normalize some of the word and limb size declarations, still not entirely happy on that front

src/curve_ed25519/curve_data.inc.c

@@ -1,9 +1,8 @@
 // FIXME move to arch or something
 #define WBITS DECAF_WORD_BITS
+#define LBITS DECAF_255_LIMB_BITS
 
 #if WBITS == 64
-#define LBITS 51
-typedef __int128_t decaf_sdword_t;
 #define LIMB(x) (x##ull)
 #define SC_LIMB(x) (x##ull)
 #else

src/curve_ed448goldilocks/curve_data.inc.c

@@ -1,15 +1,9 @@
-// FIXME move to arch or something
 #define WBITS DECAF_WORD_BITS
+// #define LBITS DECAF_448_LIMB_BITS // FIXME
 
 #if WBITS == 64
-#define LBITS 56
-typedef __int128_t decaf_sdword_t;
-#define LIMB(x) (x##ull)
 #define SC_LIMB(x) (x##ull)
 #elif WBITS == 32
-typedef int64_t decaf_sdword_t;
-#define LBITS 28
-#define LIMB(x) (x##ull)&((1ull<<LBITS)-1), (x##ull)>>LBITS
 #define SC_LIMB(x) (x##ull)&((1ull<<32)-1), (x##ull)>>32
 #else
 #error "Only supporting 32- and 64-bit platforms right now"

src/decaf.c

@@ -213,7 +213,7 @@ snv gf_canon ( gf a ) {
     gf_reduce(a);
 
     /* subtract p with borrow */
-    decaf_sdword_t carry = 0;
+    decaf_dsword_t carry = 0;
     FOR_LIMB(i, {
         carry = carry + a->limb[i] - P->limb[i];
         a->limb[i] = carry & LMASK;
@@ -273,7 +273,7 @@ snv decaf_448_subx(
     const decaf_448_scalar_t p,
     decaf_word_t extra
 ) {
-    decaf_sdword_t chain = 0;
+    decaf_dsword_t chain = 0;
     unsigned int i;
     for (i=0; i<DECAF_448_SCALAR_LIMBS; i++) {
         chain = (chain + accum[i]) - sub->limb[i];
@@ -458,7 +458,7 @@ static decaf_bool_t gf_deser(gf s, const unsigned char ser[DECAF_448_SER_BYTES])
         }
     }
     
-    decaf_sdword_t accum = 0;
+    decaf_dsword_t accum = 0;
     FOR_LIMB(i, accum = (accum + s->limb[i] - P->limb[i]) >> WBITS );
     return accum;
 }
@@ -563,7 +563,7 @@ decaf_bool_t decaf_448_scalar_decode(
         s->limb[i] = out;
     }
     
-    decaf_sdword_t accum = 0;
+    decaf_dsword_t accum = 0;
     for (i=0; i<DECAF_448_SCALAR_LIMBS; i++) {
         accum = (accum + s->limb[i] - decaf_448_scalar_p->limb[i]) >> WBITS;
     }

src/decaf_fast.c

@@ -12,6 +12,8 @@
 #define __STDC_WANT_LIB_EXT1__ 1 /* for memset_s */
 #include <decaf.h>
 #include <string.h>
+
+#include "word.h"
 #include "field.h"
 #include "decaf_config.h"
 
@@ -183,7 +185,7 @@ snv sc_subx(
     const scalar_t p,
     decaf_word_t extra
 ) {
-    decaf_sdword_t chain = 0;
+    dsword_t chain = 0;
     unsigned int i;
     for (i=0; i<SCALAR_LIMBS; i++) {
         chain = (chain + accum[i]) - sub->limb[i];
@@ -710,7 +712,7 @@ decaf_bool_t API_NS(scalar_decode)(
 ) {
     unsigned int i;
     scalar_decode_short(s, ser, SER_BYTES);
-    decaf_sdword_t accum = 0;
+    dsword_t accum = 0;
     for (i=0; i<SCALAR_LIMBS; i++) {
         accum = (accum + s->limb[i] - sc_p->limb[i]) >> WBITS;
     }

src/p25519/arch_ref64/arch_config.h

@@ -1 +1,2 @@
 #define WORD_BITS 64
+#define DECAF_255_LIMB_BITS 51

src/p25519/arch_x86_64/arch_config.h

@@ -1 +1,2 @@
 #define WORD_BITS 64
+#define DECAF_255_LIMB_BITS 51

src/p25519/arch_x86_64/f_impl.h

@@ -11,7 +11,7 @@
 #include "decaf/decaf_255.h"
 #include "word.h"
 
-#define LBITS 51
+#define DECAF_255_LIMB_BITS 51
 #define FIELD_LITERAL(a,b,c,d,e) {{ a,b,c,d,e }}
 
 /*

src/p448/arch_32/arch_config.h

@@ -1 +1,2 @@
 #define WORD_BITS 32
+#define DECAF_448_LIMB_BITS 28

src/p448/arch_arm_32/arch_config.h

@@ -1 +1,2 @@
 #define WORD_BITS 32
+#define DECAF_448_LIMB_BITS 28

src/p448/arch_neon_experimental/arch_config.h

@@ -1 +1,3 @@
 #define WORD_BITS 32
+#define DECAF_448_LIMB_BITS 28
+

src/p448/arch_ref64/arch_config.h

@@ -1 +1,3 @@
 #define WORD_BITS 64
+#define DECAF_448_LIMB_BITS 56
+

src/p448/arch_x86_64/arch_config.h

@@ -1 +1,2 @@
 #define WORD_BITS 64
+#define DECAF_448_LIMB_BITS 56

src/public_include/decaf/common.h

@@ -32,6 +32,7 @@
 /** @endcond */
 
 /* Internal word types */
+/* TODO: decide this internally, per curve, based on how it was built! */
 #if (defined(__ILP64__) || defined(__amd64__) || defined(__x86_64__) || (((__UINT_FAST32_MAX__)>>30)>>30)) \
 	 && !defined(DECAF_FORCE_32_BIT)
 #define DECAF_WORD_BITS 64