+}
+
+void SetInBoundsFlag(IR::Inst* inst, Id result) {
+    IR::Inst* const in_bounds{inst->GetAssociatedPseudoOperation(IR::Opcode::GetInBoundsFromOp)};
+    if (!in_bounds) {
+        return;
+    }
+    in_bounds->SetDefinition(result);
+    in_bounds->Invalidate();
+}
+
+Id ComputeMinThreadId(EmitContext& ctx, Id thread_id, Id segmentation_mask) {
+    return ctx.OpBitwiseAnd(ctx.U32[1], thread_id, segmentation_mask);
+}
+
+Id ComputeMaxThreadId(EmitContext& ctx, Id min_thread_id, Id clamp, Id not_seg_mask) {
+    return ctx.OpBitwiseOr(ctx.U32[1], min_thread_id,
+                           ctx.OpBitwiseAnd(ctx.U32[1], clamp, not_seg_mask));
+}
+
+Id GetMaxThreadId(EmitContext& ctx, Id thread_id, Id clamp, Id segmentation_mask) {
+    const Id not_seg_mask{ctx.OpNot(ctx.U32[1], segmentation_mask)};
+    const Id min_thread_id{ComputeMinThreadId(ctx, thread_id, segmentation_mask)};
+    return ComputeMaxThreadId(ctx, min_thread_id, clamp, not_seg_mask);
+}
+
+Id SelectValue(EmitContext& ctx, Id in_range, Id value, Id src_thread_id) {
+    return ctx.OpSelect(ctx.U32[1], in_range,
+                        ctx.OpSubgroupReadInvocationKHR(ctx.U32[1], value, src_thread_id), value);
+}
+} // Anonymous namespace
+
+Id EmitVoteAll(EmitContext& ctx, Id pred) {
+    if (!ctx.profile.warp_size_potentially_larger_than_guest) {
+        return ctx.OpSubgroupAllKHR(ctx.U1, pred);
+    }
+    const Id mask_ballot{ctx.OpSubgroupBallotKHR(ctx.U32[4], ctx.true_value)};
+    const Id active_mask{LargeWarpBallot(ctx, mask_ballot)};
+    const Id ballot{LargeWarpBallot(ctx, ctx.OpSubgroupBallotKHR(ctx.U32[4], pred))};
+    const Id lhs{ctx.OpBitwiseAnd(ctx.U32[1], ballot, active_mask)};
+    return ctx.OpIEqual(ctx.U1, lhs, active_mask);
+}
+
+Id EmitVoteAny(EmitContext& ctx, Id pred) {
+    if (!ctx.profile.warp_size_potentially_larger_than_guest) {
+        return ctx.OpSubgroupAnyKHR(ctx.U1, pred);
+    }
+    const Id mask_ballot{ctx.OpSubgroupBallotKHR(ctx.U32[4], ctx.true_value)};
+    const Id active_mask{LargeWarpBallot(ctx, mask_ballot)};
+    const Id ballot{LargeWarpBallot(ctx, ctx.OpSubgroupBallotKHR(ctx.U32[4], pred))};
+    const Id lhs{ctx.OpBitwiseAnd(ctx.U32[1], ballot, active_mask)};
+    return ctx.OpINotEqual(ctx.U1, lhs, ctx.u32_zero_value);
+}
+
+Id EmitVoteEqual(EmitContext& ctx, Id pred) {
+    if (!ctx.profile.warp_size_potentially_larger_than_guest) {
+        return ctx.OpSubgroupAllEqualKHR(ctx.U1, pred);
+    }
+    const Id mask_ballot{ctx.OpSubgroupBallotKHR(ctx.U32[4], ctx.true_value)};
+    const Id active_mask{LargeWarpBallot(ctx, mask_ballot)};
+    const Id ballot{LargeWarpBallot(ctx, ctx.OpSubgroupBallotKHR(ctx.U32[4], pred))};
+    const Id lhs{ctx.OpBitwiseXor(ctx.U32[1], ballot, active_mask)};
+    return ctx.OpLogicalOr(ctx.U1, ctx.OpIEqual(ctx.U1, lhs, ctx.u32_zero_value),
+                           ctx.OpIEqual(ctx.U1, lhs, active_mask));
+}
+
+Id EmitSubgroupBallot(EmitContext& ctx, Id pred) {
+    const Id ballot{ctx.OpSubgroupBallotKHR(ctx.U32[4], pred)};
