shader: Implement VOTE

5 files changed, 77 insertions, 1 deletions

diff --git a/src/shader_recompiler/backend/spirv/emit_context.cpp b/src/shader_recompiler/backend/spirv/emit_context.cpp
index 4a4de3676..36f130781 100644
--- a/src/shader_recompiler/backend/spirv/emit_context.cpp
+++ b/src/shader_recompiler/backend/spirv/emit_context.cpp
@@ -259,6 +259,10 @@ void EmitContext::DefineInputs(const Info& info, Stage stage) {
     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) {
+        subgroup_local_invocation_id =
+            DefineInput(*this, U32[1], spv::BuiltIn::SubgroupLocalInvocationId);
+    }
     if (info.loads_position) {
         const bool is_fragment{stage != Stage::Fragment};
         const spv::BuiltIn built_in{is_fragment ? spv::BuiltIn::Position : spv::BuiltIn::FragCoord};
diff --git a/src/shader_recompiler/backend/spirv/emit_context.h b/src/shader_recompiler/backend/spirv/emit_context.h
index 9b9e0d6b1..6e64360bf 100644
--- a/src/shader_recompiler/backend/spirv/emit_context.h
+++ b/src/shader_recompiler/backend/spirv/emit_context.h
@@ -82,6 +82,7 @@ public:
 
     Id workgroup_id{};
     Id local_invocation_id{};
+    Id subgroup_local_invocation_id{};
     Id instance_id{};
     Id instance_index{};
     Id base_instance{};
@@ -96,7 +97,7 @@ public:
     std::array<Id, 32> output_generics{};
 
     std::array<Id, 8> frag_color{};
-    Id frag_depth {};
+    Id frag_depth{};
 
     std::vector<Id> interfaces;
 
diff --git a/src/shader_recompiler/backend/spirv/emit_spirv.cpp b/src/shader_recompiler/backend/spirv/emit_spirv.cpp
index 93e851133..107403912 100644
--- a/src/shader_recompiler/backend/spirv/emit_spirv.cpp
+++ b/src/shader_recompiler/backend/spirv/emit_spirv.cpp
@@ -224,6 +224,15 @@ 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) {
+        ctx.AddExtension("SPV_KHR_shader_ballot");
+        ctx.AddCapability(spv::Capability::SubgroupBallotKHR);
+        if (!profile.warp_size_potentially_larger_than_guest) {
+            // vote ops are only used when not taking the long path
+            ctx.AddExtension("SPV_KHR_subgroup_vote");
+            ctx.AddCapability(spv::Capability::SubgroupVoteKHR);
+        }
+    }
     // TODO: Track this usage
     ctx.AddCapability(spv::Capability::ImageGatherExtended);
 }
diff --git a/src/shader_recompiler/backend/spirv/emit_spirv.h b/src/shader_recompiler/backend/spirv/emit_spirv.h
index 960d022ff..ce23200f2 100644
--- a/src/shader_recompiler/backend/spirv/emit_spirv.h
+++ b/src/shader_recompiler/backend/spirv/emit_spirv.h
@@ -346,5 +346,9 @@ Id EmitImageSampleDrefImplicitLod(EmitContext& ctx, IR::Inst* inst, const IR::Va
                                   Id coords, Id dref, Id bias_lc, Id offset);
 Id EmitImageSampleDrefExplicitLod(EmitContext& ctx, IR::Inst* inst, const IR::Value& index,
                                   Id coords, Id dref, Id lod_lc, Id offset);
+Id EmitVoteAll(EmitContext& ctx, Id pred);
+Id EmitVoteAny(EmitContext& ctx, Id pred);
+Id EmitVoteEqual(EmitContext& ctx, Id pred);
+Id EmitSubgroupBallot(EmitContext& ctx, Id pred);
 
 } // 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
new file mode 100644
index 000000000..a63677ef2
--- /dev/null
+++ b/src/shader_recompiler/backend/spirv/emit_spirv_vote.cpp
@@ -0,0 +1,58 @@
+// 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