MSL: Selectively enable fast-math in MSL code to match Vulkan CTS results.

Based on CTS testing, math optimizations between MSL and Vulkan are inconsistent.
In some cases, enabling MSL's fast-math compilation option matches Vulkan's math
results. In other cases, disabling it does. Broadly enabling or disabling fast-math
across all shaders results in some CTS test failures either way.

To fix this, selectively enable/disable fast-math optimizations in the MSL code,
using metal::fast and metal::precise function namespaces, where supported, and
the [[clang::optnone]] function attribute otherwise.

Adjust SPIRV-Cross unit test reference shaders to accommodate these changes.

4 files changed, 11 insertions, 11 deletions

diff --git a/reference/shaders-ue4/asm/frag/depth-compare.asm.frag b/reference/shaders-ue4/asm/frag/depth-compare.asm.frag
index bdeccc2f..9ee2d89b 100644
--- a/reference/shaders-ue4/asm/frag/depth-compare.asm.frag
+++ b/reference/shaders-ue4/asm/frag/depth-compare.asm.frag
@@ -246,7 +246,7 @@ fragment main0_out main0(constant type_View& View [[buffer(0)]], constant type_G
     float _218 = sqrt(mix(1.0, _213 * _213, _Globals.ShadowFadeFraction));
     float4 _219 = _107;
     _219.z = _218;
-    float3 _236 = normalize((SceneTexturesStruct_GBufferATexture.sample(SceneTexturesStruct_GBufferATextureSampler, _114, level(0.0)).xyz * float3(2.0)) - float3(1.0));
+    float3 _236 = fast::normalize((SceneTexturesStruct_GBufferATexture.sample(SceneTexturesStruct_GBufferATextureSampler, _114, level(0.0)).xyz * float3(2.0)) - float3(1.0));
     uint _240 = uint(round(SceneTexturesStruct_GBufferBTexture.sample(SceneTexturesStruct_GBufferBTextureSampler, _114, level(0.0)).w * 255.0));
     bool _248 = (_240 & 15u) == 5u;
     float _448;
@@ -259,7 +259,7 @@ fragment main0_out main0(constant type_View& View [[buffer(0)]], constant type_G
         if (_160)
         {
             float3 _278 = _152 / float3(_158);
-            float3 _280 = normalize(cross(_278, float3(0.0, 0.0, 1.0)));
+            float3 _280 = fast::normalize(cross(_278, float3(0.0, 0.0, 1.0)));
             float3 _284 = float3(_Globals.InvShadowmapResolution);
             float3 _285 = _280 * _284;
             float3 _286 = cross(_280, _278) * _284;
diff --git a/reference/shaders-ue4/asm/frag/global-constant-arrays.asm.frag b/reference/shaders-ue4/asm/frag/global-constant-arrays.asm.frag
index 210d4f98..c3d26e65 100644
--- a/reference/shaders-ue4/asm/frag/global-constant-arrays.asm.frag
+++ b/reference/shaders-ue4/asm/frag/global-constant-arrays.asm.frag
@@ -176,7 +176,7 @@ fragment main0_out main0(main0_in in [[stage_in]], constant type_Globals& _Globa
     float _644 = ((0.317398726940155029296875 + (4.25 * _Globals.WhiteTemp)) + ((4.2048167614439080352894961833954e-08 * _Globals.WhiteTemp) * _Globals.WhiteTemp)) / ((1.0 - (2.8974181986995972692966461181641e-05 * _Globals.WhiteTemp)) + ((1.6145605741257895715534687042236e-07 * _Globals.WhiteTemp) * _Globals.WhiteTemp));
     float _649 = ((2.0 * _633) - (8.0 * _644)) + 4.0;
     float2 _653 = float2((3.0 * _633) / _649, (2.0 * _644) / _649);
-    float2 _660 = normalize(float2(_633, _644));
+    float2 _660 = fast::normalize(float2(_633, _644));
     float _665 = _633 + (((-_660.y) * _Globals.WhiteTint) * 0.0500000007450580596923828125);
     float _669 = _644 + ((_660.x * _Globals.WhiteTint) * 0.0500000007450580596923828125);
     float _674 = ((2.0 * _665) - (8.0 * _669)) + 4.0;
@@ -273,7 +273,7 @@ fragment main0_out main0(main0_in in [[stage_in]], constant type_Globals& _Globa
     }
     else
     {
-        _990 = 57.2957763671875 * atan2(sqrt(3.0) * (_974 - _976), ((2.0 * _973) - _974) - _976);
+        _990 = 57.2957763671875 * precise::atan2(sqrt(3.0) * (_974 - _976), ((2.0 * _973) - _974) - _976);
     }
     float _995;
     if (_990 < 0.0)
@@ -436,7 +436,7 @@ fragment main0_out main0(main0_in in [[stage_in]], constant type_Globals& _Globa
                 }
                 else
                 {
-                    _2174 = 57.2957763671875 * atan2(sqrt(3.0) * (_2158 - _2160), ((2.0 * _2157) - _2158) - _2160);
+                    _2174 = 57.2957763671875 * precise::atan2(sqrt(3.0) * (_2158 - _2160), ((2.0 * _2157) - _2158) - _2160);
                 }
                 float _2179;
                 if (_2174 < 0.0)
@@ -909,7 +909,7 @@ fragment main0_out main0(main0_in in [[stage_in]], constant type_Globals& _Globa
                     }
                     else
                     {
