generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
vpx/vpx_dsp/x86/variance_sse2.c
Johann 2967bf355e Merge changes from libvpx/master by cherry-pick 
This commit bring all up-to-date changes from master that are
applicable to nextgenv2. Due to the remove VP10 code in master,
we had to cherry pick the following commits to get those changes:

Add default flags for arm64/armv8 builds

Allows building simple targets with sane default flags.

For example, using the Android arm64 toolchain from the NDK:
https://developer.android.com/ndk/guides/standalone_toolchain.html
./build/tools/make-standalone-toolchain.sh --arch=arm64 \
  --platform=android-24 --install-dir=/tmp/arm64
CROSS=/tmp/arm64/bin/aarch64-linux-android- \
  ~/libvpx/configure --target=arm64-linux-gcc --disable-multithread

BUG=webm:1143

vpx_lpf_horizontal_4_sse2: Remove dead load.

Change-Id: I51026c52baa1f0881fcd5b68e1fdf08a2dc0916e

Fail early when android target does not include --sdk-path

Change-Id: I07e7e63476a2e32e3aae123abdee8b7bbbdc6a8c

configure: clean up var style and set_all usage

Use quotes whenever possible and {} always for variables.

Replace multiple set_all calls with *able_feature().

Conflicts:
	build/make/configure.sh

vp9-svc: Remove some unneeded code/comment.

datarate_test,DatarateTestLarge: normalize bits type

quiets a msvc warning:
conversion from 'const int64_t' to 'size_t', possible loss of data

mips added p6600 cpu support

Removed -funroll-loops

psnr.c: use int64_t for sum of differences

Since the values can be negative.

*.asm: normalize label format

add a trailing ':', though it's optional with the tools we support, it's
more common to use it to mark a label. this also quiets the
orphan-labels warning with nasm/yasm.

BUG=b/29583530

Prevent negative variance

Due to rounding, hbd variance may become negative. This commit put in
check and clamp of negative values to 0.

configure: remove old visual studio support (<2010)

BUG=b/29583530

Conflicts:
	configure

configure: restore vs_version variable

inadvertently lost in the final patchset of:
078dff7 configure: remove old visual studio support (<2010)

this prevents an empty CONFIG_VS_VERSION and avoids make failure

Require x86inc.asm

Force enable x86inc.asm when building for x86. Previously there were
compatibility issues so a flag was added to simplify disabling this
code.

The known issues have been resolved and x86inc.asm is the preferred
abstraction layer (over x86_abi_support.asm).

BUG=b:29583530

convolve_test: fix byte offsets in hbd build

CONVERT_TO_BYTEPTR(x) was corrected in:
003a9d2 Port metric computation changes from nextgenv2
to use the more common (x) within the expansion. offsets should occur
after converting the pointer to the desired type.

+ factorized some common expressions

Conflicts:
	test/convolve_test.cc

vpx_dsp: remove x86inc.asm distinction

BUG=b:29583530

Conflicts:
	vpx_dsp/vpx_dsp.mk
	vpx_dsp/vpx_dsp_rtcd_defs.pl
	vpx_dsp/x86/highbd_variance_sse2.c
	vpx_dsp/x86/variance_sse2.c

test: remove x86inc.asm distinction

BUG=b:29583530

Conflicts:
	test/vp9_subtract_test.cc

configure: remove x86inc.asm distinction

BUG=b:29583530

Change-Id: I59a1192142e89a6a36b906f65a491a734e603617

Update vpx subpixel 1d filter ssse3 asm

Speed test shows the new vertical filters have degradation on Celeron
Chromebook. Added "X86_SUBPIX_VFILTER_PREFER_SLOW_CELERON" to control
the vertical filters activated code. Now just simply active the code
without degradation on Celeron. Later there should be 2 set of vertical
filters ssse3 functions, and let jump table to choose based on CPU type.