+    if (!ctx.profile.warp_size_potentially_larger_than_guest) {
+        return ctx.OpCompositeExtract(ctx.U32[1], ballot, 0U);
+    }
+    return LargeWarpBallot(ctx, ballot);
+}
+
+Id EmitShuffleIndex(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp,
+                    Id segmentation_mask) {
+    const Id not_seg_mask{ctx.OpNot(ctx.U32[1], segmentation_mask)};
+    const Id thread_id{ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id)};
+    const Id min_thread_id{ComputeMinThreadId(ctx, thread_id, segmentation_mask)};
+    const Id max_thread_id{ComputeMaxThreadId(ctx, min_thread_id, clamp, not_seg_mask)};
+
+    const Id lhs{ctx.OpBitwiseAnd(ctx.U32[1], index, not_seg_mask)};
+    const Id src_thread_id{ctx.OpBitwiseOr(ctx.U32[1], lhs, min_thread_id)};
+    const Id in_range{ctx.OpSLessThanEqual(ctx.U1, src_thread_id, max_thread_id)};
+
+    SetInBoundsFlag(inst, in_range);
+    return SelectValue(ctx, in_range, value, src_thread_id);
+}
+
+Id EmitShuffleUp(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp,
+                 Id segmentation_mask) {
+    const Id thread_id{ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id)};
+    const Id max_thread_id{GetMaxThreadId(ctx, thread_id, clamp, segmentation_mask)};
+    const Id src_thread_id{ctx.OpISub(ctx.U32[1], thread_id, index)};
+    const Id in_range{ctx.OpSGreaterThanEqual(ctx.U1, src_thread_id, max_thread_id)};
+
+    SetInBoundsFlag(inst, in_range);
+    return SelectValue(ctx, in_range, value, src_thread_id);
+}
+
+Id EmitShuffleDown(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp,
+                   Id segmentation_mask) {
+    const Id thread_id{ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id)};
+    const Id max_thread_id{GetMaxThreadId(ctx, thread_id, clamp, segmentation_mask)};
+    const Id src_thread_id{ctx.OpIAdd(ctx.U32[1], thread_id, index)};
+    const Id in_range{ctx.OpSLessThanEqual(ctx.U1, src_thread_id, max_thread_id)};
+
+    SetInBoundsFlag(inst, in_range);
+    return SelectValue(ctx, in_range, value, src_thread_id);
+}
+
+Id EmitShuffleButterfly(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp,
+                        Id segmentation_mask) {
+    const Id thread_id{ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id)};
+    const Id max_thread_id{GetMaxThreadId(ctx, thread_id, clamp, segmentation_mask)};
+    const Id src_thread_id{ctx.OpBitwiseXor(ctx.U32[1], thread_id, index)};
+    const Id in_range{ctx.OpSLessThanEqual(ctx.U1, src_thread_id, max_thread_id)};
+
+    SetInBoundsFlag(inst, in_range);
+    return SelectValue(ctx, in_range, value, src_thread_id);
+}
+
+} // namespace Shader::Backend::SPIRV
diff --git a/src/shader_recompiler/frontend/ir/ir_emitter.cpp b/src/shader_recompiler/frontend/ir/ir_emitter.cpp
index 6280c08f6..418b7f5ac 100644
--- a/src/shader_recompiler/frontend/ir/ir_emitter.cpp
+++ b/src/shader_recompiler/frontend/ir/ir_emitter.cpp
@@ -374,6 +374,10 @@ U1 IREmitter::GetSparseFromOp(const Value& op) {
     return Inst<U1>(Opcode::GetSparseFromOp, op);
 }
 
