- use in-place bignum operations where available

- runs approx. 30% faster

R=r
DELTA=24  (10 added, 2 deleted, 12 changed)
OCL=32984
CL=33005

diff --git a/test/bench/pidigits.go b/test/bench/pidigits.go
index 6e1e7e05342cca40e2e121575bb0a627b9b22004..b02c6e79efc14e757f28c3666e597d347fda3f0a 100644 (file)
--- a/test/bench/pidigits.go
+++ b/test/bench/pidigits.go
@@ -44,8 +44,6 @@ import (
 )
 
 var n = flag.Int("n", 27, "number of digits");
-
-// TODO for easier profiling, remove eventually
 var silent = flag.Bool("s", false, "don't print result");
 
 var (
@@ -61,13 +59,16 @@ func extract_digit() int64 {
                return -1;
        }
 
-       /* Compute (numer * 3 + accum) / denom */
-       tmp1, tmp2 = numer.MulNat(bignum.Nat(3)).Add(accum).QuoRem(denom);
+       // Compute (numer * 3 + accum) / denom
+       tmp1 = numer.Shl(1);
+       bignum.Iadd(tmp1, tmp1, numer);
+       bignum.Iadd(tmp1, tmp1, accum);
+       tmp1, tmp2 := tmp1.QuoRem(denom);
 
-       /* Now, if (numer * 4 + accum) % denom... */
-       tmp2 = tmp2.Add(numer);
+       // Now, if (numer * 4 + accum) % denom...
+       bignum.Iadd(tmp2, tmp2, numer);
 
-       /* ... is normalized, then the two divisions have the same result.  */
+       // ... is normalized, then the two divisions have the same result.
        if tmp2.Cmp(denom) >= 0 {
                return -1;
        }
@@ -79,16 +80,16 @@ func next_term(k int64) {
        y2 := k*2 + 1;
 
        tmp1 = numer.Shl(1);
-       accum = accum.Add(tmp1);
-       accum = accum.Mul1(y2);
-       numer = numer.Mul1(k);
-       denom = denom.Mul1(y2);
+       bignum.Iadd(accum, accum, tmp1);
+       bignum.Iscale(accum, y2);
+       bignum.Iscale(numer, k);
+       bignum.Iscale(denom, y2);
 }
 
 func eliminate_digit(d int64) {
-       accum = accum.Sub(denom.Mul1(d));
-       accum = accum.Mul1(10);
-       numer = numer.Mul1(10);
+       bignum.Isub(accum, accum, denom.Mul1(d));
+       bignum.Iscale(accum, 10);
+       bignum.Iscale(numer, 10);
 }
 
 func printf(s string, arg ...) {

diff --git a/test/bench/timing.log b/test/bench/timing.log
index e73d061b19afd0f7e6176d251886fbd4fc250b0f..75c92f26ee474f3b81e13a3906e486372e144cd9 100644 (file)
--- a/test/bench/timing.log
+++ b/test/bench/timing.log
@@ -231,3 +231,10 @@ chameneos 6000000
        gcc -O2 chameneosredux.c -lpthread      17.93u 323.65s 88.47r
        gc chameneosredux       21.72u 0.00s 21.73r
 
+August 10 2009
+
+# In-place versions for some bignum operations.
+pidigits 10000
+       gcc -O2 pidigits.c -lgmp        2.56u 0.00s 2.57r
+       gc pidigits     55.22u 0.04s 55.29r     # *** -23%
+       gc_B pidigits   55.49u 0.02s 55.60r     # *** -23%