Faster _fmpz_vec_scalar_divexact_ui

src/fmpz_mpoly/test/t-scalar_divexact_fmpz.c

@@ -92,7 +92,7 @@ TEST_FUNCTION_START(fmpz_mpoly_scalar_divexact_fmpz, state)
           fmpz_mpoly_randtest_bits(f, state, len, coeff_bits, exp_bits, ctx);
           fmpz_mpoly_randtest_bits(h, state, len, coeff_bits, exp_bits, ctx);
 
-          fmpz_randtest(c, state, n_randint(state, 200));
+          fmpz_randtest_not_zero(c, state, n_randint(state, 200) + 1);
 
           fmpz_mpoly_scalar_mul_fmpz(f, f, c, ctx);
 

src/fmpz_vec/profile/p-divexact_ui.c

@@ -0,0 +1,60 @@
+#include "ulong_extras.h"
+#include "fmpz.h"
+#include "fmpz_vec.h"
+#include "profiler.h"
+
+void
+_fmpz_vec_scalar_divexact_ui_naive(fmpz * vec1, const fmpz * vec2,
+                               slong len2, ulong c)
+{
+    slong i;
+    for (i = 0; i < len2; i++)
+        fmpz_divexact_ui(vec1 + i, vec2 + i, c);
+}
+
+int main()
+{
+    slong len, bits;
+
+    flint_rand_t state;
+    flint_rand_init(state);
+
+    fmpz *A, *Ac, *B;
+    ulong c;
+
+    double t1, t2, FLINT_SET_BUT_UNUSED(__);
+
+    flint_printf("   len    bits(A*c)  bits(c)      old       new    speedup\n");
+
+    for (len = 1; len <= 10; len++)
+    {
+        for (bits = 5; bits <= 10000; bits *= 2)
+        {
+            A = _fmpz_vec_init(len);
+            Ac = _fmpz_vec_init(len);
+            B = _fmpz_vec_init(len);
+
+            _fmpz_vec_randtest(A, state, len, bits);
+            c = n_randtest_not_zero(state);
+            _fmpz_vec_scalar_mul_ui(Ac, A, len, c);
+
+            TIMEIT_START
+            _fmpz_vec_scalar_divexact_ui_naive(B, Ac, len, c);
+            TIMEIT_STOP_VALUES(__, t1)
+            TIMEIT_START
+            _fmpz_vec_scalar_divexact_ui(B, Ac, len, c);
+            TIMEIT_STOP_VALUES(__, t2)
+
+            flint_printf("%6wd    %6wd   %6wd    %8g   %8g   %.3f\n",
+                    len, _fmpz_vec_max_bits(Ac, len), FLINT_BIT_COUNT(c),
+                    t1, t2, t1 / t2);
+
+            _fmpz_vec_clear(A, len);
+            _fmpz_vec_clear(Ac, len);
+            _fmpz_vec_clear(B, len);
+        }
+    }
+
+    flint_rand_clear(state);
+}
+

src/fmpz_vec/scalar.c

@@ -126,46 +126,6 @@ _fmpz_vec_scalar_addmul_ui(fmpz * vec1, const fmpz * vec2, slong len2, ulong c)
         fmpz_addmul_ui(vec1 + ix, vec2 + ix, c);
 }
 
-void
-_fmpz_vec_scalar_divexact_fmpz(fmpz * vec1, const fmpz * vec2,
-                               slong len2, const fmpz_t x)
-{
-    fmpz c = *x;
-
-    if (!COEFF_IS_MPZ(c))
-    {
-        if (c == 1)
-            _fmpz_vec_set(vec1, vec2, len2);
-        else if (c == -1)
-            _fmpz_vec_neg(vec1, vec2, len2);
-        else
-            _fmpz_vec_scalar_divexact_si(vec1, vec2, len2, c);
-    }
-    else
-    {
-        slong i;
-        for (i = 0; i < len2; i++)
-            fmpz_divexact(vec1 + i, vec2 + i, x);
-    }
-}
-
-void
-_fmpz_vec_scalar_divexact_si(fmpz * vec1, const fmpz * vec2, slong len2, slong c)
-{
-    slong i;
-    for (i = 0; i < len2; i++)
-        fmpz_divexact_si(vec1 + i, vec2 + i, c);
-}
-
-void
-_fmpz_vec_scalar_divexact_ui(fmpz * vec1, const fmpz * vec2,
-                             slong len2, ulong c)
-{
-    slong i;
-    for (i = 0; i < len2; i++)
-        fmpz_divexact_ui(vec1 + i, vec2 + i, c);
-}
-
 void
 _fmpz_vec_scalar_fdiv_q_2exp(fmpz * vec1, const fmpz * vec2, slong len2,
                              ulong exp)
@@ -209,6 +169,8 @@ _fmpz_vec_scalar_fdiv_r_2exp(fmpz * vec1, const fmpz * vec2, slong len2,
         fmpz_fdiv_r_2exp(vec1 + i, vec2 + i, exp);
 }
 
+/* todo: preinvert p when appropriate; or better, make sure we don't
+   call this function in places where we have an fmpz_mod */
 void _fmpz_vec_scalar_mod_fmpz(fmpz *res, const fmpz *vec, slong len, const fmpz_t p)
 {
     slong i;
@@ -267,6 +229,7 @@ _fmpz_vec_scalar_mul_ui(fmpz * vec1, const fmpz * vec2, slong len2, ulong c)
         fmpz_mul_ui(vec1 + i, vec2 + i, c);
 }
 
+/* todo: preinvert p when appropriate */
 void _fmpz_vec_scalar_smod_fmpz(fmpz *res, const fmpz *vec, slong len, const fmpz_t p)
 {
     slong i;