Dawn/Tint: Polyfill reflect vec2<f32> for D3D12 FXC on Intel

This CL add a toggle-controlled Tint polyfill for reflect on vec2<f32>,
and enable this toggle by default on D3D12 Intel device when using FXC.
This CL works around issue tint:1798.

Bug: tint:1798
Change-Id: If2f4de836eaf5e7374bc2c1ae3fbe06b91a5bbd5
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/121160
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Zhaoming Jiang <zhaoming.jiang@intel.com>
diff --git a/src/dawn/native/Toggles.cpp b/src/dawn/native/Toggles.cpp
index 72f59a1..39e1fc2 100644
--- a/src/dawn/native/Toggles.cpp
+++ b/src/dawn/native/Toggles.cpp
@@ -387,6 +387,11 @@
       "This toggle is off by default. It is expected to turn on or get removed when WebGPU V1 "
       "ships and stays stable.",
       "https://crbug.com/dawn/1563", ToggleStage::Device}},
+    {Toggle::D3D12PolyfillReflectVec2F32,
+     {"d3d12_polyfill_reflect_vec2_f32",
+      "Polyfill the reflect builtin for vec2<f32> for D3D12. This toggle is enabled by default on "
+      "D3D12 backends using FXC on Intel GPUs due to a driver issue on Intel D3D12 driver.",
+      "https://crbug.com/tint/1798", ToggleStage::Device}},
     {Toggle::NoWorkaroundSampleMaskBecomesZeroForAllButLastColorTarget,
      {"no_workaround_sample_mask_becomes_zero_for_all_but_last_color_target",
       "MacOS 12.0+ Intel has a bug where the sample mask is only applied for the last color "
diff --git a/src/dawn/native/Toggles.h b/src/dawn/native/Toggles.h
index 536a3fb..6d75fff 100644
--- a/src/dawn/native/Toggles.h
+++ b/src/dawn/native/Toggles.h
@@ -92,6 +92,7 @@
     UseBlitForDepthTextureToTextureCopyToNonzeroSubresource,
     D3D12ReplaceAddWithMinusWhenDstFactorIsZeroAndSrcFactorIsDstAlpha,
     DisallowDeprecatedAPIs,
+    D3D12PolyfillReflectVec2F32,
 
     // Unresolved issues.
     NoWorkaroundSampleMaskBecomesZeroForAllButLastColorTarget,
diff --git a/src/dawn/native/d3d12/AdapterD3D12.cpp b/src/dawn/native/d3d12/AdapterD3D12.cpp
index d05061e..8fd7993 100644
--- a/src/dawn/native/d3d12/AdapterD3D12.cpp
+++ b/src/dawn/native/d3d12/AdapterD3D12.cpp
@@ -595,6 +595,12 @@
         deviceToggles->ForceSet(
             Toggle::D3D12UseTempBufferInTextureToTextureCopyBetweenDifferentDimensions, true);
     }
+
+    // Polyfill reflect builtin for vec2<f32> on Intel device in usng FXC.
+    // See https://crbug.com/tint/1798 for more information.
+    if (gpu_info::IsIntel(vendorId) && !deviceToggles->IsEnabled(Toggle::UseDXC)) {
+        deviceToggles->Default(Toggle::D3D12PolyfillReflectVec2F32, true);
+    }
 }
 
 ResultOrError<Ref<DeviceBase>> Adapter::CreateDeviceImpl(const DeviceDescriptor* descriptor,
diff --git a/src/dawn/native/d3d12/ShaderModuleD3D12.cpp b/src/dawn/native/d3d12/ShaderModuleD3D12.cpp
index d821488..6b8d28f 100644
--- a/src/dawn/native/d3d12/ShaderModuleD3D12.cpp
+++ b/src/dawn/native/d3d12/ShaderModuleD3D12.cpp
@@ -95,6 +95,7 @@
     X(bool, disableSymbolRenaming)                                                          \
     X(bool, isRobustnessEnabled)                                                            \
     X(bool, disableWorkgroupInit)                                                           \
+    X(bool, polyfillReflectVec2F32)                                                         \
     X(bool, dumpShaders)
 
