[opus] [[RFC PATCH v2]: Ne10 fft fixed and previous 7/8] armv7, armv8: Extend fixed fft NE10 optimizations to mdct
Viswanath Puttagunta
viswanath.puttagunta at linaro.org
Fri May 8 09:52:58 PDT 2015
Extends usage of Neon optimized fixed point fft optimizations
in libNE10 to clt_mdct_forward and clt_mdct_backward.
Signed-off-by: Viswanath Puttagunta <viswanath.puttagunta at linaro.org>
---
celt/arm/arm_celt_map.c | 48 ++++++++++++++++++++++++-----------------------
celt/arm/celt_ne10_mdct.c | 23 ++++++++++-------------
celt/arm/mdct_arm.h | 32 +++++++++++++++++--------------
celt/mdct.h | 5 ++---
4 files changed, 55 insertions(+), 53 deletions(-)
diff --git a/celt/arm/arm_celt_map.c b/celt/arm/arm_celt_map.c
index af54b9d..41b18b1 100644
--- a/celt/arm/arm_celt_map.c
+++ b/celt/arm/arm_celt_map.c
@@ -69,27 +69,7 @@ void (*const OPUS_IFFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg,
opus_ifft_c, /* Media */
opus_ifft_neon /* Neon with NE10 */
};
-#endif
-
-# if defined(FIXED_POINT)
-opus_val32 (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *,
- const opus_val16 *, opus_val32 *, int , int) = {
- celt_pitch_xcorr_c, /* ARMv4 */
- MAY_HAVE_EDSP(celt_pitch_xcorr), /* EDSP */
- MAY_HAVE_MEDIA(celt_pitch_xcorr), /* Media */
- MAY_HAVE_NEON(celt_pitch_xcorr) /* NEON */
-};
-# else /* !FIXED_POINT */
-# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR)
-void (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *,
- const opus_val16 *, opus_val32 *, int, int) = {
- celt_pitch_xcorr_c, /* ARMv4 */
- celt_pitch_xcorr_c, /* EDSP */
- celt_pitch_xcorr_c, /* Media */
- celt_pitch_xcorr_float_neon /* Neon */
-};
-#if defined(HAVE_ARM_NE10)
void (*const CLT_MDCT_FORWARD_IMPL[OPUS_ARCHMASK+1])(const mdct_lookup *l,
kiss_fft_scalar *in,
kiss_fft_scalar * OPUS_RESTRICT out,
@@ -99,7 +79,11 @@ void (*const CLT_MDCT_FORWARD_IMPL[OPUS_ARCHMASK+1])(const mdct_lookup *l,
clt_mdct_forward_c, /* ARMv4 */
clt_mdct_forward_c, /* EDSP */
clt_mdct_forward_c, /* Media */
- clt_mdct_forward_float_neon /* Neon with NE10 */
+#if defined(FIXED_POINT)
+ clt_mdct_forward_c, /* Debugging issues with NE10 library for fixed point */
+#else
+ clt_mdct_forward_neon /* Neon with NE10 */
+#endif
};
void (*const CLT_MDCT_BACKWARD_IMPL[OPUS_ARCHMASK+1])(const mdct_lookup *l,
@@ -111,10 +95,28 @@ void (*const CLT_MDCT_BACKWARD_IMPL[OPUS_ARCHMASK+1])(const mdct_lookup *l,
clt_mdct_backward_c, /* ARMv4 */
clt_mdct_backward_c, /* EDSP */
clt_mdct_backward_c, /* Media */
- clt_mdct_backward_float_neon /* Neon with NE10 */
+ clt_mdct_backward_neon, /* Neon with NE10 */
};
-#endif /* HAVE_ARM_NE10 */
+#endif
+
+# if defined(FIXED_POINT)
+opus_val32 (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *,
+ const opus_val16 *, opus_val32 *, int , int) = {
+ celt_pitch_xcorr_c, /* ARMv4 */
+ MAY_HAVE_EDSP(celt_pitch_xcorr), /* EDSP */
+ MAY_HAVE_MEDIA(celt_pitch_xcorr), /* Media */
+ MAY_HAVE_NEON(celt_pitch_xcorr) /* NEON */
+};
+# else /* !FIXED_POINT */
+# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR)
+void (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *,
+ const opus_val16 *, opus_val32 *, int, int) = {
+ celt_pitch_xcorr_c, /* ARMv4 */
+ celt_pitch_xcorr_c, /* EDSP */
+ celt_pitch_xcorr_c, /* Media */
+ celt_pitch_xcorr_float_neon /* Neon */
+};
# endif /* OPUS_ARM_MAY_HAVE_NEON_INTR */
# endif /* FIXED_POINT */
diff --git a/celt/arm/celt_ne10_mdct.c b/celt/arm/celt_ne10_mdct.c
index 938fc93..a445088 100644
--- a/celt/arm/celt_ne10_mdct.c
+++ b/celt/arm/celt_ne10_mdct.c
@@ -43,13 +43,11 @@
#include "os_support.h"
#include "stack_alloc.h"
-#if !defined(FIXED_POINT)
-
-void clt_mdct_forward_float_neon(const mdct_lookup *l,
- kiss_fft_scalar *in,
- kiss_fft_scalar * OPUS_RESTRICT out,
- const opus_val16 *window,