improve vpx_filter_block1d* based on replace paddsw+psrlw to pmulhrsw

Make set_reference control API work in VP9

Moved the API patch from NextGenv2. An example was included.
To try it, for example, run the following command:
$ examples/vpx_cx_set_ref vp9 352 288 in.yuv out.ivf 4 30

Conflicts:
	examples.mk
	examples/vpx_cx_set_ref.c
	test/cx_set_ref.sh
	vp9/decoder/vp9_decoder.c

deblock filter : moved from vp8 code branch

The deblocking filters used in vp8 have been moved to vpx_dsp for
use by both vp8 and vp9.

vpx_thread.[hc]: update webp source reference

+ drop the blob hash, the updated reference will be updated in the
commit message

BUG=b/29583578

vpx_thread: use native windows cond var if available

BUG=b/29583578

original webp change:

commit 110ad5835ecd66995d0e7f66dca1b90dea595f5a
Author: James Zern <jzern@google.com>
Date:   Mon Nov 23 19:49:58 2015 -0800

    thread: use native windows cond var if available

    Vista / Server 2008 and up. no speed difference observed.

100644 blob 4fc372b7bc6980a9ed3618c8cce5b67ed7b0f412 src/utils/thread.c
100644 blob 840831185502d42a3246e4b7ff870121c8064791 src/utils/thread.h

vpx_thread: use InitializeCriticalSectionEx if available

BUG=b/29583578

original webp change:

commit 63fadc9ffacc77d4617526a50c696d21d558a70b
Author: James Zern <jzern@google.com>
Date:   Mon Nov 23 20:38:46 2015 -0800

    thread: use InitializeCriticalSectionEx if available

    Windows Vista / Server 2008 and up

100644 blob f84207d89b3a6bb98bfe8f3fa55cad72dfd061ff src/utils/thread.c
100644 blob 840831185502d42a3246e4b7ff870121c8064791 src/utils/thread.h

vpx_thread: use WaitForSingleObjectEx if available

BUG=b/29583578

original webp change:

commit 0fd0e12bfe83f16ce4f1c038b251ccbc13c62ac2
Author: James Zern <jzern@google.com>
Date:   Mon Nov 23 20:40:26 2015 -0800

    thread: use WaitForSingleObjectEx if available

    Windows XP and up

100644 blob d58f74e5523dbc985fc531cf5f0833f1e9157cf0 src/utils/thread.c
100644 blob 840831185502d42a3246e4b7ff870121c8064791 src/utils/thread.h

vpx_thread: use CreateThread for windows phone

BUG=b/29583578

original webp change:

commit d2afe974f9d751de144ef09d31255aea13b442c0
Author: James Zern <jzern@google.com>
Date:   Mon Nov 23 20:41:26 2015 -0800

    thread: use CreateThread for windows phone

    _beginthreadex is unavailable for winrt/uwp

    Change-Id: Ie7412a568278ac67f0047f1764e2521193d74d4d

100644 blob 93f7622797f05f6acc1126e8296c481d276e4047 src/utils/thread.c
100644 blob 840831185502d42a3246e4b7ff870121c8064791 src/utils/thread.h

vp9_postproc.c missing extern.

BUG=webm:1256

deblock: missing const on extern const.

postproc - move filling of noise buffer to vpx_dsp.

Fix encoder crashes for odd size input

clean-up vp9_intrapred_test

remove tuple and overkill VP9IntraPredBase class.

postproc: noise style fixes.