 #define D3D_BYTECODE_COMPILATION_REQUEST_MEMBERS(X) \
@@ -401,6 +402,8 @@
         options.interstage_locations = r.interstageLocations;
     }
 
+    options.polyfill_reflect_vec2_f32 = r.polyfillReflectVec2F32;
+
     TRACE_EVENT0(tracePlatform.UnsafeGetValue(), General, "tint::writer::hlsl::Generate");
     auto result = tint::writer::hlsl::Generate(&transformedProgram, options);
     DAWN_INVALID_IF(!result.success, "An error occured while generating HLSL: %s", result.error);
@@ -606,6 +609,8 @@
     req.hlsl.arrayLengthFromUniform = std::move(arrayLengthFromUniform);
     req.hlsl.substituteOverrideConfig = std::move(substituteOverrideConfig);
 
+    req.hlsl.polyfillReflectVec2F32 = device->IsToggleEnabled(Toggle::D3D12PolyfillReflectVec2F32);
+
     const CombinedLimits& limits = device->GetLimits();
     req.hlsl.limits = LimitsForCompilationRequest::Create(limits.v1);
 
diff --git a/src/tint/transform/builtin_polyfill.cc b/src/tint/transform/builtin_polyfill.cc
index 606c6ce..259eb11 100644
--- a/src/tint/transform/builtin_polyfill.cc
+++ b/src/tint/transform/builtin_polyfill.cc
@@ -582,6 +582,33 @@
         return name;
     }
 
+    /// Builds the polyfill function for the `reflect` builtin
+    /// @param ty the parameter and return type for the function
+    /// @return the polyfill function name
+    Symbol reflect(const type::Type* ty) {
+        auto name = b.Symbols().New("tint_reflect");
+
+        // WGSL polyfill function:
+        //      fn tint_reflect(e1 : T, e2 : T) -> T {
+        //          let factor = (-2.0 * dot(e1, e2));
+        //          return (e1 + (factor * e2));
+        //      }
+        // Using -2.0 instead of 2.0 in factor to prevent the optimization that cause wrong result.
+        // See https://crbug.com/tint/1798 for more details.
+        auto body = utils::Vector{
+            b.Decl(b.Let("factor", b.Mul(-2.0_a, b.Call("dot", "e1", "e2")))),
+            b.Return(b.Add("e1", b.Mul("factor", "e2"))),
+        };
+        b.Func(name,
+               utils::Vector{
+                   b.Param("e1", T(ty)),
+                   b.Param("e2", T(ty)),
+               },
+               T(ty), body);
+
+        return name;
+    }
+
     /// Builds the polyfill function for the `saturate` builtin
     /// @param ty the parameter and return type for the function
     /// @return the polyfill function name
@@ -1007,6 +1034,18 @@
                                 builtin, [&] { return s.insertBits(builtin->ReturnType()); });
                         }
                         break;
+                    case sem::BuiltinType::kReflect:
+                        // Only polyfill for vec2<f32>. See https://crbug.com/tint/1798 for more
+                        // details.
+                        if (polyfill.reflect_vec2_f32) {
+                            auto& sig = builtin->Signature();
+                            auto* vec = sig.return_type->As<type::Vector>();
+                            if (vec && vec->Width() == 2 && vec->type()->Is<type::F32>()) {
+                                fn = builtin_polyfills.GetOrCreate(
+                                    builtin, [&] { return s.reflect(builtin->ReturnType()); });
+                            }
+                        }
+                        break;
                     case sem::BuiltinType::kSaturate:
                         if (polyfill.saturate) {
                             fn = builtin_polyfills.GetOrCreate(
diff --git a/src/tint/transform/builtin_polyfill.h b/src/tint/transform/builtin_polyfill.h
index 521940c..b070248 100644
--- a/src/tint/transform/builtin_polyfill.h
+++ b/src/tint/transform/builtin_polyfill.h
@@ -70,6 +70,8 @@
         bool int_div_mod = false;
         /// Should float modulos be polyfilled to emit a precise modulo operation as per the spec?
         bool precise_float_mod = false;
+        /// Should `reflect()` be polyfilled for vec2<f32>?
+        bool reflect_vec2_f32 = false;
         /// Should `saturate()` be polyfilled?
         bool saturate = false;
         /// Should `sign()` be polyfilled for integer types?
diff --git a/src/tint/transform/builtin_polyfill_test.cc b/src/tint/transform/builtin_polyfill_test.cc
index 09d17b8..8c0d787 100644
--- a/src/tint/transform/builtin_polyfill_test.cc
+++ b/src/tint/transform/builtin_polyfill_test.cc
@@ -3045,6 +3045,177 @@
 }
 
