varsamp: build the phase table without malloc0's memset, read h forward

calc_varsamp() gained a transpose of the coefficient table, which showed up
as ~0.3 ms on create_varsamp()/setInRate_varsamp() (1.0 ms -> 1.4 ms). It is
a one-off setup cost, not on the sample path, but it is easy to trim: every
element of hp is written, so malloc0()'s memset of ~1 MB is dead, and h is
cold straight out of fir_bandpass(), so walk it along its fast axis and let
the prefetcher work.

Output is bit-identical; xvarsamp() is unchanged at ~34 us/buffer.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

varsamp.c

@@ -67,11 +67,15 @@ void calc_varsamp (VARSAMP a)
 		   line.  Transposing makes each phase contiguous; hshift() interpolates
 		   between phases hidx and hidx+1, hence R+1 of them. */
 		int p, m;
-		double* h = fir_bandpass(a->ncoef, fc_norm_low, fc_norm_high, (double)a->R, 1, 0, (double)a->R * a->gain);
+		const int R = a->R, rsize = a->rsize;
-		a->hp = (double *)malloc0 ((size_t)(a->R + 1) * a->rsize * sizeof (double));
+		double* h = fir_bandpass(a->ncoef, fc_norm_low, fc_norm_high, (double)R, 1, 0, (double)R * a->gain);
-		for (p = 0; p <= a->R; p++)
+		// every element is written below, so skip malloc0()'s memset of ~1 MB
-			for (m = 0; m < a->rsize; m++)
+		a->hp = (double *)_aligned_malloc ((size_t)(R + 1) * rsize * sizeof (double), 16);
-				a->hp[(size_t)p * a->rsize + m] = h[p + (size_t)m * a->R];
+		// walk h forward (p is its fast axis) so the prefetcher sees a linear
+		// stream; h is cold here, straight from fir_bandpass()
+		for (m = 0; m < rsize; m++)
+			for (p = 0; p <= R; p++)
+				a->hp[(size_t)p * rsize + m] = h[p + (size_t)m * R];
 		_aligned_free (h);
 	}
 	a->ringI = (double *)malloc0(a->rsize * sizeof(double));