shithub: opus

Download patch

ref: 85fb8a1d17a563d1e7989e8268a856a5c90ecfe7
parent: a1ae821c964960ed659134e8f834ba31a3db1a6b
author: Linfeng Zhang <linfengz@google.com>
date: Wed Sep 7 11:44:23 EDT 2016 

Replace call of celt_inner_prod_c() (step 2)

Should call celt_inner_prod().
This requires the API change of celt_pitch_xcorr() by passing in
"arch".

We tested on x86 and arm, and got bit exact results as original.

Change-Id: I606915da6a196f327ce81f4a5ae32811f4c1fabb

Signed-off-by: Jean-Marc Valin <jmvalin@jmvalin.ca>


--- a/celt/arm/arm_celt_map.c
+++ b/celt/arm/arm_celt_map.c
@@ -40,7 +40,7 @@
     (defined(OPUS_ARM_MAY_HAVE_MEDIA) && !defined(OPUS_ARM_PRESUME_MEDIA)) || \
     (defined(OPUS_ARM_MAY_HAVE_EDSP) && !defined(OPUS_ARM_PRESUME_EDSP)))
 opus_val32 (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *,
-    const opus_val16 *, opus_val32 *, int , int) = {
+    const opus_val16 *, opus_val32 *, int, int, int) = {
   celt_pitch_xcorr_c,               /* ARMv4 */
   MAY_HAVE_EDSP(celt_pitch_xcorr),  /* EDSP */
   MAY_HAVE_MEDIA(celt_pitch_xcorr), /* Media */
@@ -51,7 +51,7 @@
 # else /* !FIXED_POINT */
 #  if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR)
 void (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *,
-    const opus_val16 *, opus_val32 *, int, int) = {
+    const opus_val16 *, opus_val32 *, int, int, int) = {
   celt_pitch_xcorr_c,              /* ARMv4 */
   celt_pitch_xcorr_c,              /* EDSP */
   celt_pitch_xcorr_c,              /* Media */
--- a/celt/arm/celt_neon_intr.c
+++ b/celt/arm/celt_neon_intr.c
@@ -290,8 +290,9 @@
 }
 
 void celt_pitch_xcorr_float_neon(const opus_val16 *_x, const opus_val16 *_y,
-                        opus_val32 *xcorr, int len, int max_pitch) {
+                        opus_val32 *xcorr, int len, int max_pitch, int arch) {
    int i;
+   (void)arch;
    celt_assert(max_pitch > 0);
    celt_assert((((unsigned char *)_x-(unsigned char *)NULL)&3)==0);
 
--- a/celt/arm/celt_pitch_xcorr_arm.s
+++ b/celt/arm/celt_pitch_xcorr_arm.s
@@ -153,7 +153,7 @@
   ENDP
 
 ; opus_val32 celt_pitch_xcorr_neon(opus_val16 *_x, opus_val16 *_y,
-;  opus_val32 *xcorr, int len, int max_pitch)
+;  opus_val32 *xcorr, int len, int max_pitch, int arch)
 celt_pitch_xcorr_neon PROC
   ; input:
   ;   r0  = opus_val16 *_x
@@ -168,6 +168,8 @@
   ;   r6  = int         max_pitch
   ;   r12 = int         j
   ;   q15 = int         maxcorr[4] (q15 is not used by xcorr_kernel_neon())
+  ; ignored:
+  ;         int         arch
   STMFD        sp!, {r4-r6, lr}
   LDR          r6, [sp, #16]
   VMOV.S32     q15, #1
@@ -358,6 +360,8 @@
   ;   r9  = opus_val32  sum3
   ;   r1  = int         max_pitch
   ;   r12 = int         j
+  ; ignored:
+  ;         int         arch
   STMFD        sp!, {r4-r11, lr}
   MOV          r5, r1
   LDR          r1, [sp, #36]
--- a/celt/arm/pitch_arm.h
+++ b/celt/arm/pitch_arm.h
@@ -34,7 +34,7 @@
 
 #  if defined(OPUS_ARM_MAY_HAVE_NEON)
 opus_val32 celt_pitch_xcorr_neon(const opus_val16 *_x, const opus_val16 *_y,
-    opus_val32 *xcorr, int len, int max_pitch);
+    opus_val32 *xcorr, int len, int max_pitch, int arch);
 #  endif
 
 #  if defined(OPUS_ARM_MAY_HAVE_MEDIA)
@@ -43,7 +43,7 @@
 
 #  if defined(OPUS_ARM_MAY_HAVE_EDSP)
 opus_val32 celt_pitch_xcorr_edsp(const opus_val16 *_x, const opus_val16 *_y,
-    opus_val32 *xcorr, int len, int max_pitch);
+    opus_val32 *xcorr, int len, int max_pitch, int arch);
 #  endif
 
