aboutsummaryrefslogtreecommitdiffstats
path: root/erts/emulator/beam/erl_utils.h
diff options
context:
space:
mode:
authorJosé Valim <[email protected]>2016-02-01 10:50:32 +0100
committerJosé Valim <[email protected]>2016-02-15 10:53:05 +0100
commitbc1e16e334afc8255b221e3ffbb6f7c54fb91a2d (patch)
tree5baa13ffe9a1727e66e4fcd48dda1521cf94dd9e /erts/emulator/beam/erl_utils.h
parente038cbe699acc19901e10b9569a3b783e900582d (diff)
downloadotp-bc1e16e334afc8255b221e3ffbb6f7c54fb91a2d.tar.gz
otp-bc1e16e334afc8255b221e3ffbb6f7c54fb91a2d.tar.bz2
otp-bc1e16e334afc8255b221e3ffbb6f7c54fb91a2d.zip
Avoid erts_cmp jump in atom, int and float comparisons
Given the function definition below: check(X) when X >= 0, X <= 20 -> true. @nox has originally noticed that perfoming lt and ge guard tests were performing slower than they should be. Further investigation revealed that most of the cost was in jumping to the erts_cmp function. This patch brings the operations already inlined in erts_cmp into the emulator, removing the jump cost. After applying these changes, invoking the check/1 function defined above 30000 times with different values from 0 to 20 has fallen from 367us to 213us (measured as average of 3 runs). This is a considerably improvement over Erlang 18 which takes 556us on average. Floats have also dropped their time from 1126us (on Erlang 18) to 613us.
Diffstat (limited to 'erts/emulator/beam/erl_utils.h')
-rw-r--r--erts/emulator/beam/erl_utils.h34
1 files changed, 24 insertions, 10 deletions
diff --git a/erts/emulator/beam/erl_utils.h b/erts/emulator/beam/erl_utils.h
index 4058d63eaf..b8746f7edb 100644
--- a/erts/emulator/beam/erl_utils.h
+++ b/erts/emulator/beam/erl_utils.h
@@ -161,7 +161,9 @@ int eq(Eterm, Eterm);
#define EQ(x,y) (((x) == (y)) || (is_not_both_immed((x),(y)) && eq((x),(y))))
+int erts_cmp_atoms(Eterm a, Eterm b);
Sint erts_cmp(Eterm, Eterm, int, int);
+Sint erts_cmp_compound(Eterm, Eterm, int, int);
Sint cmp(Eterm a, Eterm b);
#define CMP(A,B) erts_cmp(A,B,0,0)
#define CMP_TERM(A,B) erts_cmp(A,B,1,0)
@@ -174,17 +176,29 @@ Sint cmp(Eterm a, Eterm b);
#define cmp_ge(a,b) (CMP((a),(b)) >= 0)
#define cmp_gt(a,b) (CMP((a),(b)) > 0)
-#define cmp_lt_term(a,b) (CMP_TERM((a),(b)) < 0)
-#define cmp_le_term(a,b) (CMP_TERM((a),(b)) <= 0)
-#define cmp_ge_term(a,b) (CMP_TERM((a),(b)) >= 0)
-#define cmp_gt_term(a,b) (CMP_TERM((a),(b)) > 0)
-
-#define CMP_LT(a,b) ((a) != (b) && cmp_lt((a),(b)))
-#define CMP_GE(a,b) ((a) == (b) || cmp_ge((a),(b)))
#define CMP_EQ(a,b) ((a) == (b) || cmp_eq((a),(b)))
-#define CMP_NE(a,b) ((a) != (b) && cmp_ne((a),(b)))
-#define CMP_LT_TERM(a,b) ((a) != (b) && cmp_lt_term((a),(b)))
-#define CMP_GE_TERM(a,b) ((a) == (b) || cmp_ge_term((a),(b)))
+#define CMP_EQ_ACTION(X,Y,Action) \
+ if ((X) != (Y)) { CMP_SPEC((X),(Y),!=,Action,1); }
+#define CMP_NE_ACTION(X,Y,Action) \
+ if ((X) == (Y)) { Action; } else { CMP_SPEC((X),(Y),==,Action,1); }
+#define CMP_GE_ACTION(X,Y,Action) \
+ if ((X) != (Y)) { CMP_SPEC((X),(Y),<,Action,0); }
+#define CMP_LT_ACTION(X,Y,Action) \
+ if ((X) == (Y)) { Action; } else { CMP_SPEC((X),(Y),>=,Action,0); }
+
+#define CMP_SPEC(X,Y,Op,Action,EqOnly) \
+ if (is_atom(X) && is_atom(Y)) { \
+ if (erts_cmp_atoms(X, Y) Op 0) { Action; }; \
+ } else if (is_both_small(X, Y)) { \
+ if (signed_val(X) Op signed_val(Y)) { Action; }; \
+ } else if (is_float(X) && is_float(Y)) { \
+ FloatDef af, bf; \
+ GET_DOUBLE(X, af); \
+ GET_DOUBLE(Y, bf); \
+ if (af.fd Op bf.fd) { Action; }; \
+ } else { \
+ if (erts_cmp_compound(X,Y,0,EqOnly) Op 0) { Action; }; \
+ }
#endif