+U1 IREmitter::GetInBoundsFromOp(const Value& op) {
+    return Inst<U1>(Opcode::GetInBoundsFromOp, op);
+}
+
 F16F32F64 IREmitter::FPAdd(const F16F32F64& a, const F16F32F64& b, FpControl control) {
     if (a.Type() != b.Type()) {
         throw InvalidArgument("Mismatching types {} and {}", a.Type(), b.Type());
@@ -1486,4 +1490,23 @@ U32 IREmitter::SubgroupBallot(const U1& value) {
     return Inst<U32>(Opcode::SubgroupBallot, value);
 }
 
+U32 IREmitter::ShuffleIndex(const IR::U32& value, const IR::U32& index, const IR::U32& clamp,
+                            const IR::U32& seg_mask) {
+    return Inst<U32>(Opcode::ShuffleIndex, value, index, clamp, seg_mask);
+}
+
+U32 IREmitter::ShuffleUp(const IR::U32& value, const IR::U32& index, const IR::U32& clamp,
+                         const IR::U32& seg_mask) {
+    return Inst<U32>(Opcode::ShuffleUp, value, index, clamp, seg_mask);
+}
+
+U32 IREmitter::ShuffleDown(const IR::U32& value, const IR::U32& index, const IR::U32& clamp,
+                           const IR::U32& seg_mask) {
+    return Inst<U32>(Opcode::ShuffleDown, value, index, clamp, seg_mask);
+}
+
+U32 IREmitter::ShuffleButterfly(const IR::U32& value, const IR::U32& index, const IR::U32& clamp,
+                                const IR::U32& seg_mask) {
+    return Inst<U32>(Opcode::ShuffleButterfly, value, index, clamp, seg_mask);
+}
 } // namespace Shader::IR
diff --git a/src/shader_recompiler/frontend/ir/ir_emitter.h b/src/shader_recompiler/frontend/ir/ir_emitter.h
index ebbda78a9..64738735e 100644
--- a/src/shader_recompiler/frontend/ir/ir_emitter.h
+++ b/src/shader_recompiler/frontend/ir/ir_emitter.h
@@ -104,6 +104,7 @@ public:
     [[nodiscard]] U1 GetCarryFromOp(const Value& op);
     [[nodiscard]] U1 GetOverflowFromOp(const Value& op);
     [[nodiscard]] U1 GetSparseFromOp(const Value& op);
+    [[nodiscard]] U1 GetInBoundsFromOp(const Value& op);
 
     [[nodiscard]] Value CompositeConstruct(const Value& e1, const Value& e2);
     [[nodiscard]] Value CompositeConstruct(const Value& e1, const Value& e2, const Value& e3);
@@ -147,7 +148,8 @@ public:
     [[nodiscard]] F32F64 FPRecipSqrt(const F32F64& value);
     [[nodiscard]] F32 FPSqrt(const F32& value);
     [[nodiscard]] F16F32F64 FPSaturate(const F16F32F64& value);
-    [[nodiscard]] F16F32F64 FPClamp(const F16F32F64& value, const F16F32F64& min_value, const F16F32F64& max_value);
+    [[nodiscard]] F16F32F64 FPClamp(const F16F32F64& value, const F16F32F64& min_value,
+                                    const F16F32F64& max_value);
     [[nodiscard]] F16F32F64 FPRoundEven(const F16F32F64& value, FpControl control = {});
     [[nodiscard]] F16F32F64 FPFloor(const F16F32F64& value, FpControl control = {});
     [[nodiscard]] F16F32F64 FPCeil(const F16F32F64& value, FpControl control = {});
@@ -242,6 +244,14 @@ public:
     [[nodiscard]] U1 VoteAny(const U1& value);
     [[nodiscard]] U1 VoteEqual(const U1& value);
     [[nodiscard]] U32 SubgroupBallot(const U1& value);
+    [[nodiscard]] U32 ShuffleIndex(const IR::U32& value, const IR::U32& index, const IR::U32& clamp,
+                                   const IR::U32& seg_mask);
+    [[nodiscard]] U32 ShuffleUp(const IR::U32& value, const IR::U32& index, const IR::U32& clamp,
+                                const IR::U32& seg_mask);
+    [[nodiscard]] U32 ShuffleDown(const IR::U32& value, const IR::U32& index, const IR::U32& clamp,
+                                  const IR::U32& seg_mask);
+    [[nodiscard]] U32 ShuffleButterfly(const IR::U32& value, const IR::U32& index,
+                                       const IR::U32& clamp, const IR::U32& seg_mask);
 