-                        _1337 = 57.2957763671875 * atan2(sqrt(3.0) * (_1321 - _1323), ((2.0 * _1320) - _1321) - _1323);
+                        _1337 = 57.2957763671875 * precise::atan2(sqrt(3.0) * (_1321 - _1323), ((2.0 * _1320) - _1321) - _1323);
                     }
                     float _1342;
                     if (_1337 < 0.0)
diff --git a/reference/shaders-ue4/asm/frag/padded-float-array-member-defef.asm.frag b/reference/shaders-ue4/asm/frag/padded-float-array-member-defef.asm.frag
index 67fa21aa..83857f22 100644
--- a/reference/shaders-ue4/asm/frag/padded-float-array-member-defef.asm.frag
+++ b/reference/shaders-ue4/asm/frag/padded-float-array-member-defef.asm.frag
@@ -178,7 +178,7 @@ fragment main0_out main0(main0_in in [[stage_in]], constant type_Globals& _Globa
     float _670 = ((0.317398726940155029296875 + (4.25 * _Globals.WhiteTemp)) + ((4.2048167614439080352894961833954e-08 * _Globals.WhiteTemp) * _Globals.WhiteTemp)) / ((1.0 - (2.8974181986995972692966461181641e-05 * _Globals.WhiteTemp)) + ((1.6145605741257895715534687042236e-07 * _Globals.WhiteTemp) * _Globals.WhiteTemp));
     float _675 = ((2.0 * _659) - (8.0 * _670)) + 4.0;
     float2 _679 = float2((3.0 * _659) / _675, (2.0 * _670) / _675);
-    float2 _686 = normalize(float2(_659, _670));
+    float2 _686 = fast::normalize(float2(_659, _670));
     float _691 = _659 + (((-_686.y) * _Globals.WhiteTint) * 0.0500000007450580596923828125);
     float _695 = _670 + ((_686.x * _Globals.WhiteTint) * 0.0500000007450580596923828125);
     float _700 = ((2.0 * _691) - (8.0 * _695)) + 4.0;
@@ -275,7 +275,7 @@ fragment main0_out main0(main0_in in [[stage_in]], constant type_Globals& _Globa
     }
     else
     {
-        _1016 = 57.2957763671875 * atan2(sqrt(3.0) * (_1000 - _1002), ((2.0 * _999) - _1000) - _1002);
+        _1016 = 57.2957763671875 * precise::atan2(sqrt(3.0) * (_1000 - _1002), ((2.0 * _999) - _1000) - _1002);
     }
     float _1021;
     if (_1016 < 0.0)
@@ -485,7 +485,7 @@ fragment main0_out main0(main0_in in [[stage_in]], constant type_Globals& _Globa
                 }
                 else
                 {
-                    _2278 = 57.2957763671875 * atan2(sqrt(3.0) * (_2262 - _2264), ((2.0 * _2261) - _2262) - _2264);
+                    _2278 = 57.2957763671875 * precise::atan2(sqrt(3.0) * (_2262 - _2264), ((2.0 * _2261) - _2262) - _2264);
                 }
                 float _2283;
                 if (_2278 < 0.0)
@@ -958,7 +958,7 @@ fragment main0_out main0(main0_in in [[stage_in]], constant type_Globals& _Globa
                     }
                     else
                     {
-                        _1443 = 57.2957763671875 * atan2(sqrt(3.0) * (_1427 - _1429), ((2.0 * _1426) - _1427) - _1429);
+                        _1443 = 57.2957763671875 * precise::atan2(sqrt(3.0) * (_1427 - _1429), ((2.0 * _1426) - _1427) - _1429);
                     }
                     float _1448;
                     if (_1443 < 0.0)
diff --git a/reference/shaders-ue4/asm/frag/sample-mask-not-array.asm.frag b/reference/shaders-ue4/asm/frag/sample-mask-not-array.asm.frag
index 9404aa66..834517a8 100644
--- a/reference/shaders-ue4/asm/frag/sample-mask-not-array.asm.frag
+++ b/reference/shaders-ue4/asm/frag/sample-mask-not-array.asm.frag
@@ -462,7 +462,7 @@ fragment main0_out main0(main0_in in [[stage_in]], constant type_View& View [[bu
     {
         _215 = in.in_var_TEXCOORD7;
     }
-    float3 _216 = fast::max(Material.Material_VectorExpressions[1].xyz * float3(((1.0 + dot(float3(-1.0, -1.5, 3.0) / float3(sqrt(12.25)), normalize(float3x3(in.in_var_TEXCOORD10_centroid.xyz, cross(in.in_var_TEXCOORD11_centroid.xyz, in.in_var_TEXCOORD10_centroid.xyz) * float3(in.in_var_TEXCOORD11_centroid.w), in.in_var_TEXCOORD11_centroid.xyz) * normalize((float3(0.0, 0.0, 1.0) * float3(View.View_NormalOverrideParameter.w)) + View.View_NormalOverrideParameter.xyz)))) * 0.5) + 0.20000000298023223876953125), float3(0.0));
+    float3 _216 = fast::max(Material.Material_VectorExpressions[1].xyz * float3(((1.0 + dot(float3(-1.0, -1.5, 3.0) / float3(sqrt(12.25)), fast::normalize(float3x3(in.in_var_TEXCOORD10_centroid.xyz, cross(in.in_var_TEXCOORD11_centroid.xyz, in.in_var_TEXCOORD10_centroid.xyz) * float3(in.in_var_TEXCOORD11_centroid.w), in.in_var_TEXCOORD11_centroid.xyz) * fast::normalize((float3(0.0, 0.0, 1.0) * float3(View.View_NormalOverrideParameter.w)) + View.View_NormalOverrideParameter.xyz)))) * 0.5) + 0.20000000298023223876953125), float3(0.0));
     float3 _246;
     if (View.View_OutOfBoundsMask > 0.0)
     {