Manual formatting tweaks to Curve448 code

Following running openssl-format-source there were a lot of manual tweaks
that were requried.

Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
(Merged from https://github.com/openssl/openssl/pull/5105)

24 files changed, 1693 insertions, 2748 deletions

diff --git a/crypto/ec/curve448/arch_32/f_impl.c b/crypto/ec/curve448/arch_32/f_impl.c
index 76ec9711f0..3d8a331e3e 100644
--- a/crypto/ec/curve448/arch_32/f_impl.c
+++ b/crypto/ec/curve448/arch_32/f_impl.c
@@ -24,13 +24,11 @@ void gf_mul(gf_s * __restrict__ cs, const gf as, const gf bs)
 {
     const uint32_t *a = as->limb, *b = bs->limb;
     uint32_t *c = cs->limb;
-
     uint64_t accum0 = 0, accum1 = 0, accum2 = 0;
     uint32_t mask = (1 << 28) - 1;
-
     uint32_t aa[8], bb[8];
-
     int i, j;
+
     for (i = 0; i < 8; i++) {
         aa[i] = a[i] + a[i + 8];
         bb[i] = b[i] + b[i + 8];
diff --git a/crypto/ec/curve448/arch_32/f_impl.h b/crypto/ec/curve448/arch_32/f_impl.h
index 25bfa1f79e..40a9fb93d9 100644
--- a/crypto/ec/curve448/arch_32/f_impl.h
+++ b/crypto/ec/curve448/arch_32/f_impl.h
@@ -39,9 +39,8 @@ void gf_bias(gf a, int amt)
     unsigned int i;
     uint32_t co1 = ((1 << 28) - 1) * amt, co2 = co1 - amt;
 
-    for (i = 0; i < sizeof(*a) / sizeof(a->limb[0]); i++) {
+    for (i = 0; i < sizeof(*a) / sizeof(a->limb[0]); i++)
         a->limb[i] += (i == sizeof(*a) / sizeof(a->limb[0]) / 2) ? co2 : co1;
-    }
 }
 
 void gf_weak_reduce(gf a)
@@ -51,8 +50,7 @@ void gf_weak_reduce(gf a)
     unsigned int i;
 
     a->limb[8] += tmp;
-    for (i = 15; i > 0; i--) {
+    for (i = 15; i > 0; i--)
         a->limb[i] = (a->limb[i] & mask) + (a->limb[i - 1] >> 28);
-    }
     a->limb[0] = (a->limb[0] & mask) + tmp;
 }
diff --git a/crypto/ec/curve448/arch_arm_32/arch_intrinsics.h b/crypto/ec/curve448/arch_arm_32/arch_intrinsics.h
index 73b82755c3..aa578a40e5 100644
--- a/crypto/ec/curve448/arch_arm_32/arch_intrinsics.h
+++ b/crypto/ec/curve448/arch_arm_32/arch_intrinsics.h
@@ -19,7 +19,8 @@ static __inline__ __attribute((always_inline, unused))
 uint32_t word_is_zero(uint32_t a)
 {
     uint32_t ret;
- asm("subs %0, %1, #1;\n\tsbc %0, %0, %0": "=r"(ret): "r"(a):"cc");
+
+    asm("subs %0, %1, #1;\n\tsbc %0, %0, %0": "=r"(ret): "r"(a):"cc");
     return ret;
 }
 
diff --git a/crypto/ec/curve448/arch_arm_32/f_impl.c b/crypto/ec/curve448/arch_arm_32/f_impl.c
index 25e970389b..5956d6ceba 100644
--- a/crypto/ec/curve448/arch_arm_32/f_impl.c
+++ b/crypto/ec/curve448/arch_arm_32/f_impl.c
@@ -19,9 +19,10 @@ static inline void __attribute__ ((gnu_inline, always_inline))
 #ifdef  __ARMEL__
     uint32_t lo = *acc, hi = (*acc) >> 32;
 
-    __asm__ __volatile__("smlal %[lo], %[hi], %[a], %[b]":[lo] "+&r"(lo),
-                         [hi] "+&r"(hi)
-                         :[a] "r"(a),[b] "r"(b));
+    __asm__ __volatile__ ("smlal %[lo], %[hi], %[a], %[b]"
+                          : [lo]"+&r"(lo), [hi]"+&r"(hi)
+                          : [a]"r"(a), [b]"r"(b));
+
 
     *acc = lo + (((uint64_t)hi) << 32);
 #else
@@ -35,9 +36,11 @@ static inline void __attribute__ ((gnu_inline, always_inline))
 #ifdef __ARMEL__
     uint32_t lo = *acc, hi = (*acc) >> 32;
 
-    __asm__ __volatile__("smlal %[lo], %[hi], %[a], %[b]":[lo] "+&r"(lo),
-                         [hi] "+&r"(hi)
-                         :[a] "r"(a),[b] "r"(2 * b));
+    __asm__ __volatile__ ("smlal %[lo], %[hi], %[a], %[b]"
+                          : [lo]"+&r"(lo), [hi]"+&r"(hi)
+                          : [a]"r"(a), [b]"r"(2 * b));
+
+
 
     *acc = lo + (((uint64_t)hi) << 32);
 #else
@@ -51,9 +54,9 @@ static inline void __attribute__ ((gnu_inline, always_inline))
 #ifdef __ARMEL__
     uint32_t lo, hi;
 