 private:
     IR::Block::iterator insertion_point;
diff --git a/src/shader_recompiler/frontend/ir/microinstruction.cpp b/src/shader_recompiler/frontend/ir/microinstruction.cpp
index ba3968056..be8eb4d4c 100644
--- a/src/shader_recompiler/frontend/ir/microinstruction.cpp
+++ b/src/shader_recompiler/frontend/ir/microinstruction.cpp
@@ -89,6 +89,7 @@ bool Inst::IsPseudoInstruction() const noexcept {
     case Opcode::GetCarryFromOp:
     case Opcode::GetOverflowFromOp:
     case Opcode::GetSparseFromOp:
+    case Opcode::GetInBoundsFromOp:
         return true;
     default:
         return false;
@@ -123,6 +124,9 @@ Inst* Inst::GetAssociatedPseudoOperation(IR::Opcode opcode) {
     case Opcode::GetSparseFromOp:
         CheckPseudoInstruction(associated_insts->sparse_inst, Opcode::GetSparseFromOp);
         return associated_insts->sparse_inst;
+    case Opcode::GetInBoundsFromOp:
+        CheckPseudoInstruction(associated_insts->in_bounds_inst, Opcode::GetInBoundsFromOp);
+        return associated_insts->in_bounds_inst;
     default:
         throw InvalidArgument("{} is not a pseudo-instruction", opcode);
     }
@@ -262,6 +266,10 @@ void Inst::Use(const Value& value) {
         AllocAssociatedInsts(assoc_inst);
         SetPseudoInstruction(assoc_inst->sparse_inst, this);
         break;
+    case Opcode::GetInBoundsFromOp:
+        AllocAssociatedInsts(assoc_inst);
+        SetPseudoInstruction(assoc_inst->in_bounds_inst, this);
+        break;
     default:
         break;
     }
@@ -289,6 +297,10 @@ void Inst::UndoUse(const Value& value) {
         AllocAssociatedInsts(assoc_inst);
         RemovePseudoInstruction(assoc_inst->overflow_inst, Opcode::GetOverflowFromOp);
         break;
+    case Opcode::GetInBoundsFromOp:
+        AllocAssociatedInsts(assoc_inst);
+        RemovePseudoInstruction(assoc_inst->in_bounds_inst, Opcode::GetInBoundsFromOp);
+        break;
     default:
         break;
     }
diff --git a/src/shader_recompiler/frontend/ir/microinstruction.h b/src/shader_recompiler/frontend/ir/microinstruction.h
index d5336c438..770bbd550 100644
--- a/src/shader_recompiler/frontend/ir/microinstruction.h
+++ b/src/shader_recompiler/frontend/ir/microinstruction.h
@@ -134,6 +134,7 @@ static_assert(sizeof(Inst) <= 128, "Inst size unintentionally increased");
 
