gcc-clean, though the code is slow when compiled by gcc

Makefile

@@ -11,6 +11,7 @@ else
 CC = gcc
 endif
 LD = $(CC)
+ASM ?= $(CC)
 
 ifneq (,$(findstring x86_64,$(MACHINE)))
 ARCH ?= arch_x86_64
@@ -99,7 +100,7 @@ build/timestamp:
 	touch $@
 
 build/%.o: build/%.s
-	$(CC) $(ASFLAGS) -c -o $@ $<
+	$(ASM) $(ASFLAGS) -c -o $@ $<
 
 build/%.s: src/%.c $(HEADERS)
 	$(CC) $(CFLAGS) -S -c -o $@ $<

src/arch_x86_64/p448.h

@@ -187,7 +187,7 @@ p448_cond_swap (
 ) {
     big_register_t *aa = (big_register_t*)a;
     big_register_t *bb = (big_register_t*)b;
-    big_register_t m = doswap;
+    big_register_t m = br_set_to_mask(doswap);
 
     unsigned int i;
     for (i=0; i<sizeof(*a)/sizeof(*aa); i++) {
@@ -259,7 +259,7 @@ p448_cond_neg(
     struct p448_t negated;
     big_register_t *aa = (big_register_t *)a;
     big_register_t *nn = (big_register_t*)&negated;
-    big_register_t m = doNegate;
+    big_register_t m = br_set_to_mask(doNegate);
     
     p448_neg(&negated, a);
     p448_bias(&negated, 2);

src/include/word.h

@@ -71,18 +71,18 @@ static const mask_t MASK_FAILURE = 0, MASK_SUCCESS = -1;
 typedef uint32x4_t vecmask_t;
 #else
 /* FIXME this only works on clang */
-typedef uint64_t uint64x2_t __attribute__((ext_vector_type(2)));
-typedef int64_t  int64x2_t __attribute__((ext_vector_type(2)));
-typedef uint64_t uint64x4_t __attribute__((ext_vector_type(4)));
-typedef int64_t  int64x4_t __attribute__((ext_vector_type(4)));
-typedef uint32_t uint32x4_t __attribute__((ext_vector_type(4)));
-typedef int32_t  int32x4_t __attribute__((ext_vector_type(4)));
-typedef uint32_t uint32x2_t __attribute__((ext_vector_type(2)));
-typedef int32_t  int32x2_t __attribute__((ext_vector_type(2)));
-typedef uint32_t uint32x8_t __attribute__((ext_vector_type(8)));
-typedef int32_t  int32x8_t __attribute__((ext_vector_type(8)));
+typedef uint64_t uint64x2_t __attribute__((vector_size(16)));
+typedef int64_t  int64x2_t __attribute__((vector_size(16)));
+typedef uint64_t uint64x4_t __attribute__((vector_size(32)));
+typedef int64_t  int64x4_t __attribute__((vector_size(32)));
+typedef uint32_t uint32x2_t __attribute__((vector_size(8)));
+typedef int32_t  int32x2_t __attribute__((vector_size(8)));
+typedef uint32_t uint32x4_t __attribute__((vector_size(16)));
+typedef int32_t  int32x4_t __attribute__((vector_size(16)));
+typedef uint32_t uint32x8_t __attribute__((vector_size(32)));
+typedef int32_t  int32x8_t __attribute__((vector_size(32)));
 /* TODO: vector width for procs like ARM; gcc support */
-typedef word_t vecmask_t __attribute__((ext_vector_type(4)));
+typedef word_t vecmask_t __attribute__((vector_size(32)));
 #endif
 
 #if __AVX2__
@@ -111,14 +111,15 @@ br_set_to_mask(mask_t x) {
 #else
 static __inline__ big_register_t
 br_set_to_mask(mask_t x) {
-    return (big_register_t)x;
+    big_register_t out = {x,x,x,x,x,x,x,x};
+    return out;
 }
 #endif
 
 #if __AVX2__ || __SSE2__
 static __inline__ big_register_t
 br_is_zero(big_register_t x) {
-    return (big_register_t)(x == (big_register_t)0);
+    return (big_register_t)(x == br_set_to_mask(0));
 }
 #elif __ARM_NEON__
 static __inline__ big_register_t

src/sha512.c

@@ -163,9 +163,9 @@ sha512_final (
         sha512_process_block(ctx);
         fill = 0;
     }
-    memset(ctx->block + fill, 0, 112-fill);
-    *((uint64_t *)&ctx->block[112]) = 0;
-    *((uint64_t *)&ctx->block[120]) = htobe64((ctx->nbytes * 8));
+    memset(ctx->block + fill, 0, 120-fill);
+    uint64_t size = htobe64((ctx->nbytes * 8));
+    memcpy(&ctx->block[120], &size, sizeof(size));
     sha512_process_block(ctx);
     for (i=0; i<8; i++) {
         ctx->chain[i] = htobe64(ctx->chain[i]);