Attempt more optimal codegen for OpCompositeInsert.

Speculate that we can modify the SSA value in-place. As long as it is
not used after the modify, this is fine.

Also need to make sure we don't attempt to RMW something that is
impossible to modify.

1 files changed, 4 insertions, 4 deletions

diff --git a/reference/shaders-ue4/asm/frag/depth-compare.asm.frag b/reference/shaders-ue4/asm/frag/depth-compare.asm.frag
index 5f336c3d..0a6c9841 100644
--- a/reference/shaders-ue4/asm/frag/depth-compare.asm.frag
+++ b/reference/shaders-ue4/asm/frag/depth-compare.asm.frag
@@ -244,8 +244,9 @@ fragment main0_out main0(constant type_View& View [[buffer(0)]], constant type_G
     }
     float _213 = fast::clamp(((_207 - 0.5) * _Globals.ShadowSharpen) + 0.5, 0.0, 1.0);
     float _218 = sqrt(mix(1.0, _213 * _213, _Globals.ShadowFadeFraction));
-    float4 _219 = _107;
+    float4 _219;
     _219.z = _218;
+    float4 _220 = float4(float3(1.0).x, float3(1.0).y, _219.z, float3(1.0).z);
     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;
@@ -307,9 +308,8 @@ fragment main0_out main0(constant type_View& View [[buffer(0)]], constant type_G
     {
         _448 = 1.0;
     }
-    float4 _451 = float4(float3(1.0).x, float3(1.0).y, _219.z, float3(1.0).z);
-    _451.w = _248 ? sqrt(_448) : _218;
-    out.out_var_SV_Target0 = _451;
+    _220.w = _248 ? sqrt(_448) : _218;
+    out.out_var_SV_Target0 = _220;
     return out;
 }