Fix vsubs.u

Fix the overflow cap of unsigned saturated subtraction.

Bug: 296483050
Change-Id: If675412b6b873b1dc6554c5e7a7941d9329a6e74

hdl/chisel/src/kelvin/vector/VAluInt.scala

diff --git a/hdl/chisel/src/kelvin/vector/VAluInt.scala b/hdl/chisel/src/kelvin/vector/VAluInt.scala
index 619f5d1..3f81781 100644
--- a/hdl/chisel/src/kelvin/vector/VAluInt.scala
+++ b/hdl/chisel/src/kelvin/vector/VAluInt.scala
@@ -971,7 +971,7 @@
       val satsubsel =
         Cat( signed && satsubmsb === 2.U,  // vsub.s -ve
              signed && satsubmsb === 1.U,  // vsub.s +ve
-            !signed && satsubmsb(1))       // vsub.su +ve
+            !signed && satsubmsb(1))       // vsub.su 0
       assert(PopCount(satsubsel) <= 1.U)
 
       val rsubtr = rsub_b(m,l) - rsub_a(m,l)
@@ -1330,7 +1330,7 @@
 
       val sub = MuxOR(satsubsel(2) && io.op.sub.subs, Cat(1.U(1.W), 0.U((size - 1).W))) |
                 MuxOR(satsubsel(1) && io.op.sub.subs, ~0.U((size - 1).W)) |
-                MuxOR(satsubsel(0) && io.op.sub.subs, ~0.U(size.W)) |
+                MuxOR(satsubsel(0) && io.op.sub.subs, 0.U(size.W)) |
                 MuxOR(satsubsel === 0.U && io.op.sub.subs || io.op.sub.sub, subtr(size - 1, 0)) |
                 MuxOR(io.op.sub.hsub, subtr(size, 1))
 

tests/verilator_sim/kelvin/alu_ref.h

diff --git a/tests/verilator_sim/kelvin/alu_ref.h b/tests/verilator_sim/kelvin/alu_ref.h
index 82d950f..6ab9372 100644
--- a/tests/verilator_sim/kelvin/alu_ref.h
+++ b/tests/verilator_sim/kelvin/alu_ref.h
@@ -422,9 +422,7 @@
                           std::numeric_limits<T>::max());
     return m;
   }
-  uint64_t m = static_cast<uint64_t>(a) - static_cast<uint64_t>(b);
-  m = std::min<uint64_t>(m, std::numeric_limits<T>::max());
-  return m;
+  return a < b ? 0 : a - b;
 }
 
 template <typename T>