Jib: Fix optimizations for SMT IR changes

Fixes C backend optimizations that were disabled due to changes in the
IR while working on the SMT generation.

Also add a -Oaarch64_fast option that optimizes any integer within a
struct to be an int64_t, which is safe for the ARM v8.5 spec and
improves performance significantly (reduces Linux boot times by 4-5
minutes). Eventually this should probably be a directive that can be
attached to any arbitrary struct/type.

Fixes the -c_specialize option for ARM v8.5. However this only gives a
very small performance improvment for a very large increase in
compilation time however.

2 files changed, 49 insertions, 7 deletions

diff --git a/lib/sail.c b/lib/sail.c
index 2d47939e..cc4466bb 100644
--- a/lib/sail.c
+++ b/lib/sail.c
@@ -1017,15 +1017,11 @@ void slice(lbits *rop, const lbits op, const sail_int start_mpz, const sail_int
   }
 }
 
-inline
+__attribute__((target ("bmi2")))
 sbits sslice(const fbits op, const mach_int start, const mach_int len)
 {
   sbits rop;
-#ifdef INTRINSICS
   rop.bits = _bzhi_u64(op >> start, len);
-#else
-  rop.bits = (op >> start) & safe_rshift(UINT64_MAX, 64 - len);
-#endif
   rop.len = len;
   return rop;
 }
@@ -1126,18 +1122,25 @@ void reverse_endianness(lbits *rop, const lbits op)
   }
 }
 
-inline
 bool eq_sbits(const sbits op1, const sbits op2)
 {
   return op1.bits == op2.bits;
 }
 
-inline
 bool neq_sbits(const sbits op1, const sbits op2)
 {
   return op1.bits != op2.bits;
 }
 
+__attribute__((target ("bmi2")))
+sbits not_sbits(const sbits op)
+{
+  sbits rop;
+  rop.bits = (~op.bits) & _bzhi_u64(UINT64_MAX, op.len);
+  rop.len = op.len;
+  return rop;
+}
+
 sbits xor_sbits(const sbits op1, const sbits op2)
 {
   sbits rop;
@@ -1146,6 +1149,40 @@ sbits xor_sbits(const sbits op1, const sbits op2)
   return rop;
 }
 
+sbits or_sbits(const sbits op1, const sbits op2)
+{
+  sbits rop;
+  rop.bits = op1.bits | op2.bits;
+  rop.len = op1.len;
+  return rop;
+}
+
+sbits and_sbits(const sbits op1, const sbits op2)
+{
+  sbits rop;
+  rop.bits = op1.bits & op2.bits;
+  rop.len = op1.len;
+  return rop;
+}
+
+__attribute__((target ("bmi2")))
+sbits add_sbits(const sbits op1, const sbits op2)
+{
+  sbits rop;
+  rop.bits = (op1.bits + op2.bits) & _bzhi_u64(UINT64_MAX, op1.len);
+  rop.len = op1.len;
+  return rop;
+}
+
+__attribute__((target ("bmi2")))
+sbits sub_sbits(const sbits op1, const sbits op2)
+{
+  sbits rop;
+  rop.bits = (op1.bits - op2.bits) & _bzhi_u64(UINT64_MAX, op1.len);
+  rop.len = op1.len;
+  return rop;
+}
+
 /* ***** Sail Reals ***** */
 
 void CREATE(real)(real *rop)
diff --git a/lib/sail.h b/lib/sail.h
index 05c65027..258b7a91 100644
--- a/lib/sail.h
+++ b/lib/sail.h
@@ -333,7 +333,12 @@ void reverse_endianness(lbits*, lbits);
 
 bool eq_sbits(const sbits op1, const sbits op2);
 bool neq_sbits(const sbits op1, const sbits op2);
+sbits not_sbits(const sbits op);
 sbits xor_sbits(const sbits op1, const sbits op2);
+sbits or_sbits(const sbits op1, const sbits op2);
+sbits and_sbits(const sbits op1, const sbits op2);
+sbits add_sbits(const sbits op1, const sbits op2);
+sbits sub_sbits(const sbits op1, const sbits op2);
 
 /* ***** Sail reals ***** */