-    __asm__ __volatile__("smull %[lo], %[hi], %[a], %[b]":[lo] "=&r"(lo),
-                         [hi] "=&r"(hi)
-                         :[a] "r"(a),[b] "r"(b));
+    __asm__ __volatile__ ("smull %[lo], %[hi], %[a], %[b]"
+                          : [lo]"=&r"(lo), [hi]"=&r"(hi)
+                          : [a]"r"(a), [b]"r"(b));
 
     *acc = lo + (((uint64_t)hi) << 32);
 #else
@@ -68,8 +71,8 @@ static inline void __attribute__ ((gnu_inline, always_inline))
     uint32_t lo, hi;
 
     __asm__ /*__volatile__*/ ("smull %[lo], %[hi], %[a], %[b]"
- :                           [lo] "=&r"(lo),[hi] "=&r"(hi)
- :                           [a] "r"(a),[b] "r"(2 * b));
+                              : [lo]"=&r"(lo), [hi]"=&r"(hi)
+                              : [a]"r"(a), [b]"r"(2*b));
 
     *acc = lo + (((uint64_t)hi) << 32);
 #else
@@ -729,16 +732,14 @@ void gf_sqr(gf_s * __restrict__ cs, const gf as)
 void gf_mulw_unsigned(gf_s * __restrict__ cs, const gf as, uint32_t b)
 {
     uint32_t mask = (1ull << 28) - 1;
-    assert(b <= mask);
-
     const uint32_t *a = as->limb;
     uint32_t *c = cs->limb;
-
     uint64_t accum0, accum8;
-
     int i;
-
     uint32_t c0, c8, n0, n8;
+
+    assert(b <= mask);
+
     c0 = a[0];
     c8 = a[8];
     accum0 = widemul(b, c0);
diff --git a/crypto/ec/curve448/arch_arm_32/f_impl.h b/crypto/ec/curve448/arch_arm_32/f_impl.h
index 9008619ce7..2e385d5214 100644
--- a/crypto/ec/curve448/arch_arm_32/f_impl.h
+++ b/crypto/ec/curve448/arch_arm_32/f_impl.h
@@ -23,10 +23,6 @@ void gf_add_RAW(gf out, const gf a, const gf b)
         ((uint32xn_t *) out)[i] =
             ((const uint32xn_t *)a)[i] + ((const uint32xn_t *)b)[i];
     }
-    /*
-     * for (unsigned int i=0; i<sizeof(*out)/sizeof(out->limb[0]); i++) {
-     * out->limb[i] = a->limb[i] + b->limb[i]; }
-     */
 }
 
 void gf_sub_RAW(gf out, const gf a, const gf b)
@@ -35,10 +31,6 @@ void gf_sub_RAW(gf out, const gf a, const gf b)
         ((uint32xn_t *) out)[i] =
             ((const uint32xn_t *)a)[i] - ((const uint32xn_t *)b)[i];
     }
-    /*
-     * for (unsigned int i=0; i<sizeof(*out)/sizeof(out->limb[0]); i++) {
-     * out->limb[i] = a->limb[i] - b->limb[i]; }
-     */
 }
 
 void gf_bias(gf a, int amt)
@@ -47,6 +39,7 @@ void gf_bias(gf a, int amt)
     uint32x4_t lo = { co1, co1, co1, co1 }, hi = {
     co2, co1, co1, co1};
     uint32x4_t *aa = (uint32x4_t *) a;
+
     aa[0] += lo;
     aa[1] += lo;
     aa[2] += hi;
@@ -57,6 +50,7 @@ void gf_weak_reduce(gf a)
 {
     uint64_t mask = (1ull << 28) - 1;
     uint64_t tmp = a->limb[15] >> 28;
+
     a->limb[8] += tmp;
     for (unsigned int i = 15; i > 0; i--) {
         a->limb[i] = (a->limb[i] & mask) + (a->limb[i - 1] >> 28);
diff --git a/crypto/ec/curve448/arch_neon/arch_intrinsics.h b/crypto/ec/curve448/arch_neon/arch_intrinsics.h
index 201b3735b3..3947b43485 100644
--- a/crypto/ec/curve448/arch_neon/arch_intrinsics.h
+++ b/crypto/ec/curve448/arch_neon/arch_intrinsics.h
@@ -19,7 +19,7 @@ static __inline__ __attribute((always_inline, unused))
 uint32_t word_is_zero(uint32_t a)
 {
     uint32_t ret;
- __asm__("subs %0, %1, #1;\n\tsbc %0, %0, %0": "=r"(ret): "r"(a):"cc");
+    __asm__("subs %0, %1, #1;\n\tsbc %0, %0, %0": "=r"(ret): "r"(a):"cc");
     return ret;
 }
 
diff --git a/crypto/ec/curve448/arch_neon/f_impl.c b/crypto/ec/curve448/arch_neon/f_impl.c
index fe21aee128..7f6fda24f9 100644
--- a/crypto/ec/curve448/arch_neon/f_impl.c
+++ b/crypto/ec/curve448/arch_neon/f_impl.c
@@ -12,588 +12,583 @@
 
 #include "f_field.h"
 
