diff --git a/GPU/Software/Rasterizer.cpp b/GPU/Software/Rasterizer.cpp
index b760364994..56abea7897 100644
--- a/GPU/Software/Rasterizer.cpp
+++ b/GPU/Software/Rasterizer.cpp
@@ -1159,7 +1159,7 @@ void DrawTriangleSlice(
 						int32x4_t sec = vsetq_lane_s32(0, sec_color[i].ivec, 3);
 						prim_color[i].ivec = vaddq_s32(prim_color[i].ivec, sec);
 #else
-						prim_color[i] = Vec4<int>(sec_color[i], 0);
+						prim_color[i] += Vec4<int>(sec_color[i], 0);
 #endif
 					}
 				}
diff --git a/test.py b/test.py
index c455d97224..ef639a0ebd 100755
--- a/test.py
+++ b/test.py
@@ -512,17 +512,9 @@ known_failures = {
     # The ISA returns the canonical NaN (0x7fc00000) from every operation, never the operand's
     # NaN, so a negative or signaling NaN input loses its sign and payload. Everything else passes.
     "cpu/fpu/roundmode",
-    # The software renderer's output differs from the reference by the same amount on both of
-    # these architectures, despite them using completely different SIMD paths. Unexplained.
-    "gpu/clipping/homogeneous",
-    "gpu/commands/cull",
-    "gpu/primitives/triangles",
   ],
   "loongarch64": [
     "cpu/fpu/fpu",
-    "gpu/clipping/homogeneous",
-    "gpu/commands/cull",
-    "gpu/primitives/triangles",
   ],
 }