- int overlap, int shift, int stride, int arch)
+void clt_mdct_forward_neon(const mdct_lookup *l,
+ kiss_fft_scalar *in,
+ kiss_fft_scalar * OPUS_RESTRICT out,
+ const opus_val16 *window,
+ int overlap, int shift, int stride, int arch)
{
int i;
int N, N2, N4;
@@ -159,11 +157,11 @@ void clt_mdct_forward_float_neon(const mdct_lookup *l,
RESTORE_STACK;
}
-void clt_mdct_backward_float_neon(const mdct_lookup *l,
- kiss_fft_scalar *in,
- kiss_fft_scalar * OPUS_RESTRICT out,
- const opus_val16 * OPUS_RESTRICT window,
- int overlap, int shift, int stride, int arch)
+void clt_mdct_backward_neon(const mdct_lookup *l,
+ kiss_fft_scalar *in,
+ kiss_fft_scalar * OPUS_RESTRICT out,
+ const opus_val16 * OPUS_RESTRICT window,
+ int overlap, int shift, int stride, int arch)
{
int i;
int N, N2, N4;
@@ -260,4 +258,3 @@ void clt_mdct_backward_float_neon(const mdct_lookup *l,
}
RESTORE_STACK;
}
-#endif /* !defined(FIXED_POINT) */
diff --git a/celt/arm/mdct_arm.h b/celt/arm/mdct_arm.h
index db32efe..f43c28c 100644
--- a/celt/arm/mdct_arm.h
+++ b/celt/arm/mdct_arm.h
@@ -36,25 +36,29 @@
#include "config.h"
#include "mdct.h"
-#if !defined(FIXED_POINT) && defined(HAVE_ARM_NE10)
+#if defined(HAVE_ARM_NE10)
/** Compute a forward MDCT and scale by 4/N, trashes the input array */
-void clt_mdct_forward_float_neon(const mdct_lookup *l, kiss_fft_scalar *in,
- kiss_fft_scalar * OPUS_RESTRICT out,
- const opus_val16 *window, int overlap,
- int shift, int stride, int arch);
-
-void clt_mdct_backward_float_neon(const mdct_lookup *l, kiss_fft_scalar *in,
- kiss_fft_scalar * OPUS_RESTRICT out,
- const opus_val16 *window, int overlap,
- int shift, int stride, int arch);
-
+void clt_mdct_forward_neon(const mdct_lookup *l, kiss_fft_scalar *in,
+ kiss_fft_scalar * OPUS_RESTRICT out,
+ const opus_val16 *window, int overlap,
+ int shift, int stride, int arch);
+void clt_mdct_backward_neon(const mdct_lookup *l, kiss_fft_scalar *in,
+ kiss_fft_scalar * OPUS_RESTRICT out,
+ const opus_val16 *window, int overlap,
+ int shift, int stride, int arch);
#if !defined(OPUS_HAVE_RTCD)
#define OVERRIDE_OPUS_MDCT (1)
+#if defined(FIXED_POINT)
+/* Debugging corner cases in mdct forward issues with NE10 library */
#define clt_mdct_forward(_l, _in, _out, _window, _int, _shift, _stride, _arch) \
- clt_mdct_forward_float_neon(_l, _in, _out, _window, _int, _shift, _stride, _arch)
+ clt_mdct_forward_c(_l, _in, _out, _window, _int, _shift, _stride, _arch)
+#else
+#define clt_mdct_forward(_l, _in, _out, _window, _int, _shift, _stride, _arch) \
+ clt_mdct_forward_neon(_l, _in, _out, _window, _int, _shift, _stride, _arch)
+#endif
#define clt_mdct_backward(_l, _in, _out, _window, _int, _shift, _stride, _arch) \
- clt_mdct_backward_float_neon(_l, _in, _out, _window, _int, _shift, _stride, _arch)
+ clt_mdct_backward_neon(_l, _in, _out, _window, _int, _shift, _stride, _arch)
#endif /* OPUS_HAVE_RTCD */
-#endif /* !defined(FIXED_POINT) && defined(HAVE_ARM_NE10) */
+#endif /* defined(HAVE_ARM_NE10) */
#endif
diff --git a/celt/mdct.h b/celt/mdct.h
index 3ced019..df8bcd2 100644
--- a/celt/mdct.h
+++ b/celt/mdct.h
@@ -53,7 +53,7 @@ typedef struct {
const kiss_twiddle_scalar * OPUS_RESTRICT trig;
} mdct_lookup;
-#if !defined(FIXED_POINT) && defined(HAVE_ARM_NE10)
+#if defined(HAVE_ARM_NE10)
#include "arm/mdct_arm.h"
#endif
@@ -76,8 +76,7 @@ void clt_mdct_backward_c(const mdct_lookup *l, kiss_fft_scalar *in,
#if !defined(OVERRIDE_OPUS_MDCT)
/* Is run-time CPU detection enabled on this platform? */
-#if defined(OPUS_HAVE_RTCD) && defined(HAVE_ARM_NE10) && !defined(FIXED_POINT)
-
+#if defined(OPUS_HAVE_RTCD) && defined(HAVE_ARM_NE10)
void (*const CLT_MDCT_FORWARD_IMPL[OPUS_ARCHMASK+1])(const mdct_lookup *l,
kiss_fft_scalar *in,
kiss_fft_scalar * OPUS_RESTRICT out,
--
1.9.1
More information about the opus
mailing list