-static __inline__ uint64x2_t __attribute__ ((gnu_inline, always_inline, unused))
-    xx_vaddup_u64(uint64x2_t x)
+static __inline__ uint64x2_t __attribute__((gnu_inline,always_inline,unused))
+xx_vaddup_u64(uint64x2_t x)
 {
- __asm__("vadd.s64 %f0, %e0":"+w"(x));
+    __asm__ ("vadd.s64 %f0, %e0" : "+w"(x));
     return x;
 }
 
-static __inline__ int64x2_t __attribute__ ((gnu_inline, always_inline, unused))
-    vrev128_s64(int64x2_t x)
+static __inline__ int64x2_t __attribute__((gnu_inline,always_inline,unused))
+vrev128_s64(int64x2_t x)
 {
- __asm__("vswp.s64 %e0, %f0":"+w"(x));
+    __asm__ ("vswp.s64 %e0, %f0" : "+w"(x));
     return x;
 }
 
-static __inline__ uint64x2_t __attribute__ ((gnu_inline, always_inline))
-    vrev128_u64(uint64x2_t x)
+static __inline__ uint64x2_t __attribute__((gnu_inline,always_inline))
+vrev128_u64(uint64x2_t x)
 {
- __asm__("vswp.s64 %e0, %f0":"+w"(x));
+    __asm__ ("vswp.s64 %e0, %f0" : "+w"(x));
     return x;
 }
 
-static inline void __attribute__ ((gnu_inline, always_inline, unused))
-    smlal(uint64_t *acc, const uint32_t a, const uint32_t b)
+static inline void __attribute__((gnu_inline,always_inline,unused))
+smlal(uint64_t *acc, const uint32_t a, const uint32_t b)
 {
-    *acc += (int64_t)(int32_t)a *(int64_t)(int32_t)b;
+    *acc += (int64_t)(int32_t)a * (int64_t)(int32_t)b;
 }
 
-static inline void __attribute__ ((gnu_inline, always_inline, unused))
-    smlal2(uint64_t *acc, const uint32_t a, const uint32_t b)
+static inline void __attribute__((gnu_inline,always_inline,unused))
+smlal2(uint64_t *acc, const uint32_t a, const uint32_t b)
 {
-    *acc += (int64_t)(int32_t)a *(int64_t)(int32_t)b *2;
+    *acc += (int64_t)(int32_t)a * (int64_t)(int32_t)b * 2;
 }
 
-static inline void __attribute__ ((gnu_inline, always_inline, unused))
-    smull(uint64_t *acc, const uint32_t a, const uint32_t b)
+static inline void __attribute__((gnu_inline,always_inline,unused))
+smull(uint64_t *acc, const uint32_t a, const uint32_t b)
 {
-    *acc = (int64_t)(int32_t)a *(int64_t)(int32_t)b;
+    *acc = (int64_t)(int32_t)a * (int64_t)(int32_t)b;
 }
 
-static inline void __attribute__ ((gnu_inline, always_inline, unused))
-    smull2(uint64_t *acc, const uint32_t a, const uint32_t b)
+static inline void __attribute__((gnu_inline,always_inline,unused))
+smull2(uint64_t *acc, const uint32_t a, const uint32_t b)
 {
-    *acc = (int64_t)(int32_t)a *(int64_t)(int32_t)b *2;
+    *acc = (int64_t)(int32_t)a * (int64_t)(int32_t)b * 2;
 }
 
