summaryrefslogtreecommitdiff
path: root/drivers/opus/celt/x86/pitch_sse.h
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/opus/celt/x86/pitch_sse.h')
-rw-r--r--drivers/opus/celt/x86/pitch_sse.h257
1 files changed, 145 insertions, 112 deletions
diff --git a/drivers/opus/celt/x86/pitch_sse.h b/drivers/opus/celt/x86/pitch_sse.h
index 1542b87232..2b4b54f61f 100644
--- a/drivers/opus/celt/x86/pitch_sse.h
+++ b/drivers/opus/celt/x86/pitch_sse.h
@@ -1,4 +1,5 @@
-/* Copyright (c) 2013 Jean-Marc Valin and John Ridges */
+/* Copyright (c) 2013 Jean-Marc Valin and John Ridges
+ Copyright (c) 2014, Cisco Systems, INC MingXiang WeiZhou MinPeng YanWang*/
/**
@file pitch_sse.h
@brief Pitch analysis
@@ -31,126 +32,158 @@
#ifndef PITCH_SSE_H
#define PITCH_SSE_H
+#include "opus/opus_config.h"
-#include <xmmintrin.h>
-#include "opus/celt/arch.h"
+#if defined(OPUS_X86_MAY_HAVE_SSE4_1) && defined(FIXED_POINT)
+void xcorr_kernel_sse4_1(
+ const opus_int16 *x,
+ const opus_int16 *y,
+ opus_val32 sum[4],
+ int len);
+#endif
+
+#if defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT)
+void xcorr_kernel_sse(
+ const opus_val16 *x,
+ const opus_val16 *y,
+ opus_val32 sum[4],
+ int len);
+#endif
+#if defined(OPUS_X86_PRESUME_SSE4_1) && defined(FIXED_POINT)
#define OVERRIDE_XCORR_KERNEL
-static OPUS_INLINE void xcorr_kernel(const opus_val16 *x, const opus_val16 *y, opus_val32 sum[4], int len)
-{
- int j;
- __m128 xsum1, xsum2;
- xsum1 = _mm_loadu_ps(sum);
- xsum2 = _mm_setzero_ps();
-
- for (j = 0; j < len-3; j += 4)
- {
- __m128 x0 = _mm_loadu_ps(x+j);
- __m128 yj = _mm_loadu_ps(y+j);
- __m128 y3 = _mm_loadu_ps(y+j+3);
-
- xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0x00),yj));
- xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0x55),
- _mm_shuffle_ps(yj,y3,0x49)));
- xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0xaa),
- _mm_shuffle_ps(yj,y3,0x9e)));
- xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0xff),y3));
- }
- if (j < len)
- {
- xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j)));
- if (++j < len)
- {
- xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j)));
- if (++j < len)
- {
- xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j)));
- }
- }
- }
- _mm_storeu_ps(sum,_mm_add_ps(xsum1,xsum2));
-}
+#define xcorr_kernel(x, y, sum, len, arch) \
+ ((void)arch, xcorr_kernel_sse4_1(x, y, sum, len))
-#define OVERRIDE_DUAL_INNER_PROD
-static OPUS_INLINE void dual_inner_prod(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02,
- int N, opus_val32 *xy1, opus_val32 *xy2)
-{
- int i;
- __m128 xsum1, xsum2;
- xsum1 = _mm_setzero_ps();
- xsum2 = _mm_setzero_ps();
- for (i=0;i<N-3;i+=4)
- {
- __m128 xi = _mm_loadu_ps(x+i);
- __m128 y1i = _mm_loadu_ps(y01+i);
- __m128 y2i = _mm_loadu_ps(y02+i);
- xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(xi, y1i));
- xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(xi, y2i));
- }
- /* Horizontal sum */
- xsum1 = _mm_add_ps(xsum1, _mm_movehl_ps(xsum1, xsum1));
- xsum1 = _mm_add_ss(xsum1, _mm_shuffle_ps(xsum1, xsum1, 0x55));
- _mm_store_ss(xy1, xsum1);
- xsum2 = _mm_add_ps(xsum2, _mm_movehl_ps(xsum2, xsum2));
- xsum2 = _mm_add_ss(xsum2, _mm_shuffle_ps(xsum2, xsum2, 0x55));
- _mm_store_ss(xy2, xsum2);
- for (;i<N;i++)
- {
- *xy1 = MAC16_16(*xy1, x[i], y01[i]);
- *xy2 = MAC16_16(*xy2, x[i], y02[i]);
- }
-}
+#elif defined(OPUS_X86_PRESUME_SSE) && !defined(FIXED_POINT)
+#define OVERRIDE_XCORR_KERNEL
+#define xcorr_kernel(x, y, sum, len, arch) \
+ ((void)arch, xcorr_kernel_sse(x, y, sum, len))
+
+#elif (defined(OPUS_X86_MAY_HAVE_SSE4_1) && defined(FIXED_POINT)) || (defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT))
+
+extern void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])(
+ const opus_val16 *x,
+ const opus_val16 *y,
+ opus_val32 sum[4],
+ int len);
+
+#define OVERRIDE_XCORR_KERNEL
+#define xcorr_kernel(x, y, sum, len, arch) \
+ ((*XCORR_KERNEL_IMPL[(arch) & OPUS_ARCHMASK])(x, y, sum, len))
+
+#endif
+
+#if defined(OPUS_X86_MAY_HAVE_SSE4_1) && defined(FIXED_POINT)
+opus_val32 celt_inner_prod_sse4_1(
+ const opus_int16 *x,
+ const opus_int16 *y,
+ int N);
+#endif
+
+#if defined(OPUS_X86_MAY_HAVE_SSE2) && defined(FIXED_POINT)
+opus_val32 celt_inner_prod_sse2(
+ const opus_int16 *x,
+ const opus_int16 *y,
+ int N);
+#endif
+
+#if defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(FIXED_POINT)
+opus_val32 celt_inner_prod_sse(
+ const opus_val16 *x,
+ const opus_val16 *y,
+ int N);
+#endif
+
+
+#if defined(OPUS_X86_PRESUME_SSE4_1) && defined(FIXED_POINT)
+#define OVERRIDE_CELT_INNER_PROD
+#define celt_inner_prod(x, y, N, arch) \
+ ((void)arch, celt_inner_prod_sse4_1(x, y, N))
+
+#elif defined(OPUS_X86_PRESUME_SSE2) && defined(FIXED_POINT) && !defined(OPUS_X86_MAY_HAVE_SSE4_1)
+#define OVERRIDE_CELT_INNER_PROD
+#define celt_inner_prod(x, y, N, arch) \
+ ((void)arch, celt_inner_prod_sse2(x, y, N))
+
+#elif defined(OPUS_X86_PRESUME_SSE) && !defined(FIXED_POINT)
+#define OVERRIDE_CELT_INNER_PROD
+#define celt_inner_prod(x, y, N, arch) \
+ ((void)arch, celt_inner_prod_sse(x, y, N))
+
+
+#elif ((defined(OPUS_X86_MAY_HAVE_SSE4_1) || defined(OPUS_X86_MAY_HAVE_SSE2)) && defined(FIXED_POINT)) || \
+ (defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT))
+
+extern opus_val32 (*const CELT_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])(
+ const opus_val16 *x,
+ const opus_val16 *y,
+ int N);
+
+#define OVERRIDE_CELT_INNER_PROD
+#define celt_inner_prod(x, y, N, arch) \
+ ((*CELT_INNER_PROD_IMPL[(arch) & OPUS_ARCHMASK])(x, y, N))
-#define OVERRIDE_COMB_FILTER_CONST
-static OPUS_INLINE void comb_filter_const(opus_val32 *y, opus_val32 *x, int T, int N,
