From 85fb8a1d17a563d1e7989e8268a856a5c90ecfe7 Mon Sep 17 00:00:00 2001
From: Linfeng Zhang <linfengz@google.com>
Date: Wed, 7 Sep 2016 15:44:23 -0700
Subject: [PATCH] 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>
---
celt/arm/arm_celt_map.c | 4 ++--
celt/arm/celt_neon_intr.c | 3 ++-
celt/arm/celt_pitch_xcorr_arm.s | 6 +++++-
celt/arm/pitch_arm.h | 24 +++++++++---------------
celt/pitch.c | 13 -------------
celt/pitch.h | 11 ++---------
6 files changed, 20 insertions(+), 41 deletions(-)
diff --git a/celt/arm/arm_celt_map.c b/celt/arm/arm_celt_map.c
index 4d4d069a..cc6eb45c 100644
--- 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 @@ opus_val32 (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *,
# 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 */
diff --git a/celt/arm/celt_neon_intr.c b/celt/arm/celt_neon_intr.c
index 47bbe3dc..f8f1d439 100644
--- a/celt/arm/celt_neon_intr.c
+++ b/celt/arm/celt_neon_intr.c
@@ -290,8 +290,9 @@ static void xcorr_kernel_neon_float_process1(const float32_t *x,
}
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);
diff --git a/celt/arm/celt_pitch_xcorr_arm.s b/celt/arm/celt_pitch_xcorr_arm.s
index f96e0a88..6e873afc 100644
--- a/celt/arm/celt_pitch_xcorr_arm.s
+++ b/celt/arm/celt_pitch_xcorr_arm.s
@@ -153,7 +153,7 @@ xcorr_kernel_neon_process1
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 @@ celt_pitch_xcorr_neon PROC
; 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 @@ celt_pitch_xcorr_edsp PROC
; 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]
diff --git a/celt/arm/pitch_arm.h b/celt/arm/pitch_arm.h
index 14331169..d8b022ef 100644
--- 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 @@ opus_val32 celt_pitch_xcorr_neon(const opus_val16 *_x, const opus_val16 *_y,
# 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 @@ opus_val32 celt_pitch_xcorr_edsp(const opus_val16 *_x, const opus_val16 *_y,
(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
@@ -99,25 +96,22 @@ extern void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])(
/* 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) && \
(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
diff --git a/celt/pitch.c b/celt/pitch.c
index 1ca68bf6..5b061d86 100644
--- a/celt/pitch.c
+++ b/celt/pitch.c
@@ -220,13 +220,8 @@ opus_val32
#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 @@ celt_pitch_xcorr(const opus_val16 *_x, const opus_val16 *_y,
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 @@ celt_pitch_xcorr(const opus_val16 *_x, const opus_val16 *_y,
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);
diff --git a/celt/pitch.h b/celt/pitch.h
index d3503532..d7978440 100644
--- a/celt/pitch.h
+++ b/celt/pitch.h
@@ -184,17 +184,10 @@ opus_val32
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
--
GitLab