Rewrite i386 backend as hand-written assembly

Replaces the C fallback (borrowed from PowerPC) with dedicated i386
assembly for all nine kernels, matching the amd64/arm64 convention.
Fixes two real bugs found and verified on physical Pentium 4 hardware:
scalar FP silently running on x87 instead of SSE2, and fm_int_math
calling __udivdi3 on every iteration of the highest-weighted test
because the C compiler didn't know its 0xdeadbeef divisor fits in 32
bits. Real-hardware SINGLECORE/MULTICORE scores go from 425/448 to
539/581.

src/fossmark_ppc32.c

@@ -11,10 +11,6 @@
 #include <stdio.h>
 #include <string.h>
 
-#if defined(__i386__) || defined(__x86_64__) || defined(_M_IX86) || defined(_M_X64)
-#include <emmintrin.h>
-#endif
-
 #if defined(__linux__)
 #include <sys/auxv.h>
 #endif
@@ -68,7 +64,7 @@ uint64_t fm_primes(uint64_t limit, uint8_t *sieve)
 	return count;
 }
 
-#if defined(__powerpc__) && !defined(__powerpc64__)
+#if !defined(__powerpc64__)
 static uint64_t fm_simd_scalar(uint64_t iters, void *memory)
 {
 	uint32_t *v = (uint32_t *)memory;
@@ -113,7 +109,7 @@ uint64_t fm_simd(uint64_t iters, void *memory)
 		sum ^= v[j];
 	return sum;
 }
-#elif defined(__powerpc__)
+#else
 /* These are kept in fossmark_ppc32_ext.S so this translation unit, and thus
  * the executable's default code path, only requires baseline PPC32. */
 extern void fm_simd_ps_kernel(uint64_t iters, void *memory);
@@ -196,31 +192,6 @@ uint64_t fm_simd(uint64_t iters, void *memory)
 		sum ^= v[j];
 	return sum;
 }
-#else
-uint64_t fm_simd(uint64_t iters, void *memory)
-{
-	__m128i a, b;
-	uint32_t *v = (uint32_t *)memory;
-	uint32_t sum = 0;
-	uint64_t i;
-	unsigned j;
-
-	if (!iters)
-		return 0;
-	a = _mm_loadu_si128((const __m128i *)v);
-	b = _mm_loadu_si128((const __m128i *)(v + 4));
-	for (i = 0; i < iters; i++) {
-		a = _mm_add_epi32(a, b);
-		b = _mm_xor_si128(b, a);
-		a = _mm_add_epi32(a, b);
-		b = _mm_xor_si128(b, a);
-	}
-	_mm_storeu_si128((__m128i *)v, a);
-	_mm_storeu_si128((__m128i *)(v + 4), b);
-	for (j = 0; j < 8; j++)
-		sum ^= v[j];
-	return sum;
-}
 #endif
 
 static uint32_t load32_native(const uint8_t *p)

src/main.c

@@ -42,7 +42,7 @@
 #ifndef FM_API_BASE_URL
 #  define FM_API_BASE_URL "https://fossbench.net"
 #endif
-#define FM_VERSION "0.1.3-hotfix4"
+#define FM_VERSION "0.1.4"
 
 /* ---------- platform identification (for the banner only) ---------- */