Fix 32-bit builds breakage from Stribog

[magnumripper]

Problem introduced by ebbbfc1a8

One possibility (although not the best) is to just #ifdef the whole file outside of the plugin stanza. Similar to eg. rawSHA1_ng_fmt_plug.c but in this case checking for eg. ARCH_SIZE after including arch.h

But it should be very trivial to actually fix it instead. Why does it break at all?

[jfoug]

Why does it break at all?

The intrinsic _mm_insert_epi64 is not a valid instruction under 32 bit (tested cygwin and ubuntu 32 bit)

Also, the intrisic _mm_cvtsi64_si128 is not in 32 bit land.

But it should be very trivial to actually fix it instead.

The fix 'could' be to build our own _mm_insert_epi64 that has #ifdef for 32 bit wrapper around it.

[jfoug]

This patch at least allows this file to be built on 32 bit systems. It will fail however. NOTE, I think this will build properly on a < sse4.1 system, but if run there, it will core.

$ git diff gost3411-2012-sse41_plug.c
diff --git a/src/gost3411-2012-sse41_plug.c b/src/gost3411-2012-sse41_plug.c
index 97a91e0..7e16305 100644
--- a/src/gost3411-2012-sse41_plug.c
+++ b/src/gost3411-2012-sse41_plug.c
@@ -22,6 +22,15 @@
 #include "gost3411-tables.h"
 #include "gost3411-2012-sse41.h"

+#include "arch.h"
+
+#if ARCH_BITS==32
+// this is NOT right, but at least it allows 32 bit to build.
+// NOTE, this file depends upon SSE4.1, so it should NOT be a *plug.c
+static inline __m128i _mm_cvtsi64_si128(uint64_t b) { __m128i a; return a; }
+static inline __m128i _mm_insert_epi64(__m128i a, uint64_t b, int c) { return a; }
+#endif
+
 static inline void add512(const union uint512_u* x, const union uint512_u* y, union uint512_u* r)
 {
        uint_fast8_t i, CF;

[magnumripper]

NOTE, this file depends upon SSE4.1, so it should NOT be a *plug.c

Plugs can easily depend on things like SSE4.1. Look at rawSHA1_ng_fmt_plug.c just before plugin stanza.

[magnumripper]

Should be easy to write proper (albeit slow) emulation functions though. With some luck we may already have something usable in pseudo-intrinsics.c

[magnumripper]

This is proper emulation (works fine):

static inline __m128i _mm_cvtsi64_si128(long long a) {
    return _mm_set_epi32(0, 0, (unsigned int)(a >> 32), (unsigned int)a);
}

static inline __m128i _mm_insert_epi64(__m128i a, uint64_t b, int c) {
    c <<= 1;
    a = _mm_insert_epi32(a, (unsigned int)b, c);
    return _mm_insert_epi32(a, (unsigned int)(b >> 32), c + 1);
}