Fix issue #29 "Make FFTS work on all architectures"

Modify macros-alpha.h to provide scalar operations on all platforms.
Using union and memcpy to avoid strict aliasing issues.

1 files changed, 121 insertions, 93 deletions

diff --git a/src/macros-alpha.h b/src/macros-alpha.h
index be5ec20..f4efaf8 100644
--- a/src/macros-alpha.h
+++ b/src/macros-alpha.h
@@ -1,10 +1,10 @@
 /*
- 
+
  This file is part of FFTS -- The Fastest Fourier Transform in the South
-  
- Copyright (c) 2013, Michael J. Cree <mcree@orcon.net.nz> 
+
+ Copyright (c) 2013, Michael J. Cree <mcree@orcon.net.nz>
  Copyright (c) 2012, 2013, Anthony M. Blake <amb@anthonix.com>
- 
+
  All rights reserved.
 
  Redistribution and use in source and binary forms, with or without
@@ -31,177 +31,205 @@
 
 */
 
-#ifndef __MACROS_ALPHA_H__
-#define __MACROS_ALPHA_H__
+#ifndef FFTS_MACROS_ALPHA_H
+#define FFTS_MACROS_ALPHA_H
 
-#include <math.h>
+#include <string.h>
 
-#ifdef __alpha__
-#define restrict
-#endif
-
-typedef struct {float r1, i1, r2, i2;} V;
+typedef union {
+    struct {
+        float r1;
+        float i1;
+        float r2;
+        float i2;
+    } r;
+    uint32_t u[4];
+} V;
 
 #define FFTS_MALLOC(d,a) malloc(d)
 #define FFTS_FREE(d) free(d)
 
-#define VLIT4(f3,f2,f1,f0) ((V){f0,f1,f2,f3})
-
-static inline V VADD(V x, V y)
+static FFTS_ALWAYS_INLINE V VLIT4(float f3, float f2, float f1, float f0)
 {
     V z;
-    z.r1 = x.r1 + y.r1;
-    z.i1 = x.i1 + y.i1;
-    z.r2 = x.r2 + y.r2;
-    z.i2 = x.i2 + y.i2;
+
+    z.r.r1 = f0;
+    z.r.i1 = f1;
+    z.r.r2 = f2;
+    z.r.i2 = f3;
+
     return z;
 }
 
-
-static inline V VSUB(V x, V y)
+static FFTS_ALWAYS_INLINE V VADD(V x, V y)
 {
     V z;
-    z.r1 = x.r1 - y.r1;
-    z.i1 = x.i1 - y.i1;
-    z.r2 = x.r2 - y.r2;
-    z.i2 = x.i2 - y.i2;
+
+    z.r.r1 = x.r.r1 + y.r.r1;
+    z.r.i1 = x.r.i1 + y.r.i1;
+    z.r.r2 = x.r.r2 + y.r.r2;
+    z.r.i2 = x.r.i2 + y.r.i2;
+
     return z;
 }
 
-
-static inline V VMUL(V x, V y)
+static FFTS_ALWAYS_INLINE V VSUB(V x, V y)
 {
     V z;
-    z.r1 = x.r1 * y.r1;
-    z.i1 = x.i1 * y.i1;
-    z.r2 = x.r2 * y.r2;
-    z.i2 = x.i2 * y.i2;
+
+    z.r.r1 = x.r.r1 - y.r.r1;
+    z.r.i1 = x.r.i1 - y.r.i1;
+    z.r.r2 = x.r.r2 - y.r.r2;
+    z.r.i2 = x.r.i2 - y.r.i2;
+
     return z;
 }
 
-static inline V VXOR(V x, V y)
+static FFTS_ALWAYS_INLINE V VMUL(V x, V y)
 {
-    V r;
-    r.r1 = (uint32_t)x.r1 ^ (uint32_t)y.r1;
-    r.i1 = (uint32_t)x.i1 ^ (uint32_t)y.i1;
-    r.r2 = (uint32_t)x.r2 ^ (uint32_t)y.r2;
-    r.i2 = (uint32_t)x.i2 ^ (uint32_t)y.i2;
-    return r;
+    V z;
+
+    z.r.r1 = x.r.r1 * y.r.r1;
+    z.r.i1 = x.r.i1 * y.r.i1;
+    z.r.r2 = x.r.r2 * y.r.r2;
+    z.r.i2 = x.r.i2 * y.r.i2;
+
+    return z;
 }
 
-static inline V VSWAPPAIRS(V x)
+static FFTS_ALWAYS_INLINE V VXOR(V x, V y)
 {
     V z;
-    z.r1 = x.i1;
-    z.i1 = x.r1;
-    z.r2 = x.i2;
-    z.i2 = x.r2;
+
+    z.u[0] = x.u[0] ^ y.u[0];
+    z.u[1] = x.u[1] ^ y.u[1];
+    z.u[2] = x.u[2] ^ y.u[2];
+    z.u[3] = x.u[3] ^ y.u[3];
+
     return z;
 }
 