gtest-all.cc: quiet an unused variable warning

under windows / mingw builds

vp9_intrapred_test: follow-up cleanup

address few comments from ce050afaf3e288895c3bee4160336e2d2133b6ea

Change-Id: I3eece7efa9335f4210303993ef6c1857ad5c29c8
2016-07-18 10:31:10 -07:00

690 lines
26 KiB
C
Raw Permalink Blame History

/*
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include <emmintrin.h> // SSE2
#include "./vpx_config.h"
#include "./vpx_dsp_rtcd.h"
#include "vpx_ports/mem.h"
typedef void (*getNxMvar_fn_t) (const unsigned char *src, int src_stride,
const unsigned char *ref, int ref_stride,
unsigned int *sse, int *sum);
unsigned int vpx_get_mb_ss_sse2(const int16_t *src) {
__m128i vsum = _mm_setzero_si128();
int i;
for (i = 0; i < 32; ++i) {
const __m128i v = _mm_loadu_si128((const __m128i *)src);
vsum = _mm_add_epi32(vsum, _mm_madd_epi16(v, v));
src += 8;
}
vsum = _mm_add_epi32(vsum, _mm_srli_si128(vsum, 8));
vsum = _mm_add_epi32(vsum, _mm_srli_si128(vsum, 4));
return _mm_cvtsi128_si32(vsum);
}
#define READ64(p, stride, i) \
_mm_unpacklo_epi8(_mm_cvtsi32_si128(*(const uint32_t *)(p + i * stride)), \
_mm_cvtsi32_si128(*(const uint32_t *)(p + (i + 1) * stride)))
static void get4x4var_sse2(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
unsigned int *sse, int *sum) {
const __m128i zero = _mm_setzero_si128();
const __m128i src0 = _mm_unpacklo_epi8(READ64(src, src_stride, 0), zero);
const __m128i src1 = _mm_unpacklo_epi8(READ64(src, src_stride, 2), zero);
const __m128i ref0 = _mm_unpacklo_epi8(READ64(ref, ref_stride, 0), zero);
const __m128i ref1 = _mm_unpacklo_epi8(READ64(ref, ref_stride, 2), zero);
const __m128i diff0 = _mm_sub_epi16(src0, ref0);
const __m128i diff1 = _mm_sub_epi16(src1, ref1);
// sum
__m128i vsum = _mm_add_epi16(diff0, diff1);
vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 8));
vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 4));
vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 2));
*sum = (int16_t)_mm_extract_epi16(vsum, 0);
// sse
vsum = _mm_add_epi32(_mm_madd_epi16(diff0, diff0),
_mm_madd_epi16(diff1, diff1));
vsum = _mm_add_epi32(vsum, _mm_srli_si128(vsum, 8));
vsum = _mm_add_epi32(vsum, _mm_srli_si128(vsum, 4));
*sse = _mm_cvtsi128_si32(vsum);
}
void vpx_get8x8var_sse2(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
unsigned int *sse, int *sum) {
const __m128i zero = _mm_setzero_si128();
__m128i vsum = _mm_setzero_si128();
__m128i vsse = _mm_setzero_si128();
int i;
for (i = 0; i < 8; i += 2) {
const __m128i src0 = _mm_unpacklo_epi8(_mm_loadl_epi64(
(const __m128i *)(src + i * src_stride)), zero);
const __m128i ref0 = _mm_unpacklo_epi8(_mm_loadl_epi64(
(const __m128i *)(ref + i * ref_stride)), zero);
const __m128i diff0 = _mm_sub_epi16(src0, ref0);
const __m128i src1 = _mm_unpacklo_epi8(_mm_loadl_epi64(
(const __m128i *)(src + (i + 1) * src_stride)), zero);
const __m128i ref1 = _mm_unpacklo_epi8(_mm_loadl_epi64(
