Make CELT FFT twiddle complex type aligned

This makes kiss_twiddle_cpx 4-byte aligned (instead of 2-byte) for
fixed-point builds. Tested with an armv6j+nofp development board, CELT
encoding becomes 1.4x as fast, and decoding over 2x.

Performance gain is mostly attributed to the proper alignment of the
static const array mdct_twiddles960.

Co-authored-by: David Gao <davidgao@google.com>
Signed-off-by: Felicia Lim <flim@google.com>

celt/kiss_fft.h

@@ -52,6 +52,10 @@ extern "C" {
 #  define kiss_fft_scalar opus_int32
 #  define kiss_twiddle_scalar opus_int16
 
+/* Some 32-bit CPUs would load/store a kiss_twiddle_cpx with a single memory
+ * access, and could benefit from additional alignment.
+ */
+#  define KISS_TWIDDLE_CPX_ALIGNMENT (sizeof(opus_int32))
 
 #else
 # ifndef kiss_fft_scalar
@@ -62,6 +66,12 @@ extern "C" {
 # endif
 #endif
 
+#if defined(__GNUC__) && defined(KISS_TWIDDLE_CPX_ALIGNMENT)
+#define KISS_TWIDDLE_CPX_ALIGNED __attribute__((aligned(KISS_TWIDDLE_CPX_ALIGNMENT)))
+#else
+#define KISS_TWIDDLE_CPX_ALIGNED
+#endif
+
 typedef struct {
     kiss_fft_scalar r;
     kiss_fft_scalar i;
@@ -70,7 +80,7 @@ typedef struct {
 typedef struct {
    kiss_twiddle_scalar r;
    kiss_twiddle_scalar i;
-}kiss_twiddle_cpx;
+} KISS_TWIDDLE_CPX_ALIGNED kiss_twiddle_cpx;
 
 #define MAXFACTORS 8
 /* e.g. an fft of length 128 has 4 factors