 ////////////////////////////////////////////////////////////////////////////////
+// reflect for vec2<f32>
+////////////////////////////////////////////////////////////////////////////////
+DataMap polyfillReflectVec2F32() {
+    BuiltinPolyfill::Builtins builtins;
+    builtins.reflect_vec2_f32 = true;
+    DataMap data;
+    data.Add<BuiltinPolyfill::Config>(builtins);
+    return data;
+}
+
+TEST_F(BuiltinPolyfillTest, ShouldRunReflect_vec2_f32) {
+    auto* src = R"(
+fn f() {
+  let e1 = vec2<f32>(1.0f);
+  let e2 = vec2<f32>(1.0f);
+  let x = reflect(e1, e2);
+}
+)";
+
+    EXPECT_FALSE(ShouldRun<BuiltinPolyfill>(src));
+    EXPECT_TRUE(ShouldRun<BuiltinPolyfill>(src, polyfillReflectVec2F32()));
+}
+
+TEST_F(BuiltinPolyfillTest, ShouldRunReflect_vec2_f16) {
+    auto* src = R"(
+enable f16;
+
+fn f() {
+  let e1 = vec2<f16>(1.0h);
+  let e2 = vec2<f16>(1.0h);
+  let x = reflect(e1, e2);
+}
+)";
+
+    EXPECT_FALSE(ShouldRun<BuiltinPolyfill>(src));
+    EXPECT_FALSE(ShouldRun<BuiltinPolyfill>(src, polyfillReflectVec2F32()));
+}
+
+TEST_F(BuiltinPolyfillTest, ShouldRunReflect_vec3_f32) {
+    auto* src = R"(
+fn f() {
+  let e1 = vec3<f32>(1.0f);
+  let e2 = vec3<f32>(1.0f);
+  let x = reflect(e1, e2);
+}
+)";
+
+    EXPECT_FALSE(ShouldRun<BuiltinPolyfill>(src));
+    EXPECT_FALSE(ShouldRun<BuiltinPolyfill>(src, polyfillReflectVec2F32()));
+}
+
+TEST_F(BuiltinPolyfillTest, ShouldRunReflect_vec3_f16) {
+    auto* src = R"(
+enable f16;
+
+fn f() {
+  let e1 = vec3<f16>(1.0h);
+  let e2 = vec3<f16>(1.0h);
+  let x = reflect(e1, e2);
+}
+)";
+
+    EXPECT_FALSE(ShouldRun<BuiltinPolyfill>(src));
+    EXPECT_FALSE(ShouldRun<BuiltinPolyfill>(src, polyfillReflectVec2F32()));
+}
+
+TEST_F(BuiltinPolyfillTest, ShouldRunReflect_vec4_f32) {
+    auto* src = R"(
+fn f() {
+  let e1 = vec3<f32>(1.0f);
+  let e2 = vec3<f32>(1.0f);
+  let x = reflect(e1, e2);
+}
+)";
+
+    EXPECT_FALSE(ShouldRun<BuiltinPolyfill>(src));
+    EXPECT_FALSE(ShouldRun<BuiltinPolyfill>(src, polyfillReflectVec2F32()));
+}
+
+TEST_F(BuiltinPolyfillTest, ShouldRunReflect_vec4_f16) {
+    auto* src = R"(
+enable f16;
+
+fn f() {
+  let e1 = vec3<f16>(1.0h);
+  let e2 = vec3<f16>(1.0h);
+  let x = reflect(e1, e2);
+}
+)";
+
+    EXPECT_FALSE(ShouldRun<BuiltinPolyfill>(src));
+    EXPECT_FALSE(ShouldRun<BuiltinPolyfill>(src, polyfillReflectVec2F32()));
+}
+
+TEST_F(BuiltinPolyfillTest, Reflect_ConstantExpression) {
+    auto* src = R"(
+fn f() {
+  let r : vec2<f32> = reflect(vec2<f32>(1.0), vec2<f32>(1.0));
+}
+)";
+
+    EXPECT_FALSE(ShouldRun<BuiltinPolyfill>(src, polyfillReflectVec2F32()));
+}
+
+TEST_F(BuiltinPolyfillTest, Reflect_vec2_f32) {
+    auto* src = R"(
+fn f() {
+  let v = 0.5f;