(const __m128i *)(ref + (i + 1) * ref_stride)), zero);
const __m128i diff1 = _mm_sub_epi16(src1, ref1);
vsum = _mm_add_epi16(vsum, diff0);
vsum = _mm_add_epi16(vsum, diff1);
vsse = _mm_add_epi32(vsse, _mm_madd_epi16(diff0, diff0));
vsse = _mm_add_epi32(vsse, _mm_madd_epi16(diff1, diff1));
}
// sum
vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 8));
vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 4));
vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 2));
*sum = (int16_t)_mm_extract_epi16(vsum, 0);
// sse
vsse = _mm_add_epi32(vsse, _mm_srli_si128(vsse, 8));
vsse = _mm_add_epi32(vsse, _mm_srli_si128(vsse, 4));
*sse = _mm_cvtsi128_si32(vsse);
}
void vpx_get16x16var_sse2(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
unsigned int *sse, int *sum) {
const __m128i zero = _mm_setzero_si128();
__m128i vsum = _mm_setzero_si128();
__m128i vsse = _mm_setzero_si128();
int i;
for (i = 0; i < 16; ++i) {
const __m128i s = _mm_loadu_si128((const __m128i *)src);
const __m128i r = _mm_loadu_si128((const __m128i *)ref);
const __m128i src0 = _mm_unpacklo_epi8(s, zero);
const __m128i ref0 = _mm_unpacklo_epi8(r, zero);
const __m128i diff0 = _mm_sub_epi16(src0, ref0);
const __m128i src1 = _mm_unpackhi_epi8(s, zero);
const __m128i ref1 = _mm_unpackhi_epi8(r, zero);
const __m128i diff1 = _mm_sub_epi16(src1, ref1);
vsum = _mm_add_epi16(vsum, diff0);
vsum = _mm_add_epi16(vsum, diff1);
vsse = _mm_add_epi32(vsse, _mm_madd_epi16(diff0, diff0));
vsse = _mm_add_epi32(vsse, _mm_madd_epi16(diff1, diff1));
src += src_stride;
ref += ref_stride;
}
// sum
vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 8));
vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 4));
*sum = (int16_t)_mm_extract_epi16(vsum, 0) +
(int16_t)_mm_extract_epi16(vsum, 1);
// sse
vsse = _mm_add_epi32(vsse, _mm_srli_si128(vsse, 8));
vsse = _mm_add_epi32(vsse, _mm_srli_si128(vsse, 4));
*sse = _mm_cvtsi128_si32(vsse);
}
static void variance_sse2(const unsigned char *src, int src_stride,
const unsigned char *ref, int ref_stride,
int w, int h, unsigned int *sse, int *sum,
getNxMvar_fn_t var_fn, int block_size) {
int i, j;
*sse = 0;
*sum = 0;
for (i = 0; i < h; i += block_size) {
for (j = 0; j < w; j += block_size) {
unsigned int sse0;
int sum0;
var_fn(src + src_stride * i + j, src_stride,
ref + ref_stride * i + j, ref_stride, &sse0, &sum0);
*sse += sse0;
*sum += sum0;
}
}
}
unsigned int vpx_variance4x4_sse2(const unsigned char *src, int src_stride,
const unsigned char *ref, int ref_stride,
unsigned int *sse) {
int sum;
get4x4var_sse2(src, src_stride, ref, ref_stride, sse, &sum);
return *sse - ((sum * sum) >> 4);
}
unsigned int vpx_variance8x4_sse2(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
unsigned int *sse) {
int sum;
variance_sse2(src, src_stride, ref, ref_stride, 8, 4,
sse, &sum, get4x4var_sse2, 4);
return *sse - ((sum * sum) >> 5);
}
unsigned int vpx_variance4x8_sse2(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
unsigned int *sse) {