+static FFTS_ALWAYS_INLINE V VSWAPPAIRS(V x)
+{
+    V z;
+
+    z.r.r1 = x.r.i1;
+    z.r.i1 = x.r.r1;
+    z.r.r2 = x.r.i2;
+    z.r.i2 = x.r.r2;
+
+    return z;
+}
 
-static inline V VBLEND(V x, V y)
+static FFTS_ALWAYS_INLINE V VBLEND(V x, V y)
 {
     V z;
-    z.r1 = x.r1;
-    z.i1 = x.i1;
-    z.r2 = y.r2;
-    z.i2 = y.i2;
+
+    z.r.r1 = x.r.r1;
+    z.r.i1 = x.r.i1;
+    z.r.r2 = y.r.r2;
+    z.r.i2 = y.r.i2;
+
     return z;
 }
 
-static inline V VUNPACKHI(V x, V y)
+static FFTS_ALWAYS_INLINE V VUNPACKHI(V x, V y)
 {
     V z;
-    z.r1 = x.r2;
-    z.i1 = x.i2;
-    z.r2 = y.r2;
-    z.i2 = y.i2;
+
+    z.r.r1 = x.r.r2;
+    z.r.i1 = x.r.i2;
+    z.r.r2 = y.r.r2;
+    z.r.i2 = y.r.i2;
+
     return z;
 }
 
-static inline V VUNPACKLO(V x, V y)
+static FFTS_ALWAYS_INLINE V VUNPACKLO(V x, V y)
 {
     V z;
-    z.r1 = x.r1;
-    z.i1 = x.i1;
-    z.r2 = y.r1;
-    z.i2 = y.i1;
+
+    z.r.r1 = x.r.r1;
+    z.r.i1 = x.r.i1;
+    z.r.r2 = y.r.r1;
+    z.r.i2 = y.r.i1;
+
     return z;
 }
 
-static inline V VDUPRE(V x)
+static FFTS_ALWAYS_INLINE V VDUPRE(V x)
 {
     V z;
-    z.r1 = x.r1;
-    z.i1 = x.r1;
-    z.r2 = x.r2;
-    z.i2 = x.r2;
+
+    z.r.r1 = x.r.r1;
+    z.r.i1 = x.r.r1;
+    z.r.r2 = x.r.r2;
+    z.r.i2 = x.r.r2;
+
     return z;
 }
 
-static inline V VDUPIM(V x)
+static FFTS_ALWAYS_INLINE V VDUPIM(V x)
 {
     V z;
-    z.r1 = x.i1;
-    z.i1 = x.i1;
-    z.r2 = x.i2;
-    z.i2 = x.i2;
+
+    z.r.r1 = x.r.i1;
+    z.r.i1 = x.r.i1;
+    z.r.r2 = x.r.i2;
+    z.r.i2 = x.r.i2;
+
     return z;
 }
 
-static inline V IMUL(V d, V re, V im)
+static FFTS_ALWAYS_INLINE V IMUL(V d, V re, V im)
 {
     re = VMUL(re, d);
     im = VMUL(im, VSWAPPAIRS(d));
-    return VSUB(re, im);  
+    return VSUB(re, im);
 }
 
-
-static inline V IMULJ(V d, V re, V im)
+static FFTS_ALWAYS_INLINE V IMULJ(V d, V re, V im)
 {
     re = VMUL(re, d);
     im = VMUL(im, VSWAPPAIRS(d));
     return VADD(re, im);
 }
 
-static inline V MULI(int inv, V x)
+static FFTS_ALWAYS_INLINE V MULI(int inv, V x)
 {
     V z;
 
     if (inv) {
-	z.r1 = -x.r1;
-	z.i1 = x.i1;
-	z.r2 = -x.r2;
-	z.i2 = x.i2;
-    }else{
-	z.r1 = x.r1;
-	z.i1 = -x.i1;
-	z.r2 = x.r2;
-	z.i2 = -x.i2;
+        z.r.r1 = -x.r.r1;
+        z.r.i1 =  x.r.i1;
+        z.r.r2 = -x.r.r2;
+        z.r.i2 =  x.r.i2;
+    } else {
+        z.r.r1 =  x.r.r1;
+        z.r.i1 = -x.r.i1;
+        z.r.r2 =  x.r.r2;
+        z.r.i2 = -x.r.i2;
     }
+
     return z;
 }
 
-
-static inline V IMULI(int inv, V x)
+static FFTS_ALWAYS_INLINE V IMULI(int inv, V x)
 {
     return VSWAPPAIRS(MULI(inv, x));
 }
 
-
-static inline V VLD(const void *s)
+static FFTS_ALWAYS_INLINE V VLD(const void *s)
 {
-    V *d = (V *)s;
-    return *d;
+    V z;
+    memcpy(&z, s, sizeof(z));
+    return z;
 }
 
-
-static inline void VST(void *d, V s)
+static FFTS_ALWAYS_INLINE void VST(void *d, V s)
 {
-    V *r = (V *)d;
+    V *r = (V*) d;
     *r = s;
 }
 
-#endif
-// vim: set autoindent noexpandtab tabstop=3 shiftwidth=3:
+#endif /* FFTS_MACROS_ALPHA_H */
\ No newline at end of file