-void gf_mul(gf_s * __restrict__ cs, const gf as, const gf bs)
+void gf_mul(gf_s *__restrict__ cs, const gf as, const gf bs)
 {
-#define _bl0 "q0"
-#define _bl0_0 "d0"
-#define _bl0_1 "d1"
-#define _bh0 "q1"
-#define _bh0_0 "d2"
-#define _bh0_1 "d3"
-#define _bs0 "q2"
-#define _bs0_0 "d4"
-#define _bs0_1 "d5"
-#define _bl2 "q3"
-#define _bl2_0 "d6"
-#define _bl2_1 "d7"
-#define _bh2 "q4"
-#define _bh2_0 "d8"
-#define _bh2_1 "d9"
-#define _bs2 "q5"
-#define _bs2_0 "d10"
-#define _bs2_1 "d11"
-
-#define _as0 "q6"
-#define _as0_0 "d12"
-#define _as0_1 "d13"
-#define _as2 "q7"
-#define _as2_0 "d14"
-#define _as2_1 "d15"
-#define _al0 "q8"
-#define _al0_0 "d16"
-#define _al0_1 "d17"
-#define _ah0 "q9"
-#define _ah0_0 "d18"
-#define _ah0_1 "d19"
-#define _al2 "q10"
-#define _al2_0 "d20"
-#define _al2_1 "d21"
-#define _ah2 "q11"
-#define _ah2_0 "d22"
-#define _ah2_1 "d23"
-
-#define _a0a "q12"
-#define _a0a_0 "d24"
-#define _a0a_1 "d25"
-#define _a0b "q13"
-#define _a0b_0 "d26"
-#define _a0b_1 "d27"
-#define _a1a "q14"
-#define _a1a_0 "d28"
-#define _a1a_1 "d29"
-#define _a1b "q15"
-#define _a1b_0 "d30"
-#define _a1b_1 "d31"
-#define VMAC(op,result,a,b,n) #op" "result", "a", "b"[" #n "]\n\t"
-#define VOP3(op,result,a,b)   #op" "result", "a", "b"\n\t"
-#define VOP2(op,result,a)     #op" "result", "a"\n\t"
-
-    int32x2_t *vc = (int32x2_t *) cs->limb;
-
-    __asm__ __volatile__("vld2.32 {" _al0_0 "," _al0_1 "," _ah0_0 "," _ah0_1
-                         "}, [%[a],:128]!" "\n\t" VOP3(vadd.i32, _as0, _al0,
-                                                       _ah0)
-                         "vld2.32 {" _bl0_0 "," _bl0_1 "," _bh0_0 "," _bh0_1
-                         "}, [%[b],:128]!" "\n\t" VOP3(vadd.i32, _bs0_1, _bl0_1,
-                                                       _bh0_1) VOP3(vsub.i32,
-                                                                    _bs0_0,
-                                                                    _bl0_0,
-                                                                    _bh0_0)
-                         "vld2.32 {" _bl2_0 "," _bl2_1 "," _bh2_0 "," _bh2_1
-                         "}, [%[b],:128]!" "\n\t" VOP3(vadd.i32, _bs2, _bl2,
-                                                       _bh2)
-                         "vld2.32 {" _al2_0 "," _al2_1 "," _ah2_0 "," _ah2_1
-                         "}, [%[a],:128]!" "\n\t" VOP3(vadd.i32, _as2, _al2,
-                                                       _ah2)
-                         VMAC(vmull.s32, _a0b, _as0_1, _bs2_1,
-                              0) VMAC(vmlal.s32, _a0b, _as2_0, _bs2_0,
-                                      0) VMAC(vmlal.s32, _a0b, _as2_1, _bs0_1,
-                                              0) VMAC(vmlal.s32, _a0b, _as0_0,
-                                                      _bh0_0, 0)
-                         VMAC(vmull.s32, _a1b, _as0_1, _bs2_1,
-                              1) VMAC(vmlal.s32, _a1b, _as2_0, _bs2_0,
-                                      1) VMAC(vmlal.s32, _a1b, _as2_1, _bs0_1,
-                                              1) VMAC(vmlal.s32, _a1b, _as0_0,
-                                                      _bh0_0, 1)
-                         VOP2(vmov, _a0a, _a0b) VMAC(vmlal.s32, _a0a, _ah0_1,
-                                                     _bh2_1, 0) VMAC(vmlal.s32,
-                                                                     _a0a,
-                                                                     _ah2_0,
-                                                                     _bh2_0,
-                                                                     0)
-                         VMAC(vmlal.s32, _a0a, _ah2_1, _bh0_1,
-                              0) VMAC(vmlal.s32, _a0a, _ah0_0, _bl0_0, 0)
-                         VMAC(vmlsl.s32, _a0b, _al0_1, _bl2_1,
-                              0) VMAC(vmlsl.s32, _a0b, _al2_0, _bl2_0,
-                                      0) VMAC(vmlsl.s32, _a0b, _al2_1, _bl0_1,