- opus_val16 g10, opus_val16 g11, opus_val16 g12)
-{
- int i;
- __m128 x0v;
- __m128 g10v, g11v, g12v;
- g10v = _mm_load1_ps(&g10);
- g11v = _mm_load1_ps(&g11);
- g12v = _mm_load1_ps(&g12);
- x0v = _mm_loadu_ps(&x[-T-2]);
- for (i=0;i<N-3;i+=4)
- {
- __m128 yi, yi2, x1v, x2v, x3v, x4v;
- const opus_val32 *xp = &x[i-T-2];
- yi = _mm_loadu_ps(x+i);
- x4v = _mm_loadu_ps(xp+4);
-#if 0
- /* Slower version with all loads */
- x1v = _mm_loadu_ps(xp+1);
- x2v = _mm_loadu_ps(xp+2);
- x3v = _mm_loadu_ps(xp+3);
-#else
- x2v = _mm_shuffle_ps(x0v, x4v, 0x4e);
- x1v = _mm_shuffle_ps(x0v, x2v, 0x99);
- x3v = _mm_shuffle_ps(x2v, x4v, 0x99);
#endif
- yi = _mm_add_ps(yi, _mm_mul_ps(g10v,x2v));
-#if 0 /* Set to 1 to make it bit-exact with the non-SSE version */
- yi = _mm_add_ps(yi, _mm_mul_ps(g11v,_mm_add_ps(x3v,x1v)));
- yi = _mm_add_ps(yi, _mm_mul_ps(g12v,_mm_add_ps(x4v,x0v)));
+#if defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT)
+
+#define OVERRIDE_DUAL_INNER_PROD
+#define OVERRIDE_COMB_FILTER_CONST
+
+#undef dual_inner_prod
+#undef comb_filter_const
+
+void dual_inner_prod_sse(const opus_val16 *x,
+ const opus_val16 *y01,
+ const opus_val16 *y02,
+ int N,
+ opus_val32 *xy1,
+ opus_val32 *xy2);
+
+void comb_filter_const_sse(opus_val32 *y,
+ opus_val32 *x,
+ int T,
+ int N,
+ opus_val16 g10,
+ opus_val16 g11,
+ opus_val16 g12);
+
+
+#if defined(OPUS_X86_PRESUME_SSE)
+# define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch) \
+ ((void)(arch),dual_inner_prod_sse(x, y01, y02, N, xy1, xy2))
+
+# define comb_filter_const(y, x, T, N, g10, g11, g12, arch) \
+ ((void)(arch),comb_filter_const_sse(y, x, T, N, g10, g11, g12))
#else
- /* Use partial sums */
- yi2 = _mm_add_ps(_mm_mul_ps(g11v,_mm_add_ps(x3v,x1v)),
- _mm_mul_ps(g12v,_mm_add_ps(x4v,x0v)));
- yi = _mm_add_ps(yi, yi2);
+
+extern void (*const DUAL_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])(
+ const opus_val16 *x,
+ const opus_val16 *y01,
+ const opus_val16 *y02,
+ int N,
+ opus_val32 *xy1,
+ opus_val32 *xy2);
+
+#define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch) \
+ ((*DUAL_INNER_PROD_IMPL[(arch) & OPUS_ARCHMASK])(x, y01, y02, N, xy1, xy2))
+
+extern void (*const COMB_FILTER_CONST_IMPL[OPUS_ARCHMASK + 1])(
+ opus_val32 *y,
+ opus_val32 *x,
+ int T,
+ int N,
+ opus_val16 g10,
+ opus_val16 g11,
+ opus_val16 g12);
+
+#define comb_filter_const(y, x, T, N, g10, g11, g12, arch) \
+ ((*COMB_FILTER_CONST_IMPL[(arch) & OPUS_ARCHMASK])(y, x, T, N, g10, g11, g12))
+
+#define NON_STATIC_COMB_FILTER_CONST_C
+
#endif
- x0v=x4v;
- _mm_storeu_ps(y+i, yi);
- }
-#ifdef CUSTOM_MODES
- for (;i<N;i++)
- {
- y[i] = x[i]
- + MULT16_32_Q15(g10,x[i-T])
- + MULT16_32_Q15(g11,ADD32(x[i-T+1],x[i-T-1]))
- + MULT16_32_Q15(g12,ADD32(x[i-T+2],x[i-T-2]));
- }
#endif
-}
#endif
generated by cgit v1.2.3 (git 2.25.1) at 2025-03-22 08:15:27 +0000