precompute/precomputed sm works. needs demagication. slight perf regression in this build for some reason?

Makefile

@@ -70,7 +70,7 @@ LIBCOMPONENTS= build/goldilocks.o build/barrett_field.o build/crandom.o \
   build/$(FIELD).o build/ec_point.o build/scalarmul.o build/sha512.o build/magic.o \
 	build/f_arithmetic.o build/arithmetic.o
 
-DECAFCOMPONENTS= build/$(DECAF).o build/shake.o build/decaf_crypto.o
+DECAFCOMPONENTS= build/$(DECAF).o build/shake.o build/decaf_crypto.o build/decaf_tables.o
 
 TESTCOMPONENTS=build/test.o build/test_scalarmul.o build/test_sha512.o \
 	build/test_pointops.o build/test_arithmetic.o build/test_goldilocks.o build/magic.o \
@@ -113,6 +113,7 @@ else
 	ln -sf `basename $@` build/goldilocks.so.1
 endif
 
+
 build/decaf.so: $(DECAFCOMPONENTS)
 	rm -f $@
 ifeq ($(UNAME),Darwin)
@@ -131,6 +132,15 @@ build/timestamp:
 build/%.o: build/%.s
 	$(ASM) $(ASFLAGS) -c -o $@ $<
 
+build/decaf_gen_tables: build/decaf_gen_tables.o build/$(DECAF).o build/$(FIELD).o build/f_arithmetic.o
+	$(LD) $(LDFLAGS) -o $@ $^
+	
+build/decaf_tables.c: build/decaf_gen_tables
+	./$< > $@
+	
+build/decaf_tables.s: build/decaf_tables.c $(HEADERS)
+	$(CC) $(CFLAGS) -S -c -o $@ $<
+	
 build/%.s: src/%.c $(HEADERS)
 	$(CC) $(CFLAGS) -S -c -o $@ $<
 

include/decaf.h

@@ -359,7 +359,7 @@ decaf_bool_t decaf_448_direct_scalarmul (
  * @param [in] b Any point.
  */
 void decaf_448_precompute (
-    struct decaf_448_precomputed_s *a,
+    decaf_448_precomputed_s *a,
     const decaf_448_point_t b
 ) API_VIS NONNULL2;
 
@@ -376,7 +376,7 @@ void decaf_448_precompute (
  */
 void decaf_448_precomputed_scalarmul (
     decaf_448_point_t scaled,
-    const struct decaf_448_precomputed_s *base,
+    const decaf_448_precomputed_s *base,
     const decaf_448_scalar_t scalar
 ) API_VIS NONNULL3;
 

src/decaf.c

@@ -89,8 +89,13 @@ struct decaf_448_precomputed_s {
     decaf_448_point_t p[1];
 };
 
-const struct decaf_448_precomputed_s *decaf_448_precomputed_base =
-    (const struct decaf_448_precomputed_s *)decaf_448_point_base;
+/* FIXME: restore */
+// const struct decaf_448_precomputed_s *decaf_448_precomputed_base =
+//     (const struct decaf_448_precomputed_s *)decaf_448_point_base;
+
+extern const decaf_word_t decaf_448_precomputed_base_as_words[];
+const decaf_448_precomputed_s *decaf_448_precomputed_base =
+    (const decaf_448_precomputed_s *) &decaf_448_precomputed_base_as_words;
 
 const size_t sizeof_decaf_448_precomputed_s = sizeof(struct decaf_448_precomputed_s);
 const size_t alignof_decaf_448_precomputed_s = 32;

src/decaf_fast.c

@@ -15,6 +15,7 @@
 
  /* TODO REMOVE */
 #include "constant_time.h"
+#include <stdio.h>
 
 #define WBITS DECAF_WORD_BITS
 
@@ -94,10 +95,11 @@ typedef struct { gf a, b, c; } niels_s, niels_t[1];
 typedef struct { niels_t n; gf z; } pniels_s, pniels_t[1];
 struct decaf_448_precomputed_s { niels_t table [5<<4]; /* MAGIC */ };
 
-const struct decaf_448_precomputed_s decaf_448_precomputed_base_s,
-    *decaf_448_precomputed_base = &decaf_448_precomputed_base_s;
+extern const decaf_word_t decaf_448_precomputed_base_as_words[];
+const decaf_448_precomputed_s *decaf_448_precomputed_base =
+    (const decaf_448_precomputed_s *) &decaf_448_precomputed_base_as_words;
 