 #  if defined(OPUS_HAVE_RTCD) && \
@@ -52,18 +52,15 @@
      (defined(OPUS_ARM_MAY_HAVE_EDSP) && !defined(OPUS_ARM_PRESUME_EDSP)))
 extern opus_val32
 (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *,
-      const opus_val16 *, opus_val32 *, int, int);
+      const opus_val16 *, opus_val32 *, int, int, int);
 #   define OVERRIDE_PITCH_XCORR (1)
-#   define celt_pitch_xcorr(_x, _y, xcorr, len, max_pitch, arch) \
-  ((*CELT_PITCH_XCORR_IMPL[(arch)&OPUS_ARCHMASK])(_x, _y, \
-        xcorr, len, max_pitch))
+#   define celt_pitch_xcorr (*CELT_PITCH_XCORR_IMPL[(arch)&OPUS_ARCHMASK])
 
 #  elif defined(OPUS_ARM_PRESUME_EDSP) || \
     defined(OPUS_ARM_PRESUME_MEDIA) || \
     defined(OPUS_ARM_PRESUME_NEON)
 #   define OVERRIDE_PITCH_XCORR (1)
-#   define celt_pitch_xcorr(_x, _y, xcorr, len, max_pitch, arch) \
-  ((void)(arch),PRESUME_NEON(celt_pitch_xcorr)(_x, _y, xcorr, len, max_pitch))
+#   define celt_pitch_xcorr (PRESUME_NEON(celt_pitch_xcorr))
 
 #  endif
 
@@ -99,7 +96,7 @@
 /* Float case */
 #if defined(OPUS_ARM_MAY_HAVE_NEON_INTR)
 void celt_pitch_xcorr_float_neon(const opus_val16 *_x, const opus_val16 *_y,
-                                 opus_val32 *xcorr, int len, int max_pitch);
+                                 opus_val32 *xcorr, int len, int max_pitch, int arch);
 #endif
 
 #  if defined(OPUS_HAVE_RTCD) && \
@@ -106,18 +103,15 @@
     (defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR))
 extern void
 (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *,
-      const opus_val16 *, opus_val32 *, int, int);
+      const opus_val16 *, opus_val32 *, int, int, int);
 
 #  define OVERRIDE_PITCH_XCORR (1)
-#  define celt_pitch_xcorr(_x, _y, xcorr, len, max_pitch, arch) \
-  ((*CELT_PITCH_XCORR_IMPL[(arch)&OPUS_ARCHMASK])(_x, _y, \
-        xcorr, len, max_pitch))
+#  define celt_pitch_xcorr (*CELT_PITCH_XCORR_IMPL[(arch)&OPUS_ARCHMASK])
 
 #  elif defined(OPUS_ARM_PRESUME_NEON_INTR)
 
 #   define OVERRIDE_PITCH_XCORR (1)
-#   define celt_pitch_xcorr(_x, _y, xcorr, len, max_pitch, arch) \
-   ((void)(arch),celt_pitch_xcorr_float_neon(_x, _y, xcorr, len, max_pitch))
+#   define celt_pitch_xcorr celt_pitch_xcorr_float_neon
 
 #  endif
 
--- a/celt/pitch.c
+++ b/celt/pitch.c
@@ -220,13 +220,8 @@
 #else
 void
 #endif
-#if defined(OVERRIDE_PITCH_XCORR)
 celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y,
-      opus_val32 *xcorr, int len, int max_pitch)
-#else
-celt_pitch_xcorr(const opus_val16 *_x, const opus_val16 *_y,
       opus_val32 *xcorr, int len, int max_pitch, int arch)
-#endif
 {
 
 #if 0 /* This is a simple version of the pitch correlation that should work
@@ -265,11 +260,7 @@
    for (i=0;i<max_pitch-3;i+=4)
    {
       opus_val32 sum[4]={0,0,0,0};
-#if defined(OVERRIDE_PITCH_XCORR)
-      xcorr_kernel_c(_x, _y+i, sum, len);
-#else
       xcorr_kernel(_x, _y+i, sum, len, arch);
-#endif
       xcorr[i]=sum[0];
       xcorr[i+1]=sum[1];
       xcorr[i+2]=sum[2];
@@ -285,11 +276,7 @@
    for (;i<max_pitch;i++)
    {
       opus_val32 sum;
-#if defined(OVERRIDE_PITCH_XCORR)
-      sum = celt_inner_prod_c(_x, _y+i, len);
-#else
       sum = celt_inner_prod(_x, _y+i, len, arch);
-#endif
       xcorr[i] = sum;
 #ifdef FIXED_POINT
       maxcorr = MAX32(maxcorr, sum);
--- a/celt/pitch.h
+++ b/celt/pitch.h
@@ -184,17 +184,10 @@
 void
 #endif
 celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y,
-      opus_val32 *xcorr, int len, int max_pitch);
-
-#if !defined(OVERRIDE_PITCH_XCORR)
-#ifdef FIXED_POINT
-opus_val32
-#else
-void
-#endif
-celt_pitch_xcorr(const opus_val16 *_x, const opus_val16 *_y,
       opus_val32 *xcorr, int len, int max_pitch, int arch);
 
+#ifndef OVERRIDE_PITCH_XCORR
+# define celt_pitch_xcorr celt_pitch_xcorr_c
 #endif
 
 #endif