From 6c51f496320f698e123207c09ca61e55180a31b5 Mon Sep 17 00:00:00 2001
From: ameerj <52414509+ameerj@users.noreply.github.com>
Date: Sun, 28 Mar 2021 22:23:45 -0400
Subject: shader: Implement FSWZADD

---
 src/shader_recompiler/backend/spirv/emit_context.cpp    |  8 ++++++++
 src/shader_recompiler/backend/spirv/emit_context.h      |  2 ++
 src/shader_recompiler/backend/spirv/emit_spirv.h        |  1 +
 src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp | 16 ++++++++++++++++
 4 files changed, 27 insertions(+)

(limited to 'src/shader_recompiler/backend')

diff --git a/src/shader_recompiler/backend/spirv/emit_context.cpp b/src/shader_recompiler/backend/spirv/emit_context.cpp
index 96d0e9b4d..7531f8b21 100644
--- a/src/shader_recompiler/backend/spirv/emit_context.cpp
+++ b/src/shader_recompiler/backend/spirv/emit_context.cpp
@@ -393,6 +393,14 @@ void EmitContext::DefineInputs(const Info& info) {
         subgroup_local_invocation_id =
             DefineInput(*this, U32[1], spv::BuiltIn::SubgroupLocalInvocationId);
     }
+    if (info.uses_fswzadd) {
+        const Id f32_one{Constant(F32[1], 1.0f)};
+        const Id f32_minus_one{Constant(F32[1], -1.0f)};
+        const Id f32_zero{Constant(F32[1], 0.0f)};
+        fswzadd_lut_a = ConstantComposite(F32[4], f32_minus_one, f32_one, f32_minus_one, f32_zero);
+        fswzadd_lut_b =
+            ConstantComposite(F32[4], f32_minus_one, f32_minus_one, f32_one, f32_minus_one);
+    }
     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 1a4e8221a..ffac39c4f 100644
--- a/src/shader_recompiler/backend/spirv/emit_context.h
+++ b/src/shader_recompiler/backend/spirv/emit_context.h
@@ -103,6 +103,8 @@ public:
     Id vertex_index{};
     Id base_vertex{};
     Id front_face{};
+    Id fswzadd_lut_a{};
+    Id fswzadd_lut_b{};
 
     Id local_memory{};
 
diff --git a/src/shader_recompiler/backend/spirv/emit_spirv.h b/src/shader_recompiler/backend/spirv/emit_spirv.h
index 02648d769..3d0c6f7ba 100644
--- a/src/shader_recompiler/backend/spirv/emit_spirv.h
+++ b/src/shader_recompiler/backend/spirv/emit_spirv.h
@@ -397,5 +397,6 @@ Id EmitShuffleDown(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clam
                    Id segmentation_mask);
 Id EmitShuffleButterfly(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp,
                         Id segmentation_mask);
+Id EmitFSwizzleAdd(EmitContext& ctx, Id op_a, Id op_b, Id swizzle);
 
 } // 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
index 44d8a347f..cbc5b1c96 100644
--- a/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp
+++ b/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp
@@ -132,4 +132,20 @@ Id EmitShuffleButterfly(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id
     return SelectValue(ctx, in_range, value, src_thread_id);
 }
 
+Id EmitFSwizzleAdd(EmitContext& ctx, Id op_a, Id op_b, Id swizzle) {
+    const Id three{ctx.Constant(ctx.U32[1], 3)};
+    Id mask{ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id)};
+    mask = ctx.OpBitwiseAnd(ctx.U32[1], mask, three);
+    mask = ctx.OpShiftLeftLogical(ctx.U32[1], mask, ctx.Constant(ctx.U32[1], 1));
+    mask = ctx.OpShiftRightLogical(ctx.U32[1], swizzle, mask);
+    mask = ctx.OpBitwiseAnd(ctx.U32[1], mask, three);
+
+    const Id modifier_a{ctx.OpVectorExtractDynamic(ctx.F32[1], ctx.fswzadd_lut_a, mask)};
+    const Id modifier_b{ctx.OpVectorExtractDynamic(ctx.F32[1], ctx.fswzadd_lut_b, mask)};
+
+    const Id result_a{ctx.OpFMul(ctx.F32[1], op_a, modifier_a)};
+    const Id result_b{ctx.OpFMul(ctx.F32[1], op_b, modifier_b)};
+    return ctx.OpFAdd(ctx.F32[1], result_a, result_b);
+}
+
 } // namespace Shader::Backend::SPIRV
-- 
cgit v1.2.3