-const size_t sizeof_decaf_448_precomputed_s = sizeof(struct decaf_448_precomputed_s);
+const size_t sizeof_decaf_448_precomputed_s = sizeof(decaf_448_precomputed_s);
 const size_t alignof_decaf_448_precomputed_s = 32;
 
 #if (defined(__OPTIMIZE__) && !defined(__OPTIMIZE_SIZE__)) || defined(DECAF_FORCE_UNROLL)
@@ -1021,7 +1023,7 @@ void gf_batch_invert (
 
 void
 decaf_448_precompute (
-    struct decaf_448_precomputed_s *table,
+    decaf_448_precomputed_s *table,
     const decaf_448_point_t base
 ) { 
     const int n = 5, t = 5, s = 18; // TODO MAGIC
@@ -1095,14 +1097,11 @@ decaf_448_precompute (
 
 void decaf_448_precomputed_scalarmul (
     decaf_448_point_t out,
-    const struct decaf_448_precomputed_s *table,
+    const decaf_448_precomputed_s *table,
     const decaf_448_scalar_t scalar
 ) {
     unsigned int i,j,k;
-    const int n = 5, t = 5, s = 18, nbits = 450; // TODO MAGIC
-    
-    unsigned int scalar2_words = (nbits + WBITS - 1)/WBITS;
-    if (scalar2_words < SCALAR_WORDS) scalar2_words = SCALAR_WORDS;
+    const int n = 5, t = 5, s = 18; // TODO MAGIC
     
     decaf_448_scalar_t scalar2, onehalf = {{{0}}}, two = {{{2}}}, arrr;
     onehalf->limb[SCALAR_WORDS-1] = 1ull<<(WBITS-1);
@@ -1125,7 +1124,7 @@ void decaf_448_precomputed_scalarmul (
             
             for (k=0; k<t; k++) {
                 unsigned int bit = (s-1-i) + k*s + j*(s*t);
-                if (bit < scalar2_words * WBITS) {
+                if (bit < SCALAR_WORDS * WBITS) {
                     tab |= (scalar2->limb[bit/WBITS] >> (bit%WBITS) & 1) << k;
                 }
             }
@@ -1138,10 +1137,8 @@ void decaf_448_precomputed_scalarmul (
             cond_neg_niels(ni, invert);
             if (i||j) {
                 add_niels_to_pt(out, ni, j==n-1 && i<s-1);
-                assert(decaf_point_valid(out));
             } else {
                 niels_to_pt(out, ni);
-                assert(decaf_point_valid(out));
             }
         }
     }

src/decaf_gen_tables.c

@@ -0,0 +1,44 @@
+/* Copyright (c) 2015 Cryptography Research, Inc.
+ * Released under the MIT License.  See LICENSE.txt for license information.
+ */
+
+/**
+ * @file decaf_precompute.c
+ * @author Mike Hamburg
+ * @brief Decaf global constant table precomputation.
+ */
+
+#define _XOPEN_SOURCE 600 /* for posix_memalign */
+#include <stdio.h>
+#include <stdlib.h>
+#include "decaf.h"
+
+const decaf_word_t decaf_448_precomputed_base_as_words[1]; /* To satisfy linker. */
+
+int main(int argc, char **argv) {
+    (void)argc; (void)argv;
+    
+    decaf_448_precomputed_s *pre;
+    posix_memalign((void**)&pre, alignof_decaf_448_precomputed_s, sizeof_decaf_448_precomputed_s);
+    if (!pre) return 1;
+    decaf_448_precompute(pre, decaf_448_point_base);
+    
+    const decaf_word_t *output = (const decaf_word_t *)pre;
+    unsigned i;
+    
+    printf("/** @warning: this file was automatically generated. */\n");
+    printf("#include \"decaf.h\"\n\n");
+    printf("const decaf_word_t decaf_448_precomputed_base_as_words[%d]\n", 
+        (int)(sizeof_decaf_448_precomputed_s / sizeof(decaf_word_t)));
+    printf("__attribute__((aligned(%d))) = {\n  ", (int)alignof_decaf_448_precomputed_s);
+    
+    for (i=0; i < sizeof_decaf_448_precomputed_s; i+=sizeof(decaf_word_t)) {
+        if (i && (i%8==0)) printf(",\n  ");
+        else if (i) printf(", ");
+        printf("0x%0*llxull", (int)sizeof(decaf_word_t)*2, (unsigned long long)*output );
+        output++;
+    }
+    printf("\n};\n");
+    
+    return 0;
+}