Correct vec_perm() application on little-endian 64-bit PowerPC

The LE transformation for vec_perm has an implicit assumption that the
permutation is being used to reorder vector elements (in this case 4-byte
integer word elements), not to reorder bytes within those elements.  Although
this is legal behavior, it is not anticipated by the transformation performed
by the compilers.

This causes pygame-1.9.1 test failure on PPC64LE because blitted pixmaps are
corrupted there due to how SDL uses vec_perm().

<https://bugzilla.redhat.com/show_bug.cgi?id=1392465>

--- SDL-1.2.15/src/video/SDL_blit_N.c.ori	2017-09-04 05:56:17.759347525 -0400
+++ SDL-1.2.15/src/video/SDL_blit_N.c	2017-09-06 05:36:20.570789610 -0400
@@ -146,6 +146,32 @@ static vector unsigned char calc_swizzle
     return(vswiz);
 }
 
+/* reorder bytes for PowerPC little endian */
+static vector unsigned char reorder_ppc64le_vec(vector unsigned char vpermute)
+{
+    /* The result vector of calc_swizzle32 reorder bytes using vec_perm.
+       The LE transformation for vec_perm has an implicit assumption
+       that the permutation is being used to reorder vector elements,
+       not to reorder bytes within those elements.  
+       Unfortunatly the result order is not the expected one for powerpc
+       little endian when the two first vector parameters of vec_perm are
+       not of type 'vector char'. This is because the numbering from the
+       left for BE, and numbering from the right for LE, produces a
+       different interpretation of what the odd and even lanes are.
+       Refer to fedora bug 1392465
+     */
+
+    const vector unsigned char ppc64le_reorder = VECUINT8_LITERAL(
+                                      0x01, 0x00, 0x03, 0x02,
+                                      0x05, 0x04, 0x07, 0x06,
+                                      0x09, 0x08, 0x0B, 0x0A,
+                                      0x0D, 0x0C, 0x0F, 0x0E );
+
+    vector unsigned char vswiz_ppc64le;
+    vswiz_ppc64le = vec_perm(vpermute, vpermute, ppc64le_reorder);
+    return(vswiz_ppc64le);
+}
+
 static void Blit_RGB888_RGB565(SDL_BlitInfo *info);
 static void Blit_RGB888_RGB565Altivec(SDL_BlitInfo *info) {
     int height = info->d_height;
@@ -631,6 +657,12 @@ static void Blit32to32KeyAltivec(SDL_Bli
                 vsel = (vector unsigned char)vec_and(vs, vrgbmask);
                 vsel = (vector unsigned char)vec_cmpeq(vs, vckey);
                 /* permute the src vec to the dest format */
+
+#if defined(__powerpc__) && (SDL_BYTEORDER == SDL_LIL_ENDIAN)
+            /* reorder bytes for PowerPC little endian */
+            vpermute = reorder_ppc64le_vec(vpermute);
+#endif
+
                 vs = vec_perm(vs, valpha, vpermute);
                 /* load the destination vec */
                 vd = vec_ld(0, dstp);
@@ -704,6 +736,12 @@ static void ConvertAltivec32to32_noprefe
             src += 4;
             width -= 4;
             vbits = vec_perm(vbits, voverflow, valigner);  /* src is ready. */
+
+#if defined(__powerpc__) && (SDL_BYTEORDER == SDL_LIL_ENDIAN)
+            /* reorder bytes for PowerPC little endian */
+            vpermute = reorder_ppc64le_vec(vpermute);
+#endif
+
             vbits = vec_perm(vbits, vzero, vpermute);  /* swizzle it. */
             vec_st(vbits, 0, dst);  /* store it back out. */
             dst += 4;
@@ -786,6 +824,12 @@ static void ConvertAltivec32to32_prefetc
             src += 4;
             width -= 4;
             vbits = vec_perm(vbits, voverflow, valigner);  /* src is ready. */
+
+#if defined(__powerpc__) && (SDL_BYTEORDER == SDL_LIL_ENDIAN) 
+            /* reorder bytes for PowerPC little endian */
+            vpermute = reorder_ppc64le_vec(vpermute);
+#endif
+
             vbits = vec_perm(vbits, vzero, vpermute);  /* swizzle it. */
             vec_st(vbits, 0, dst);  /* store it back out. */
             dst += 4;