-                                              0) VMAC(vmlal.s32, _a0b, _al0_0,
-                                                      _bs0_0, 0)
-                         VOP2(vmov, _a1a, _a1b) VMAC(vmlal.s32, _a1a, _ah0_1,
-                                                     _bh2_1, 1) VMAC(vmlal.s32,
-                                                                     _a1a,
-                                                                     _ah2_0,
-                                                                     _bh2_0,
-                                                                     1)
-                         VMAC(vmlal.s32, _a1a, _ah2_1, _bh0_1,
-                              1) VMAC(vmlal.s32, _a1a, _ah0_0, _bl0_0, 1)
-                         VOP2(vswp, _a0b_1, _a0a_0)
-                         VMAC(vmlsl.s32, _a1b, _al0_1, _bl2_1, 1)
-                         VMAC(vmlsl.s32, _a1b, _al2_0, _bl2_0, 1)
-                         VMAC(vmlsl.s32, _a1b, _al2_1, _bl0_1, 1)
-                         VMAC(vmlal.s32, _a1b, _al0_0, _bs0_0, 1)
-                         VOP3(vsra.u64, _a0a, _a0b, "#28")
-                         VOP3(vsub.i32, _bs0_1, _bl0_1, _bh0_1)
-                         VOP2(vmovn.i64, _a0b_0, _a0b)
-                         VOP2(vswp, _a1b_1, _a1a_0)
-                         VOP3(vadd.i64, _a1b, _a0a, _a1b)
-                         VMAC(vmull.s32, _a0a, _as2_0, _bs2_1, 0)
-                         VOP2(vmovn.i64, _a0b_1, _a1b)
-                         VMAC(vmlal.s32, _a0a, _as2_1, _bs2_0, 0)
-                         VOP3(vsra.u64, _a1a, _a1b, "#28")
-                         VMAC(vmlal.s32, _a0a, _as0_0, _bh0_1, 0)
-                         VOP2(vbic.i32, _a0b, "#0xf0000000")
-                         VMAC(vmlal.s32, _a0a, _as0_1, _bh0_0, 0)
-                         "vstmia %[c]!, {" _a0b_0 ", " _a0b_1 "}" "\n\t"
-                         VMAC(vmull.s32, _a1b, _as2_0, _bs2_1, 1)
-                         VMAC(vmlal.s32, _a1b, _as2_1, _bs2_0, 1)
-                         VMAC(vmlal.s32, _a1b, _as0_0, _bh0_1, 1)
-                         VMAC(vmlal.s32, _a1b, _as0_1, _bh0_0, 1)
-                         VOP2(vmov, _a0b_1, _a0a_1)
-                         VOP3(vadd.i64, _a0b_0, _a0a_0, _a1a_0)
-                         VOP3(vadd.i64, _a0a_0, _a0a_0, _a1a_1)
-                         VMAC(vmlal.s32, _a0a, _ah2_0, _bh2_1, 0)
-                         VMAC(vmlal.s32, _a0a, _ah2_1, _bh2_0, 0)
-                         VMAC(vmlal.s32, _a0a, _ah0_0, _bl0_1, 0)
-                         VMAC(vmlal.s32, _a0a, _ah0_1, _bl0_0, 0)
-                         VMAC(vmlsl.s32, _a0b, _al2_0, _bl2_1, 0)
-                         VMAC(vmlsl.s32, _a0b, _al2_1, _bl2_0, 0)
-                         VMAC(vmlal.s32, _a0b, _al0_0, _bs0_1, 0)
-                         VMAC(vmlal.s32, _a0b, _al0_1, _bs0_0, 0)
-                         VOP2(vmov, _a1a, _a1b)
-                         VMAC(vmlal.s32, _a1a, _ah2_0, _bh2_1, 1)
-                         VMAC(vmlal.s32, _a1a, _ah2_1, _bh2_0, 1)
-                         VMAC(vmlal.s32, _a1a, _ah0_0, _bl0_1, 1)
-                         VMAC(vmlal.s32, _a1a, _ah0_1, _bl0_0, 1)
-                         VOP2(vswp, _a0b_1, _a0a_0)
-                         VMAC(vmlsl.s32, _a1b, _al2_0, _bl2_1, 1)
-                         VMAC(vmlsl.s32, _a1b, _al2_1, _bl2_0, 1)
-                         VMAC(vmlal.s32, _a1b, _al0_0, _bs0_1, 1)
-                         VMAC(vmlal.s32, _a1b, _al0_1, _bs0_0, 1)
-                         VOP3(vsra.u64, _a0a, _a0b, "#28")
-                         VOP3(vsub.i32, _bs2_0, _bl2_0, _bh2_0)
-                         VOP2(vmovn.i64, _a0b_0, _a0b)
-                         VOP2(vswp, _a1b_1, _a1a_0)
-                         VOP3(vadd.i64, _a1b, _a0a, _a1b)
-                         VMAC(vmull.s32, _a0a, _as2_1, _bs2_1, 0)
-                         VOP2(vmovn.i64, _a0b_1, _a1b)
-                         VMAC(vmlal.s32, _a0a, _as0_0, _bh2_0, 0)