+  let r : vec2<f32> = reflect(vec2<f32>(v), vec2<f32>(v));
+}
+)";
+
+    auto* expect = R"(
+fn tint_reflect(e1 : vec2<f32>, e2 : vec2<f32>) -> vec2<f32> {
+  let factor = (-2.0 * dot(e1, e2));
+  return (e1 + (factor * e2));
+}
+
+fn f() {
+  let v = 0.5f;
+  let r : vec2<f32> = tint_reflect(vec2<f32>(v), vec2<f32>(v));
+}
+)";
+
+    auto got = Run<BuiltinPolyfill>(src, polyfillReflectVec2F32());
+
+    EXPECT_EQ(expect, str(got));
+}
+
+TEST_F(BuiltinPolyfillTest, Reflect_multiple_types) {
+    auto* src = R"(
+enable f16;
+
+fn f() {
+  let in_f32 = 0.5f;
+  let out_f32_vec2 : vec2<f32> = reflect(vec2<f32>(in_f32), vec2<f32>(in_f32));
+  let out_f32_vec3 : vec3<f32> = reflect(vec3<f32>(in_f32), vec3<f32>(in_f32));
+  let out_f32_vec4 : vec4<f32> = reflect(vec4<f32>(in_f32), vec4<f32>(in_f32));
+  let in_f16 = 0.5h;
+  let out_f16_vec2 : vec2<f16> = reflect(vec2<f16>(in_f16), vec2<f16>(in_f16));
+  let out_f16_vec3 : vec3<f16> = reflect(vec3<f16>(in_f16), vec3<f16>(in_f16));
+  let out_f16_vec4 : vec4<f16> = reflect(vec4<f16>(in_f16), vec4<f16>(in_f16));
+}
+)";
+
+    auto* expect = R"(
+enable f16;
+
+fn tint_reflect(e1 : vec2<f32>, e2 : vec2<f32>) -> vec2<f32> {
+  let factor = (-2.0 * dot(e1, e2));
+  return (e1 + (factor * e2));
+}
+
+fn f() {
+  let in_f32 = 0.5f;
+  let out_f32_vec2 : vec2<f32> = tint_reflect(vec2<f32>(in_f32), vec2<f32>(in_f32));
+  let out_f32_vec3 : vec3<f32> = reflect(vec3<f32>(in_f32), vec3<f32>(in_f32));
+  let out_f32_vec4 : vec4<f32> = reflect(vec4<f32>(in_f32), vec4<f32>(in_f32));
+  let in_f16 = 0.5h;
+  let out_f16_vec2 : vec2<f16> = reflect(vec2<f16>(in_f16), vec2<f16>(in_f16));
+  let out_f16_vec3 : vec3<f16> = reflect(vec3<f16>(in_f16), vec3<f16>(in_f16));
+  let out_f16_vec4 : vec4<f16> = reflect(vec4<f16>(in_f16), vec4<f16>(in_f16));
+}
+)";
+
+    auto got = Run<BuiltinPolyfill>(src, polyfillReflectVec2F32());
+
+    EXPECT_EQ(expect, str(got));
+}
+
+////////////////////////////////////////////////////////////////////////////////
 // saturate
 ////////////////////////////////////////////////////////////////////////////////
 DataMap polyfillSaturate() {
diff --git a/src/tint/writer/hlsl/generator.h b/src/tint/writer/hlsl/generator.h
index c624943..80dd518 100644
--- a/src/tint/writer/hlsl/generator.h
+++ b/src/tint/writer/hlsl/generator.h
@@ -61,6 +61,8 @@
     /// Interstage locations actually used as inputs in the next stage of the pipeline.
     /// This is potentially used for truncating unused interstage outputs at current shader stage.
     std::bitset<16> interstage_locations;
+    /// Set to `true` to generate polyfill for `reflect` builtin for vec2<f32>
+    bool polyfill_reflect_vec2_f32 = false;
 