 struct AssociatedInsts {
     union {
+        Inst* in_bounds_inst;
         Inst* sparse_inst;
         Inst* zero_inst{};
     };
diff --git a/src/shader_recompiler/frontend/ir/opcodes.inc b/src/shader_recompiler/frontend/ir/opcodes.inc
index dd17212a1..a2479c46a 100644
--- a/src/shader_recompiler/frontend/ir/opcodes.inc
+++ b/src/shader_recompiler/frontend/ir/opcodes.inc
@@ -159,6 +159,7 @@ OPCODE(GetSignFromOp,                                       U1,             Opaq
 OPCODE(GetCarryFromOp,                                      U1,             Opaque,                                                                         )
 OPCODE(GetOverflowFromOp,                                   U1,             Opaque,                                                                         )
 OPCODE(GetSparseFromOp,                                     U1,             Opaque,                                                                         )
+OPCODE(GetInBoundsFromOp,                                   U1,             Opaque,                                                                         )
 
 // Floating-point operations
 OPCODE(FPAbs16,                                             F16,            F16,                                                                            )
@@ -363,8 +364,12 @@ OPCODE(ImageSampleExplicitLod,                              F32x4,          U32,
 OPCODE(ImageSampleDrefImplicitLod,                          F32,            U32,            Opaque,         F32,            Opaque,         Opaque,         )
 OPCODE(ImageSampleDrefExplicitLod,                          F32,            U32,            Opaque,         F32,            Opaque,         Opaque,         )
 
-// Vote operations
+// Warp operations
 OPCODE(VoteAll,                                             U1,             U1,                                                                             )
 OPCODE(VoteAny,                                             U1,             U1,                                                                             )
 OPCODE(VoteEqual,                                           U1,             U1,                                                                             )
 OPCODE(SubgroupBallot,                                      U32,            U1,                                                                             )
+OPCODE(ShuffleIndex,                                        U32,            U32,            U32,            U32,            U32,                            )
+OPCODE(ShuffleUp,                                           U32,            U32,            U32,            U32,            U32,                            )
+OPCODE(ShuffleDown,                                         U32,            U32,            U32,            U32,            U32,                            )
+OPCODE(ShuffleButterfly,                                    U32,            U32,            U32,            U32,            U32,                            )
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/integer_scaled_add.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/integer_scaled_add.cpp
index 5469e445a..42fd42bb1 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/integer_scaled_add.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/integer_scaled_add.cpp
@@ -53,8 +53,8 @@ void TranslatorVisitor::ISCADD_reg(u64 insn) {
     ISCADD(*this, insn, GetReg20(insn));
 }
 
-void TranslatorVisitor::ISCADD_cbuf(u64) {
-    throw NotImplementedException("ISCADD (cbuf)");
+void TranslatorVisitor::ISCADD_cbuf(u64 insn) {
+    ISCADD(*this, insn, GetCbuf(insn));
 }
 
 void TranslatorVisitor::ISCADD_imm(u64 insn) {
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp
index c0e36a7e2..3ccd7b925 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp
@@ -301,10 +301,6 @@ void TranslatorVisitor::SETLMEMBASE(u64) {
     ThrowNotImplemented(Opcode::SETLMEMBASE);
 }
 
