diff --git a/src/include/decaf_255_config.h b/src/include/decaf_255_config.h
index be9d978..01de56a 100644
--- a/src/include/decaf_255_config.h
+++ b/src/include/decaf_255_config.h
@@ -20,7 +20,7 @@
  * kilobytes at the cost of perhaps 5-10% overhead in direct scalarmul
  * time.
  */
-#define DECAF_USE_MONTGOMERY_LADDER 1
+#define DECAF_USE_MONTGOMERY_LADDER 0 /* FUTURE */
 
 /** The number of comb tables for fixed base scalarmul. */
 #define DECAF_COMBS_N 3
diff --git a/src/p25519/arch_x86_64/p25519.c b/src/p25519/arch_x86_64/p25519.c
index 5959b0d..8d9044a 100644
--- a/src/p25519/arch_x86_64/p25519.c
+++ b/src/p25519/arch_x86_64/p25519.c
@@ -14,10 +14,6 @@ p255_mul (
     const uint64_t *a = as->limb, *b = bs->limb, mask = ((1ull<<51)-1);
     uint64_t *c = cs->limb;
     
-    uint64_t bh[4];
-    int i;
-    for (i=0; i<4; i++) bh[i] = b[i+1] * 19;
-    
     __uint128_t accum0, accum1, accum2;
     
     uint64_t ai = a[0];
@@ -26,24 +22,26 @@ p255_mul (
     accum2 = widemul_rm(ai, &b[2]);
     
     ai = a[1];
-    mac_rm(&accum0, ai, &bh[3]);
-    mac_rm(&accum1, ai, &b[0]);
     mac_rm(&accum2, ai, &b[1]);
+    mac_rm(&accum1, ai, &b[0]);
+    ai *= 19;
+    mac_rm(&accum0, ai, &b[4]);
     
     ai = a[2];
-    mac_rm(&accum0, ai, &bh[2]);
-    mac_rm(&accum1, ai, &bh[3]);
     mac_rm(&accum2, ai, &b[0]);
+    ai *= 19;
+    mac_rm(&accum0, ai, &b[3]);
+    mac_rm(&accum1, ai, &b[4]);
     
-    ai = a[3];
-    mac_rm(&accum0, ai, &bh[1]);
-    mac_rm(&accum1, ai, &bh[2]);
-    mac_rm(&accum2, ai, &bh[3]);
+    ai = a[3] * 19;
+    mac_rm(&accum0, ai, &b[2]);
+    mac_rm(&accum1, ai, &b[3]);
+    mac_rm(&accum2, ai, &b[4]);
     
-    ai = a[4];
-    mac_rm(&accum0, ai, &bh[0]);
-    mac_rm(&accum1, ai, &bh[1]);
-    mac_rm(&accum2, ai, &bh[2]);
+    ai = a[4] * 19;
+    mac_rm(&accum0, ai, &b[1]);
+    mac_rm(&accum1, ai, &b[2]);
+    mac_rm(&accum2, ai, &b[3]);
     
     uint64_t c0 = accum0 & mask;
     accum1 += accum0 >> 51;
@@ -52,6 +50,8 @@ p255_mul (
     c[2] = accum2 & mask;
     
     accum0 = accum2 >> 51;
+
+    mac_rm(&accum0, ai, &b[4]);
     
     ai = a[0];
     mac_rm(&accum0, ai, &b[3]);
@@ -70,7 +70,6 @@ p255_mul (
     mac_rm(&accum1, ai, &b[1]);
     
     ai = a[4];
-    mac_rm(&accum0, ai, &bh[3]);
     mac_rm(&accum1, ai, &b[0]);
     
     c[3] = accum0 & mask;
@@ -88,33 +87,6 @@ p255_mul (
     c[1] = c1 + (accum1>>51);
 }
 
-void
-p255_mulw (
-    p255_t *__restrict__ cs,
-    const p255_t *as,
-    uint64_t b
-) {
-    const uint64_t *a = as->limb, mask = ((1ull<<51)-1);
-    int i;
-    
-    uint64_t *c = cs->limb;
-
-    __uint128_t accum = 0;
-    for (i=0; i<5; i++) {
-        mac_rm(&accum, b, &a[i]);
-        c[i] = accum & mask;
-        accum >>= 51;
-    }
-    /* PERF: parallelize? eh well this is reference */
-    accum *= 19;
-    accum += c[0];
-    c[0] = accum & mask;
-    accum >>= 51;
-    
-    assert(accum < mask);
-    c[1] += accum;
-}
-
 void
 p255_sqr (
     p255_t *__restrict__ cs,
@@ -123,10 +95,6 @@ p255_sqr (
     const uint64_t *a = as->limb, mask = ((1ull<<51)-1);
     uint64_t *c = cs->limb;
     
-    uint64_t ah[4];
-    int i;
-    for (i=0; i<4; i++) ah[i] = a[i+1] * 19;
-    
     __uint128_t accum0, accum1, accum2;
     
     uint64_t ai = a[0];
@@ -137,17 +105,17 @@ p255_sqr (
     
     ai = a[1];
     mac_rr(&accum2, ai, ai);
-    ai *= 2;
-    mac_rm(&accum0, ai, &ah[3]);
+    ai *= 38;
+    mac_rm(&accum0, ai, &a[4]);
     
-    ai = a[2] * 2;
-    mac_rm(&accum0, ai, &ah[2]);
-    mac_rm(&accum1, ai, &ah[3]);
+    ai = a[2] * 38;
+    mac_rm(&accum0, ai, &a[3]);
+    mac_rm(&accum1, ai, &a[4]);
     
-    ai = a[3];
-    mac_rm(&accum1, ai, &ah[2]);
+    ai = a[3] * 19;
+    mac_rm(&accum1, ai, &a[3]);
     ai *= 2;
-    mac_rm(&accum2, ai, &ah[3]);
+    mac_rm(&accum2, ai, &a[4]);
     
     uint64_t c0 = accum0 & mask;
     accum1 += accum0 >> 51;
@@ -165,7 +133,7 @@ p255_sqr (
     mac_rm(&accum0, ai, &a[2]);
     mac_rm(&accum1, ai, &a[3]);
     
-    mac_rm(&accum0, a[4], &ah[3]);
+    mac_rr(&accum0, a[4]*19, a[4]);
     mac_rr(&accum1, a[2], a[2]);
     
     c[3] = accum0 & mask;
@@ -183,6 +151,33 @@ p255_sqr (
     c[1] = c1 + (accum1>>51);
 }
 
+void
+p255_mulw (
+    p255_t *__restrict__ cs,
+    const p255_t *as,
+    uint64_t b
+) {
+    const uint64_t *a = as->limb, mask = ((1ull<<51)-1);
+    int i;
+    
+    uint64_t *c = cs->limb;
+
+    __uint128_t accum = 0;
+    for (i=0; i<5; i++) {
+        mac_rm(&accum, b, &a[i]);
+        c[i] = accum & mask;
+        accum >>= 51;
+    }
+    /* PERF: parallelize? eh well this is reference */
+    accum *= 19;
+    accum += c[0];
+    c[0] = accum & mask;
+    accum >>= 51;
+    
+    assert(accum < mask);
+    c[1] += accum;
+}
+
 void
 p255_strong_reduce (
     p255_t *a