int sum;
variance_sse2(src, src_stride, ref, ref_stride, 4, 8,
sse, &sum, get4x4var_sse2, 4);
return *sse - ((sum * sum) >> 5);
}
unsigned int vpx_variance8x8_sse2(const unsigned char *src, int src_stride,
const unsigned char *ref, int ref_stride,
unsigned int *sse) {
int sum;
vpx_get8x8var_sse2(src, src_stride, ref, ref_stride, sse, &sum);
return *sse - ((sum * sum) >> 6);
}
unsigned int vpx_variance16x8_sse2(const unsigned char *src, int src_stride,
const unsigned char *ref, int ref_stride,
unsigned int *sse) {
int sum;
variance_sse2(src, src_stride, ref, ref_stride, 16, 8,
sse, &sum, vpx_get8x8var_sse2, 8);
return *sse - ((sum * sum) >> 7);
}
unsigned int vpx_variance8x16_sse2(const unsigned char *src, int src_stride,
const unsigned char *ref, int ref_stride,
unsigned int *sse) {
int sum;
variance_sse2(src, src_stride, ref, ref_stride, 8, 16,
sse, &sum, vpx_get8x8var_sse2, 8);
return *sse - ((sum * sum) >> 7);
}
unsigned int vpx_variance16x16_sse2(const unsigned char *src, int src_stride,
const unsigned char *ref, int ref_stride,
unsigned int *sse) {
int sum;
vpx_get16x16var_sse2(src, src_stride, ref, ref_stride, sse, &sum);
return *sse - (((uint32_t)((int64_t)sum * sum)) >> 8);
}
unsigned int vpx_variance32x32_sse2(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
unsigned int *sse) {
int sum;
variance_sse2(src, src_stride, ref, ref_stride, 32, 32,
sse, &sum, vpx_get16x16var_sse2, 16);
return *sse - (((int64_t)sum * sum) >> 10);
}
unsigned int vpx_variance32x16_sse2(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
unsigned int *sse) {
int sum;
variance_sse2(src, src_stride, ref, ref_stride, 32, 16,
sse, &sum, vpx_get16x16var_sse2, 16);
return *sse - (((int64_t)sum * sum) >> 9);
}
unsigned int vpx_variance16x32_sse2(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
unsigned int *sse) {
int sum;
variance_sse2(src, src_stride, ref, ref_stride, 16, 32,
sse, &sum, vpx_get16x16var_sse2, 16);
return *sse - (((int64_t)sum * sum) >> 9);
}
unsigned int vpx_variance64x64_sse2(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
unsigned int *sse) {
int sum;
variance_sse2(src, src_stride, ref, ref_stride, 64, 64,
sse, &sum, vpx_get16x16var_sse2, 16);
return *sse - (((int64_t)sum * sum) >> 12);
}
unsigned int vpx_variance64x32_sse2(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
unsigned int *sse) {
int sum;
variance_sse2(src, src_stride, ref, ref_stride, 64, 32,
sse, &sum, vpx_get16x16var_sse2, 16);
return *sse - (((int64_t)sum * sum) >> 11);
}
unsigned int vpx_variance32x64_sse2(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
unsigned int *sse) {
int sum;
variance_sse2(src, src_stride, ref, ref_stride, 32, 64,
sse, &sum, vpx_get16x16var_sse2, 16);
return *sse - (((int64_t)sum * sum) >> 11);
}
unsigned int vpx_mse8x8_sse2(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
unsigned int *sse) {
vpx_variance8x8_sse2(src, src_stride, ref, ref_stride, sse);
return *sse;