-void TranslatorVisitor::SHFL(u64) {
-    ThrowNotImplemented(Opcode::SHFL);
-}
-
 void TranslatorVisitor::SSY() {
     // SSY is a no-op
 }
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/warp_shuffle.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/warp_shuffle.cpp
new file mode 100644
index 000000000..550fed55c
--- /dev/null
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/warp_shuffle.cpp
@@ -0,0 +1,69 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <optional>
+
+#include "common/bit_field.h"
+#include "common/common_types.h"
+#include "shader_recompiler/frontend/maxwell/translate/impl/impl.h"
+
+namespace Shader::Maxwell {
+namespace {
+enum class ShuffleMode : u64 {
+    IDX,
+    UP,
+    DOWN,
+    BFLY,
+};
+
+[[nodiscard]] IR::U32 ShuffleOperation(IR::IREmitter& ir, const IR::U32& value,
+                                       const IR::U32& index, const IR::U32& mask,
+                                       ShuffleMode shfl_op) {
+    const IR::U32 clamp{ir.BitFieldExtract(mask, ir.Imm32(0), ir.Imm32(5))};
+    const IR::U32 seg_mask{ir.BitFieldExtract(mask, ir.Imm32(8), ir.Imm32(5))};
+    switch (shfl_op) {
+    case ShuffleMode::IDX:
+        return ir.ShuffleIndex(value, index, clamp, seg_mask);
+    case ShuffleMode::UP:
+        return ir.ShuffleUp(value, index, clamp, seg_mask);
+    case ShuffleMode::DOWN:
+        return ir.ShuffleDown(value, index, clamp, seg_mask);
+    case ShuffleMode::BFLY:
+        return ir.ShuffleButterfly(value, index, clamp, seg_mask);
+    default:
+        throw NotImplementedException("Invalid SHFL op {}", shfl_op);
+    }
+}
+
+void Shuffle(TranslatorVisitor& v, u64 insn, const IR::U32& index, const IR::U32& mask) {
+    union {
+        u64 insn;
+        BitField<0, 8, IR::Reg> dest_reg;
+        BitField<8, 8, IR::Reg> src_reg;
+        BitField<30, 2, ShuffleMode> mode;
+        BitField<48, 3, IR::Pred> pred;
+    } const shfl{insn};
+
+    const IR::U32 result{ShuffleOperation(v.ir, v.X(shfl.src_reg), index, mask, shfl.mode)};
+    v.ir.SetPred(shfl.pred, v.ir.GetInBoundsFromOp(result));
+    v.X(shfl.dest_reg, result);
+}
+} // Anonymous namespace
+
+void TranslatorVisitor::SHFL(u64 insn) {
+    union {
+        u64 insn;
+        BitField<20, 5, u64> src_a_imm;
+        BitField<28, 1, u64> src_a_flag;
+        BitField<29, 1, u64> src_b_flag;
+        BitField<34, 13, u64> src_b_imm;
+    } const flags{insn};
+    const IR::U32 src_a{flags.src_a_flag != 0 ? ir.Imm32(static_cast<u32>(flags.src_a_imm))
+                                              : GetReg20(insn)};
+    const IR::U32 src_b{flags.src_b_flag != 0 ? ir.Imm32(static_cast<u32>(flags.src_b_imm))
+                                              : GetReg39(insn)};
+    Shuffle(*this, insn, src_a, src_b);
+}
+
+} // namespace Shader::Maxwell
diff --git a/src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp b/src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp
index 32f276f3b..61cc314c7 100644
--- a/src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp
+++ b/src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp
@@ -307,6 +307,12 @@ void VisitUsages(Info& info, IR::Inst& inst) {
     case IR::Opcode::LocalInvocationId:
         info.uses_local_invocation_id = true;
         break;
+    case IR::Opcode::ShuffleIndex:
+    case IR::Opcode::ShuffleUp:
+    case IR::Opcode::ShuffleDown:
+    case IR::Opcode::ShuffleButterfly:
+        info.uses_subgroup_invocation_id = true;
+        break;
     case IR::Opcode::GetCbufU8:
     case IR::Opcode::GetCbufS8:
     case IR::Opcode::GetCbufU16:
diff --git a/src/shader_recompiler/shader_info.h b/src/shader_recompiler/shader_info.h
index 3d9f04d1a..27e61a5f9 100644
--- a/src/shader_recompiler/shader_info.h
+++ b/src/shader_recompiler/shader_info.h
@@ -56,6 +56,7 @@ struct Info {
 
     bool uses_workgroup_id{};
     bool uses_local_invocation_id{};
+    bool uses_subgroup_invocation_id{};
 
     std::array<bool, 32> loads_generics{};
     bool loads_position{};