Revert amd_bitalign/amd_bfe polyfills

src/backend/opencl/cl/cn/wolf-aes.cl

@@ -1,12 +1,47 @@
 #ifndef WOLF_AES_CL
 #define WOLF_AES_CL
 
+#ifdef STATIC
+#   undef STATIC
+#endif
 #ifdef cl_amd_media_ops2
-#pragma OPENCL EXTENSION cl_amd_media_ops2 : enable
-#define STATIC static
+#   define STATIC static
+#   pragma OPENCL EXTENSION cl_amd_media_ops2 : enable
 #else
-#define amd_bfe(src0, offset, width)   ((src0 << (32 - (offset) - width)) >> (32 - width))
-#define STATIC
+#   define STATIC
+/* taken from: https://www.khronos.org/registry/OpenCL/extensions/amd/cl_amd_media_ops2.txt
+ *     Built-in Function:
+ *     uintn amd_bfe (uintn src0, uintn src1, uintn src2)
+ *   Description
+ *     NOTE: operator >> below represent logical right shift
+ *     offset = src1.s0 & 31;
+ *     width = src2.s0 & 31;
+ *     if width = 0
+ *         dst.s0 = 0;
+ *     else if (offset + width) < 32
+ *         dst.s0 = (src0.s0 << (32 - offset - width)) >> (32 - width);
+ *     else
+ *         dst.s0 = src0.s0 >> offset;
+ *     similar operation applied to other components of the vectors
+ */
+inline int amd_bfe(const uint src0, const uint offset, const uint width)
+{
+    /* casts are removed because we can implement everything as uint
+     * int offset = src1;
+     * int width = src2;
+     * remove check for edge case, this function is always called with
+     * `width==8`
+     * @code
+     *   if ( width == 0 )
+     *      return 0;
+     * @endcode
+     */
+    if ((offset + width) < 32u) {
+        return (src0 << (32u - offset - width)) >> (32u - width);
+    }
+
+    return src0 >> offset;
+}
 #endif
 
 // AES table - the other three are generated on the fly