-                         VOP3(vsra.u64, _a1a, _a1b, "#28")
-                         VMAC(vmlal.s32, _a0a, _as0_1, _bh0_1, 0)
-                         VOP2(vbic.i32, _a0b, "#0xf0000000")
-                         VMAC(vmlal.s32, _a0a, _as2_0, _bh0_0, 0)
-                         "vstmia %[c]!, {" _a0b_0 ", " _a0b_1 "}" "\n\t"
-                         VMAC(vmull.s32, _a1b, _as2_1, _bs2_1, 1)
-                         VMAC(vmlal.s32, _a1b, _as0_0, _bh2_0, 1)
-                         VMAC(vmlal.s32, _a1b, _as0_1, _bh0_1, 1)
-                         VMAC(vmlal.s32, _a1b, _as2_0, _bh0_0, 1)
-                         VOP2(vmov, _a0b_1, _a0a_1)
-                         VOP3(vadd.i64, _a0b_0, _a0a_0, _a1a_0)
-                         VOP3(vadd.i64, _a0a_0, _a0a_0, _a1a_1)
-                         VMAC(vmlal.s32, _a0a, _ah2_1, _bh2_1, 0)
-                         VMAC(vmlal.s32, _a0a, _ah0_0, _bl2_0, 0)
-                         VMAC(vmlal.s32, _a0a, _ah0_1, _bl0_1, 0)
-                         VMAC(vmlal.s32, _a0a, _ah2_0, _bl0_0, 0)
-                         VMAC(vmlsl.s32, _a0b, _al2_1, _bl2_1, 0)
-                         VMAC(vmlal.s32, _a0b, _al0_0, _bs2_0, 0)
-                         VMAC(vmlal.s32, _a0b, _al0_1, _bs0_1, 0)
-                         VMAC(vmlal.s32, _a0b, _al2_0, _bs0_0, 0)
-                         VOP2(vmov, _a1a, _a1b)
-                         VMAC(vmlal.s32, _a1a, _ah2_1, _bh2_1, 1)
-                         VMAC(vmlal.s32, _a1a, _ah0_0, _bl2_0, 1)
-                         VMAC(vmlal.s32, _a1a, _ah0_1, _bl0_1, 1)
-                         VMAC(vmlal.s32, _a1a, _ah2_0, _bl0_0, 1)
-                         VOP2(vswp, _a0b_1, _a0a_0)
-                         VMAC(vmlsl.s32, _a1b, _al2_1, _bl2_1, 1)
-                         VMAC(vmlal.s32, _a1b, _al0_0, _bs2_0, 1)
-                         VMAC(vmlal.s32, _a1b, _al0_1, _bs0_1, 1)
-                         VMAC(vmlal.s32, _a1b, _al2_0, _bs0_0, 1)
-                         VOP3(vsub.i32, _bs2_1, _bl2_1, _bh2_1)
-                         VOP3(vsra.u64, _a0a, _a0b, "#28")
-                         VOP2(vmovn.i64, _a0b_0, _a0b)
-                         VOP2(vswp, _a1b_1, _a1a_0)
-                         VOP3(vadd.i64, _a1b, _a0a, _a1b)
-                         VMAC(vmull.s32, _a0a, _as0_0, _bh2_1, 0)
-                         VOP2(vmovn.i64, _a0b_1, _a1b)
-                         VMAC(vmlal.s32, _a0a, _as0_1, _bh2_0, 0)
-                         VOP3(vsra.u64, _a1a, _a1b, "#28")
-                         VMAC(vmlal.s32, _a0a, _as2_0, _bh0_1, 0)
-                         VOP2(vbic.i32, _a0b, "#0xf0000000")
-                         VMAC(vmlal.s32, _a0a, _as2_1, _bh0_0, 0)
-                         "vstmia %[c]!, {" _a0b_0 ", " _a0b_1 "}" "\n\t"
-                         VMAC(vmull.s32, _a1b, _as0_0, _bh2_1, 1)
-                         VMAC(vmlal.s32, _a1b, _as0_1, _bh2_0, 1)
-                         VMAC(vmlal.s32, _a1b, _as2_0, _bh0_1, 1)
-                         VMAC(vmlal.s32, _a1b, _as2_1, _bh0_0, 1)
-                         VOP2(vmov, _a0b_1, _a0a_1)
-                         VOP3(vadd.i64, _a0b_0, _a0a_0, _a1a_0)
-                         VOP3(vadd.i64, _a0a_0, _a0a_0, _a1a_1)
-                         VMAC(vmlal.s32, _a0a, _ah0_0, _bl2_1, 0)
-                         VMAC(vmlal.s32, _a0a, _ah0_1, _bl2_0, 0)
-                         VMAC(vmlal.s32, _a0a, _ah2_0, _bl0_1, 0)
-                         VMAC(vmlal.s32, _a0a, _ah2_1, _bl0_0, 0)
-                         VMAC(vmlal.s32, _a0b, _al0_0, _bs2_1, 0)
-                         VMAC(vmlal.s32, _a0b, _al0_1, _bs2_0, 0)
-                         VMAC(vmlal.s32, _a0b, _al2_0, _bs0_1, 0)
-                         VMAC(vmlal.s32, _a0b, _al2_1, _bs0_0, 0)
-                         VOP2(vmov, _a1a, _a1b)
-                         VMAC(vmlal.s32, _a1a, _ah0_0, _bl2_1, 1)
-                         VMAC(vmlal.s32, _a1a, _ah0_1, _bl2_0, 1)
-                         VMAC(vmlal.s32, _a1a, _ah2_0, _bl0_1, 1)