     /// Reflect the fields of this class so that it can be used by tint::ForeachField()
     TINT_REFLECT(root_constant_binding_point,
diff --git a/src/tint/writer/hlsl/generator_impl.cc b/src/tint/writer/hlsl/generator_impl.cc
index 787962b..14f1d0c 100644
--- a/src/tint/writer/hlsl/generator_impl.cc
+++ b/src/tint/writer/hlsl/generator_impl.cc
@@ -181,6 +181,7 @@
         polyfills.insert_bits = transform::BuiltinPolyfill::Level::kFull;
         polyfills.int_div_mod = true;
         polyfills.precise_float_mod = true;
+        polyfills.reflect_vec2_f32 = options.polyfill_reflect_vec2_f32;
         polyfills.texture_sample_base_clamp_to_edge_2d_f32 = true;
         polyfills.workgroup_uniform_load = true;
         data.Add<transform::BuiltinPolyfill::Config>(polyfills);
diff --git a/webgpu-cts/expectations.txt b/webgpu-cts/expectations.txt
index 38fbcca..c1adddb 100644
--- a/webgpu-cts/expectations.txt
+++ b/webgpu-cts/expectations.txt
@@ -233,9 +233,6 @@
 crbug.com/tint/1796 [ intel-gen-9 ubuntu ] webgpu:shader,execution,expression,call,builtin,asin:f32:inputSource="uniform";vectorize=4 [ Failure ]
 crbug.com/tint/1796 [ intel-gen-9 win10 ] webgpu:shader,execution,expression,call,builtin,asin:f32:inputSource="uniform";vectorize=4 [ Failure ]
 crbug.com/tint/1796 [ nvidia-0x2184 win10 ] webgpu:shader,execution,expression,call,builtin,asin:f32:inputSource="uniform";vectorize=4 [ Failure ]
-crbug.com/tint/1798 [ intel-gen-9 win10 ] webgpu:shader,execution,expression,call,builtin,reflect:f32_vec2:inputSource="storage_r" [ Failure ]
-crbug.com/tint/1798 [ intel-gen-9 win10 ] webgpu:shader,execution,expression,call,builtin,reflect:f32_vec2:inputSource="storage_rw" [ Failure ]
-crbug.com/tint/1798 [ intel-gen-9 win10 ] webgpu:shader,execution,expression,call,builtin,reflect:f32_vec2:inputSource="uniform" [ Failure ]
 crbug.com/tint/1801 [ nvidia-0x2184 target-cpu-32 ] webgpu:shader,execution,shader_io,compute_builtins:inputs:method="mixed";dispatch="indirect";groupSize={"x":1,"y":1,"z":1};numGroups={"x":1,"y":1,"z":1} [ Failure ]
 crbug.com/tint/1801 [ nvidia-0x2184 target-cpu-32 ] webgpu:shader,execution,shader_io,compute_builtins:inputs:method="mixed";dispatch="indirect";groupSize={"x":1,"y":1,"z":1};numGroups={"x":8,"y":4,"z":2} [ Failure ]
 crbug.com/tint/1801 [ nvidia-0x2184 target-cpu-32 ] webgpu:shader,execution,shader_io,compute_builtins:inputs:method="mixed";dispatch="indirect";groupSize={"x":3,"y":7,"z":5};numGroups={"x":13,"y":9,"z":11} [ Failure ]