}
unsigned int vpx_mse8x16_sse2(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
unsigned int *sse) {
vpx_variance8x16_sse2(src, src_stride, ref, ref_stride, sse);
return *sse;
}
unsigned int vpx_mse16x8_sse2(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
unsigned int *sse) {
vpx_variance16x8_sse2(src, src_stride, ref, ref_stride, sse);
return *sse;
}
unsigned int vpx_mse16x16_sse2(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
unsigned int *sse) {
vpx_variance16x16_sse2(src, src_stride, ref, ref_stride, sse);
return *sse;
}
// The 2 unused parameters are place holders for PIC enabled build.
// These definitions are for functions defined in subpel_variance.asm
#define DECL(w, opt) \
int vpx_sub_pixel_variance##w##xh_##opt(const uint8_t *src, \
ptrdiff_t src_stride, \
int x_offset, int y_offset, \
const uint8_t *dst, \
ptrdiff_t dst_stride, \
int height, unsigned int *sse, \
void *unused0, void *unused)
#define DECLS(opt1, opt2) \
DECL(4, opt1); \
DECL(8, opt1); \
DECL(16, opt1)
DECLS(sse2, sse2);
DECLS(ssse3, ssse3);
#undef DECLS
#undef DECL
#define FN(w, h, wf, wlog2, hlog2, opt, cast_prod, cast) \
unsigned int vpx_sub_pixel_variance##w##x##h##_##opt(const uint8_t *src, \
int src_stride, \
int x_offset, \
int y_offset, \
const uint8_t *dst, \
int dst_stride, \
unsigned int *sse_ptr) { \
unsigned int sse; \
int se = vpx_sub_pixel_variance##wf##xh_##opt(src, src_stride, x_offset, \
y_offset, dst, dst_stride, \
h, &sse, NULL, NULL); \
if (w > wf) { \
unsigned int sse2; \
int se2 = vpx_sub_pixel_variance##wf##xh_##opt(src + 16, src_stride, \
x_offset, y_offset, \
dst + 16, dst_stride, \
h, &sse2, NULL, NULL); \
se += se2; \
sse += sse2; \
if (w > wf * 2) { \
se2 = vpx_sub_pixel_variance##wf##xh_##opt(src + 32, src_stride, \
x_offset, y_offset, \
dst + 32, dst_stride, \
h, &sse2, NULL, NULL); \
se += se2; \
sse += sse2; \
se2 = vpx_sub_pixel_variance##wf##xh_##opt(src + 48, src_stride, \
x_offset, y_offset, \
dst + 48, dst_stride, \
h, &sse2, NULL, NULL); \
se += se2; \
sse += sse2; \
} \
} \
*sse_ptr = sse; \
return sse - (cast_prod (cast se * se) >> (wlog2 + hlog2)); \
}
#define FNS(opt1, opt2) \
FN(64, 64, 16, 6, 6, opt1, (int64_t), (int64_t)); \
FN(64, 32, 16, 6, 5, opt1, (int64_t), (int64_t)); \
FN(32, 64, 16, 5, 6, opt1, (int64_t), (int64_t)); \
FN(32, 32, 16, 5, 5, opt1, (int64_t), (int64_t)); \
FN(32, 16, 16, 5, 4, opt1, (int64_t), (int64_t)); \
FN(16, 32, 16, 4, 5, opt1, (int64_t), (int64_t)); \
FN(16, 16, 16, 4, 4, opt1, (uint32_t), (int64_t)); \
FN(16, 8, 16, 4, 3, opt1, (int32_t), (int32_t)); \
FN(8, 16, 8, 3, 4, opt1, (int32_t), (int32_t)); \
FN(8, 8, 8, 3, 3, opt1, (int32_t), (int32_t)); \
FN(8, 4, 8, 3, 2, opt1, (int32_t), (int32_t)); \
FN(4, 8, 4, 2, 3, opt1, (int32_t), (int32_t)); \
FN(4, 4, 4, 2, 2, opt1, (int32_t), (int32_t))
FNS(sse2, sse2);
FNS(ssse3, ssse3);
#undef FNS
#undef FN
// The 2 unused parameters are place holders for PIC enabled build.