-                         VMAC(vmlal.s32, _a1a, _ah2_1, _bl0_0, 1)
-                         VOP2(vswp, _a0b_1, _a0a_0)
-                         VMAC(vmlal.s32, _a1b, _al0_0, _bs2_1, 1)
-                         VMAC(vmlal.s32, _a1b, _al0_1, _bs2_0, 1)
-                         VMAC(vmlal.s32, _a1b, _al2_0, _bs0_1, 1)
-                         VMAC(vmlal.s32, _a1b, _al2_1, _bs0_0, 1)
-                         VOP3(vsra.u64, _a0a, _a0b, "#28")
-                         VOP2(vmovn.i64, _a0b_0, _a0b)
-                         VOP2(vswp, _a1b_1, _a1a_0)
-                         VOP3(vadd.i64, _a0a, _a0a, _a1b)
-                         VOP2(vmovn.i64, _a0b_1, _a0a)
-                         VOP3(vsra.u64, _a1a, _a0a, "#28")
-                         VOP2(vbic.i32, _a0b, "#0xf0000000")
-                         VOP2(vswp, _a1a_0, _a1a_1)
-                         "vstmia %[c]!, {" _a0b_0 ", " _a0b_1 "}" "\n\t"
-                         "sub %[c], #64" "\n\t"
-                         VOP3(vadd.i64, _a1a_1, _a1a_1, _a1a_0)
-                         "vldmia %[c], {" _a0a_0 ", " _a0a_1 ", " _a0b_0 "}"
-                         "\n\t" VOP2(vaddw.s32, _a1a, _a0a_0) VOP2(vmovn.i64,
-                                                                   _a0a_0,
-                                                                   _a1a)
-                         VOP2(vshr.s64, _a1a, "#28")
-                         VOP2(vaddw.s32, _a1a, _a0a_1) VOP2(vmovn.i64, _a0a_1,
-                                                            _a1a) VOP2(vshr.s64,
-                                                                       _a1a,
-                                                                       "#28")
-                         VOP2(vbic.i32, _a0a, "#0xf0000000")
-                         VOP2(vaddw.s32, _a1a, _a0b_0)
-                         VOP2(vmovn.i64, _a0b_0, _a1a)
-                         "vstmia %[c], {" _a0a_0 ", " _a0a_1 ", " _a0b_0 "}"
-                         "\n\t":[a] "+r"(as)
-                         ,[b] "+r"(bs)
-                         ,[c] "+r"(vc)
-
-                         ::"q0", "q1", "q2", "q3",
-                         "q4", "q5", "q6", "q7",
-                         "q8", "q9", "q10", "q11",
-                         "q12", "q13", "q14", "q15", "memory");
+    #define _bl0 "q0"
+    #define _bl0_0 "d0"
+    #define _bl0_1 "d1"
+    #define _bh0 "q1"
+    #define _bh0_0 "d2"
+    #define _bh0_1 "d3"
+    #define _bs0 "q2"
+    #define _bs0_0 "d4"
+    #define _bs0_1 "d5"
+    #define _bl2 "q3"
+    #define _bl2_0 "d6"
+    #define _bl2_1 "d7"
+    #define _bh2 "q4"
+    #define _bh2_0 "d8"
+    #define _bh2_1 "d9"
+    #define _bs2 "q5"
+    #define _bs2_0 "d10"
+    #define _bs2_1 "d11"
+
+    #define _as0 "q6"
+    #define _as0_0 "d12"
+    #define _as0_1 "d13"
+    #define _as2 "q7"
+    #define _as2_0 "d14"
+    #define _as2_1 "d15"
+    #define _al0 "q8"
+    #define _al0_0 "d16"
+    #define _al0_1 "d17"
+    #define _ah0 "q9"
+    #define _ah0_0 "d18"
+    #define _ah0_1 "d19"
+    #define _al2 "q10"
+    #define _al2_0 "d20"
+    #define _al2_1 "d21"
+    #define _ah2 "q11"
+    #define _ah2_0 "d22"
+    #define _ah2_1 "d23"
+
+    #define _a0a "q12"
+    #define _a0a_0 "d24"
+    #define _a0a_1 "d25"
+    #define _a0b "q13"
+    #define _a0b_0 "d26"
+    #define _a0b_1 "d27"
+    #define _a1a "q14"
+    #define _a1a_0 "d28"
+    #define _a1a_1 "d29"
+    #define _a1b "q15"
+    #define _a1b_0 "d30"
+    #define _a1b_1 "d31"
+    #define VMAC(op,result,a,b,n) #op" "result", "a", "b"[" #n "]\n\t"
+    #define VOP3(op,result,a,b)   #op" "result", "a", "b"\n\t"
+    #define VOP2(op,result,a)     #op" "result", "a"\n\t"
+
+    int32x2_t *vc = (int32x2_t*) cs->limb;
+
+    __asm__ __volatile__(
+        
+        "vld2.32 {"_al0_0","_al0_1","_ah0_0","_ah0_1"}, [%[a],:128]!" "\n\t"
+        VOP3(vadd.i32,_as0,_al0,_ah0)
+        
+        "vld2.32 {"_bl0_0","_bl0_1","_bh0_0","_bh0_1"}, [%[b],:128]!" "\n\t"
