temporarily removed avx2 fixes for want of further testing

hmm it allows the compiler to pick fpu regs on its own, but is it _truly_ equivalent to the asm code?

crypto/libntrup/src/avx/mult.c

@@ -8,12 +8,12 @@
 #define MULSTEP_gcc(j,h0,h1,h2,h3,h4) \
   gj = g[j]; \
   h0 += f0 * gj; \
-  _mm256_storeu_ps((float*)&h[i + j],h0); \
+  _mm256_storeu_ps(&h[i + j],h0); \
   h1 += f1 * gj; \
   h2 += f2 * gj; \
   h3 += f3 * gj; \
   h4 += f4 * gj; \
-  h0 = _mm256_loadu_ps((float*)&h[i + j + 5]); \
+  h0 = _mm256_loadu_ps(&h[i + j + 5]); \
   h0 += f5 * gj;
 
 #define MULSTEP_asm(j,h0,h1,h2,h3,h4) \
@@ -30,19 +30,9 @@
     : "+x"(h0),"+x"(h1),"+x"(h2),"+x"(h3),"+x"(h4) \
     : "x"(gj),"x"(f0),"x"(f1),"x"(f2),"x"(f3),"x"(f4),"x"(f5),"m"(h[i+j]),"m"(h[i+j+5]));
 
-#define MULSTEP MULSTEP_gcc
+#define MULSTEP MULSTEP_asm
 
 #define MULSTEP_noload(j,h0,h1,h2,h3,h4) \
-  gj = g[j]; \
-  h0 += gj*f0; \
-  _mm256_storeu_ps((float*)&h[i+j], h0); \
-  h1 += gj*f1; \
-  h2 += gj*f2; \
-  h3 += gj*f3; \
-  h4 += gj*f4; \
-  h0 = gj* f5;
-
-#define MULSTEP_noload_asm(j,h0,h1,h2,h3,h4) \
   gj = g[j]; \
   __asm__( \
     "vfmadd231ps %5,%6,%0 \n\t" \
@@ -56,16 +46,6 @@
     : "x"(gj),"x"(f0),"x"(f1),"x"(f2),"x"(f3),"x"(f4),"x"(f5),"m"(h[i+j]));
 
 #define MULSTEP_fromzero(j,h0,h1,h2,h3,h4) \
-  gj = g[j]; \
-  h0 = gj*f0; \
-  _mm256_storeu_ps((float*)&h[i+j], h0); \
-  h1 = gj*f1; \
-  h2 = gj*f2; \
-  h3 = gj*f3; \
-  h4 = gj*f4; \
-  h0 = gj*f5;
-
-#define MULSTEP_fromzero_asm(j,h0,h1,h2,h3,h4) \
   gj = g[j]; \
   __asm__( \
     "vmulps %5,%6,%0 \n\t" \

crypto/libntrup/src/avx/rq.c

@@ -21,11 +21,6 @@
 #define broadcast(r) _mm256_set1_pd(r)
 #define floor(x) _mm256_floor_pd(x)
 
-#ifndef __amd64__
-#define _mm_extract_epi64(X, N) (__extension__ ({ __v2di __a = (__v2di)(X); \
-                                                  __a[N];}))
-#endif
-
 void rq_encode(unsigned char *c,const modq *f)
 {
   crypto_int32 f0, f1, f2, f3, f4;

crypto/libntrup/src/avx/rq_mod3.c

@@ -1,6 +1,5 @@
 #if __AVX2__
 #include <immintrin.h>
-#include <smmintrin.h>
 #include "mod3.h"
 #include "rq.h"
 
@@ -10,11 +9,6 @@
 #define v4591_16 _mm256_set1_epi16(4591)
 #define v10923_16 _mm256_set1_epi16(10923)
 
-#ifndef __amd64__
-#define _mm_extract_epi64(X, N) (__extension__ ({ __v2di __a = (__v2di)(X); \
-                                                  __a[N];}))
-#endif
-
 static inline __m256i squeeze(__m256i x)
 {
   __m256i q = _mm256_mulhrs_epi16(x,v7);