#define DECL(w, opt) \
int vpx_sub_pixel_avg_variance##w##xh_##opt(const uint8_t *src, \
ptrdiff_t src_stride, \
int x_offset, int y_offset, \
const uint8_t *dst, \
ptrdiff_t dst_stride, \
const uint8_t *sec, \
ptrdiff_t sec_stride, \
int height, unsigned int *sse, \
void *unused0, void *unused)
#define DECLS(opt1, opt2) \
DECL(4, opt1); \
DECL(8, opt1); \
DECL(16, opt1)
DECLS(sse2, sse2);
DECLS(ssse3, ssse3);
#undef DECL
#undef DECLS
#define FN(w, h, wf, wlog2, hlog2, opt, cast_prod, cast) \
unsigned int vpx_sub_pixel_avg_variance##w##x##h##_##opt(const uint8_t *src, \
int src_stride, \
int x_offset, \
int y_offset, \
const uint8_t *dst, \
int dst_stride, \
unsigned int *sseptr, \
const uint8_t *sec) { \
unsigned int sse; \
int se = vpx_sub_pixel_avg_variance##wf##xh_##opt(src, src_stride, x_offset, \
y_offset, dst, dst_stride, \
sec, w, h, &sse, NULL, \
NULL); \
if (w > wf) { \
unsigned int sse2; \
int se2 = vpx_sub_pixel_avg_variance##wf##xh_##opt(src + 16, src_stride, \
x_offset, y_offset, \
dst + 16, dst_stride, \
sec + 16, w, h, &sse2, \
NULL, NULL); \
se += se2; \
sse += sse2; \
if (w > wf * 2) { \
se2 = vpx_sub_pixel_avg_variance##wf##xh_##opt(src + 32, src_stride, \
x_offset, y_offset, \
dst + 32, dst_stride, \
sec + 32, w, h, &sse2, \
NULL, NULL); \
se += se2; \
sse += sse2; \
se2 = vpx_sub_pixel_avg_variance##wf##xh_##opt(src + 48, src_stride, \
x_offset, y_offset, \
dst + 48, dst_stride, \
sec + 48, w, h, &sse2, \
NULL, NULL); \
se += se2; \
sse += sse2; \
} \
} \
*sseptr = sse; \
return sse - (cast_prod (cast se * se) >> (wlog2 + hlog2)); \
}
#define FNS(opt1, opt2) \
FN(64, 64, 16, 6, 6, opt1, (int64_t), (int64_t)); \
FN(64, 32, 16, 6, 5, opt1, (int64_t), (int64_t)); \
FN(32, 64, 16, 5, 6, opt1, (int64_t), (int64_t)); \
FN(32, 32, 16, 5, 5, opt1, (int64_t), (int64_t)); \
FN(32, 16, 16, 5, 4, opt1, (int64_t), (int64_t)); \
FN(16, 32, 16, 4, 5, opt1, (int64_t), (int64_t)); \
FN(16, 16, 16, 4, 4, opt1, (uint32_t), (int64_t)); \
FN(16, 8, 16, 4, 3, opt1, (uint32_t), (int32_t)); \
FN(8, 16, 8, 3, 4, opt1, (uint32_t), (int32_t)); \
FN(8, 8, 8, 3, 3, opt1, (uint32_t), (int32_t)); \
FN(8, 4, 8, 3, 2, opt1, (uint32_t), (int32_t)); \
FN(4, 8, 4, 2, 3, opt1, (uint32_t), (int32_t)); \
FN(4, 4, 4, 2, 2, opt1, (uint32_t), (int32_t))
FNS(sse2, sse);
FNS(ssse3, ssse3);
#undef FNS
#undef FN
void vpx_upsampled_pred_sse2(uint8_t *comp_pred,
int width, int height,
const uint8_t *ref, int ref_stride) {
int i, j;
int stride = ref_stride << 3;
if (width >= 16) {
// read 16 points at one time
for (i = 0; i < height; i++) {
for (j = 0; j < width; j+= 16) {
__m128i s0 = _mm_loadu_si128((const __m128i *)ref);
__m128i s1 = _mm_loadu_si128((const __m128i *)(ref + 16));
__m128i s2 = _mm_loadu_si128((const __m128i *)(ref + 32));
__m128i s3 = _mm_loadu_si128((const __m128i *)(ref + 48));
__m128i s4 = _mm_loadu_si128((const __m128i *)(ref + 64));
__m128i s5 = _mm_loadu_si128((const __m128i *)(ref + 80));
__m128i s6 = _mm_loadu_si128((const __m128i *)(ref + 96));
__m128i s7 = _mm_loadu_si128((const __m128i *)(ref + 112));
__m128i t0, t1, t2, t3;
t0 = _mm_unpacklo_epi8(s0, s1);
