diff src/madplug/SFMT-sse2.h @ 1386:b8dd67ad7b86

update SFMT files to version 1.3. please let me know if it break on altivec box.

--- a/src/madplug/SFMT-sse2.h	Fri Jul 27 04:53:35 2007 +0200
+++ b/src/madplug/SFMT-sse2.h	Fri Jul 27 18:56:43 2007 +0900
@@ -1,36 +1,121 @@
 /** 
- * @file SFMT-sse2.h 
- *
- * @brief SIMD oriented Fast Mersenne Twister(SFMT)
- * pseudorandom number generator
+ * @file  SFMT-sse2.h
+ * @brief SIMD oriented Fast Mersenne Twister(SFMT) for Intel SSE2
  *
  * @author Mutsuo Saito (Hiroshima University)
  * @author Makoto Matsumoto (Hiroshima University)
  *
- * Copyright (C) 2007 Mutsuo Saito, Makoto Matsumoto and Hiroshima
+ * @note We assume LITTLE ENDIAN in this file
+ *
+ * Copyright (C) 2006, 2007 Mutsuo Saito, Makoto Matsumoto and Hiroshima
  * University. All rights reserved.
  *
- * The new BSD License is applied to this software.
- * see LICENSE.txt
+ * The new BSD License is applied to this software, see LICENSE.txt
  */
 
 #ifndef SFMT_SSE2_H
 #define SFMT_SSE2_H
-#include <emmintrin.h>
+
+inline static __m128i mm_recursion(__m128i *a, __m128i *b, __m128i c,
+				   __m128i d, __m128i mask) ALWAYSINLINE;
 
-union W128_T {
-    __m128i si;
-    uint32_t u[4];
-};
+/**
+ * This function represents the recursion formula.
+ * @param a a 128-bit part of the interal state array
+ * @param b a 128-bit part of the interal state array
+ * @param c a 128-bit part of the interal state array
+ * @param d a 128-bit part of the interal state array
+ * @param mask 128-bit mask
+ * @return output
+ */
+inline static __m128i mm_recursion(__m128i *a, __m128i *b, 
+				   __m128i c, __m128i d, __m128i mask) {
+    __m128i v, x, y, z;
+    
+    x = _mm_load_si128(a);
+    y = _mm_srli_epi32(*b, SR1);
+    z = _mm_srli_si128(c, SR2);
+    v = _mm_slli_epi32(d, SL1);
+    z = _mm_xor_si128(z, x);
+    z = _mm_xor_si128(z, v);
+    x = _mm_slli_si128(x, SL2);
+    y = _mm_and_si128(y, mask);
+    z = _mm_xor_si128(z, x);
+    z = _mm_xor_si128(z, y);
+    return z;
+}
 
-typedef union W128_T w128_t;
+/**
+ * This function fills the internal state array with pseudorandom
+ * integers.
+ */
+inline static void gen_rand_all(void) {
+    int i;
+    __m128i r, r1, r2, mask;
+    mask = _mm_set_epi32(MSK4, MSK3, MSK2, MSK1);
 
-#ifdef __GNUC__
-inline static __m128i mm_recursion(__m128i *a, __m128i *b, __m128i c,
-				   __m128i d, __m128i mask)
-    __attribute__((always_inline));
-#else
-inline static __m128i mm_recursion(__m128i *a, __m128i *b,
-				   __m128i c, __m128i d, __m128i mask);
+    r1 = _mm_load_si128(&sfmt[N - 2].si);
+    r2 = _mm_load_si128(&sfmt[N - 1].si);
+    for (i = 0; i < N - POS1; i++) {
+	r = mm_recursion(&sfmt[i].si, &sfmt[i + POS1].si, r1, r2, mask);
+	_mm_store_si128(&sfmt[i].si, r);
+	r1 = r2;
+	r2 = r;
+    }
+    for (; i < N; i++) {
+	r = mm_recursion(&sfmt[i].si, &sfmt[i + POS1 - N].si, r1, r2, mask);
+	_mm_store_si128(&sfmt[i].si, r);
+	r1 = r2;
+	r2 = r;
+    }
+}
+
+/**
+ * This function fills the user-specified array with pseudorandom
+ * integers.
+ *
+ * @param array an 128-bit array to be filled by pseudorandom numbers.  
+ * @param size number of 128-bit pesudorandom numbers to be generated.
+ */
+inline static void gen_rand_array(w128_t *array, int size) {
+    int i, j;
+    __m128i r, r1, r2, mask;
+    mask = _mm_set_epi32(MSK4, MSK3, MSK2, MSK1);
+
+    r1 = _mm_load_si128(&sfmt[N - 2].si);
+    r2 = _mm_load_si128(&sfmt[N - 1].si);
+    for (i = 0; i < N - POS1; i++) {
+	r = mm_recursion(&sfmt[i].si, &sfmt[i + POS1].si, r1, r2, mask);
+	_mm_store_si128(&array[i].si, r);
+	r1 = r2;
+	r2 = r;
+    }
+    for (; i < N; i++) {
+	r = mm_recursion(&sfmt[i].si, &array[i + POS1 - N].si, r1, r2, mask);
+	_mm_store_si128(&array[i].si, r);
+	r1 = r2;
+	r2 = r;
+    }
+    /* main loop */
+    for (; i < size - N; i++) {
+	r = mm_recursion(&array[i - N].si, &array[i + POS1 - N].si, r1, r2,
+			 mask);
+	_mm_store_si128(&array[i].si, r);
+	r1 = r2;
+	r2 = r;
+    }
+    for (j = 0; j < 2 * N - size; j++) {
+	r = _mm_load_si128(&array[j + size - N].si);
+	_mm_store_si128(&sfmt[j].si, r);
+    }
+    for (; i < size; i++) {
+	r = mm_recursion(&array[i - N].si, &array[i + POS1 - N].si, r1, r2,
+			 mask);
+	_mm_store_si128(&array[i].si, r);
+	_mm_store_si128(&sfmt[j++].si, r);
+	r1 = r2;
+	r2 = r;
+    }
+}
+
 #endif
-#endif