+        VOP3(vadd.i32,_bs0_1,_bl0_1,_bh0_1)
+        VOP3(vsub.i32,_bs0_0,_bl0_0,_bh0_0)
+            
+        "vld2.32 {"_bl2_0","_bl2_1","_bh2_0","_bh2_1"}, [%[b],:128]!" "\n\t"
+        VOP3(vadd.i32,_bs2,_bl2,_bh2)
+            
+        "vld2.32 {"_al2_0","_al2_1","_ah2_0","_ah2_1"}, [%[a],:128]!" "\n\t"
+        VOP3(vadd.i32,_as2,_al2,_ah2)
+        
+        VMAC(vmull.s32,_a0b,_as0_1,_bs2_1,0)
+        VMAC(vmlal.s32,_a0b,_as2_0,_bs2_0,0)
+        VMAC(vmlal.s32,_a0b,_as2_1,_bs0_1,0)
+        VMAC(vmlal.s32,_a0b,_as0_0,_bh0_0,0)
+            
+        VMAC(vmull.s32,_a1b,_as0_1,_bs2_1,1)
+        VMAC(vmlal.s32,_a1b,_as2_0,_bs2_0,1)
+        VMAC(vmlal.s32,_a1b,_as2_1,_bs0_1,1)
+        VMAC(vmlal.s32,_a1b,_as0_0,_bh0_0,1)
+            
+        VOP2(vmov,_a0a,_a0b)
+        VMAC(vmlal.s32,_a0a,_ah0_1,_bh2_1,0)
+        VMAC(vmlal.s32,_a0a,_ah2_0,_bh2_0,0)
+        VMAC(vmlal.s32,_a0a,_ah2_1,_bh0_1,0)
+        VMAC(vmlal.s32,_a0a,_ah0_0,_bl0_0,0)
+            
+        VMAC(vmlsl.s32,_a0b,_al0_1,_bl2_1,0)
+        VMAC(vmlsl.s32,_a0b,_al2_0,_bl2_0,0)
+        VMAC(vmlsl.s32,_a0b,_al2_1,_bl0_1,0)
+        VMAC(vmlal.s32,_a0b,_al0_0,_bs0_0,0)
+            
+        VOP2(vmov,_a1a,_a1b)
+        VMAC(vmlal.s32,_a1a,_ah0_1,_bh2_1,1)
+        VMAC(vmlal.s32,_a1a,_ah2_0,_bh2_0,1)
+        VMAC(vmlal.s32,_a1a,_ah2_1,_bh0_1,1)
+        VMAC(vmlal.s32,_a1a,_ah0_0,_bl0_0,1)
+            
+            VOP2(vswp,_a0b_1,_a0a_0)
+            
+        VMAC(vmlsl.s32,_a1b,_al0_1,_bl2_1,1)
+        VMAC(vmlsl.s32,_a1b,_al2_0,_bl2_0,1)
+        VMAC(vmlsl.s32,_a1b,_al2_1,_bl0_1,1)
+        VMAC(vmlal.s32,_a1b,_al0_0,_bs0_0,1)
+                
+            VOP3(vsra.u64,_a0a,_a0b,"#28")
+            VOP3(vsub.i32,_bs0_1,_bl0_1,_bh0_1)
+            VOP2(vmovn.i64,_a0b_0,_a0b)
+                
+            VOP2(vswp,_a1b_1,_a1a_0)
+            VOP3(vadd.i64,_a1b,_a0a,_a1b)
+                    
+                    
+        VMAC(vmull.s32,_a0a,_as2_0,_bs2_1,0)
+            VOP2(vmovn.i64,_a0b_1,_a1b)
+        VMAC(vmlal.s32,_a0a,_as2_1,_bs2_0,0)
+            VOP3(vsra.u64,_a1a,_a1b,"#28")
+        VMAC(vmlal.s32,_a0a,_as0_0,_bh0_1,0)
+            VOP2(vbic.i32,_a0b,"#0xf0000000")
+        VMAC(vmlal.s32,_a0a,_as0_1,_bh0_0,0)
+            "vstmia %[c]!, {"_a0b_0", "_a0b_1"}" "\n\t"
+                    
+        VMAC(vmull.s32,_a1b,_as2_0,_bs2_1,1)
+        VMAC(vmlal.s32,_a1b,_as2_1,_bs2_0,1)
+        VMAC(vmlal.s32,_a1b,_as0_0,_bh0_1,1)
+        VMAC(vmlal.s32,_a1b,_as0_1,_bh0_0,1)
+
+        VOP2(vmov,_a0b_1,_a0a_1)
+        VOP3(vadd.i64,_a0b_0,_a0a_0,_a1a_0)
+        VOP3(vadd.i64,_a0a_0,_a0a_0,_a1a_1)
+        VMAC(vmlal.s32,_a0a,_ah2_0,_bh2_1,0)
+        VMAC(vmlal.s32,_a0a,_ah2_1,_bh2_0,0)
+        VMAC(vmlal.s32,_a0a,_ah0_0,_bl0_1,0)
+        VMAC(vmlal.s32,_a0a,_ah0_1,_bl0_0,0)
+
+        VMAC(vmlsl.s32,_a0b,_al2_0,_bl2_1,0)
+        VMAC(vmlsl.s32,_a0b,_al2_1,_bl2_0,0)
+        VMAC(vmlal.s32,_a0b,_al0_0,_bs0_1,0)
+        VMAC(vmlal.s32,_a0b,_al0_1,_bs0_0,0)
+
+        VOP2(vmov,_a1a,_a1b)
+        VMAC(vmlal.s32,_a1a,_ah2_0,_bh2_1,1)
+        VMAC(vmlal.s32,_a1a,_ah2_1,_bh2_0,1)
+        VMAC(vmlal.s32,_a1a,_ah0_0,_bl0_1,1)
+        VMAC(vmlal.s32,_a1a,_ah0_1,_bl0_0,1)
+
+            VOP2(vswp,_a0b_1,_a0a_0)
+
+        VMAC(vmlsl.s32,_a1b,_al2_0,_bl2_1,1)
+        VMAC(vmlsl.s32,_a1b,_al2_1,_bl2_0,1)
+        VMAC(vmlal.s32,_a1b,_al0_0,_bs0_1,1)
+        VMAC(vmlal.s32,_a1b,_al0_1,_bs0_0,1)
+                                        
+            VOP3(vsra.u64,_a0a,_a0b,"#28")
+            VOP3(vsub.i32,_bs2_0,_bl2_0,_bh2_0)
+            VOP2(vmovn.i64,_a0b_0,_a0b)
+                        
+            VOP2(vswp,_a1b_1,_a1a_0)
+            VOP3(vadd.i64,_a1b,_a0a,_a1b)
+
+        VMAC(vmull.s32,_a0a,_as2_1,_bs2_1,0)
+            VOP2(vmovn.i64,_a0b_1,_a1b)
+        VMAC(vmlal.s32,_a0a,_as0_0,_bh2_0,0)
+            VOP3(vsra.u64,_a1a,_a1b,"#28")
+        VMAC(vmlal.s32,_a0a,_as0_1,_bh0_1,0)
+            VOP2(vbic.i32,_a0b,"#0xf0000000")
+        VMA