s1 = _mm_unpackhi_epi8(s0, s1);
t1 = _mm_unpacklo_epi8(s2, s3);
s3 = _mm_unpackhi_epi8(s2, s3);
t2 = _mm_unpacklo_epi8(s4, s5);
s5 = _mm_unpackhi_epi8(s4, s5);
t3 = _mm_unpacklo_epi8(s6, s7);
s7 = _mm_unpackhi_epi8(s6, s7);
s0 = _mm_unpacklo_epi8(t0, s1);
s2 = _mm_unpacklo_epi8(t1, s3);
s4 = _mm_unpacklo_epi8(t2, s5);
s6 = _mm_unpacklo_epi8(t3, s7);
s0 = _mm_unpacklo_epi32(s0, s2);
s4 = _mm_unpacklo_epi32(s4, s6);
s0 = _mm_unpacklo_epi64(s0, s4);
_mm_storeu_si128((__m128i *)(comp_pred), s0);
comp_pred += 16;
ref += 16 * 8;
}
ref += stride - (width << 3);
}
} else if (width >= 8) {
// read 8 points at one time
for (i = 0; i < height; i++) {
for (j = 0; j < width; j+= 8) {
__m128i s0 = _mm_loadu_si128((const __m128i *)ref);
__m128i s1 = _mm_loadu_si128((const __m128i *)(ref + 16));
__m128i s2 = _mm_loadu_si128((const __m128i *)(ref + 32));
__m128i s3 = _mm_loadu_si128((const __m128i *)(ref + 48));
__m128i t0, t1;
t0 = _mm_unpacklo_epi8(s0, s1);
s1 = _mm_unpackhi_epi8(s0, s1);
t1 = _mm_unpacklo_epi8(s2, s3);
s3 = _mm_unpackhi_epi8(s2, s3);
s0 = _mm_unpacklo_epi8(t0, s1);
s2 = _mm_unpacklo_epi8(t1, s3);
s0 = _mm_unpacklo_epi32(s0, s2);
_mm_storel_epi64((__m128i *)(comp_pred), s0);
comp_pred += 8;
ref += 8 * 8;
}
ref += stride - (width << 3);
}
} else {
// read 4 points at one time
for (i = 0; i < height; i++) {
for (j = 0; j < width; j+= 4) {
__m128i s0 = _mm_loadu_si128((const __m128i *)ref);
__m128i s1 = _mm_loadu_si128((const __m128i *)(ref + 16));
__m128i t0;
t0 = _mm_unpacklo_epi8(s0, s1);
s1 = _mm_unpackhi_epi8(s0, s1);
s0 = _mm_unpacklo_epi8(t0, s1);
*(int *)comp_pred = _mm_cvtsi128_si32(s0);
comp_pred += 4;
ref += 4 * 8;
}
ref += stride - (width << 3);
}
}
}
void vpx_comp_avg_upsampled_pred_sse2(uint8_t *comp_pred, const uint8_t *pred,
int width, int height,
const uint8_t *ref, int ref_stride) {
const __m128i zero = _mm_set1_epi16(0);
const __m128i one = _mm_set1_epi16(1);
int i, j;
int stride = ref_stride << 3;
if (width >= 16) {
// read 16 points at one time
for (i = 0; i < height; i++) {
for (j = 0; j < width; j+= 16) {
__m128i s0 = _mm_loadu_si128((const __m128i *)ref);
__m128i s1 = _mm_loadu_si128((const __m128i *)(ref + 16));
__m128i s2 = _mm_loadu_si128((const __m128i *)(ref + 32));
__m128i s3 = _mm_loadu_si128((const __m128i *)(ref + 48));
__m128i s4 = _mm_loadu_si128((const __m128i *)(ref + 64));
__m128i s5 = _mm_loadu_si128((const __m128i *)(ref + 80));
__m128i s6 = _mm_loadu_si128((const __m128i *)(ref + 96));
__m128i s7 = _mm_loadu_si128((const __m128i *)(ref + 112));
__m128i p0 = _mm_loadu_si128((const __m128i *)pred);
__m128i p1;
__m128i t0, t1, t2, t3;
t0 = _mm_unpacklo_epi8(s0, s1);
s1 = _mm_unpackhi_epi8(s0, s1);
t1 = _mm_unpacklo_epi8(s2, s3);
s3 = _mm_unpackhi_epi8(s2, s3);
t2 = _mm_unpacklo_epi8(s4, s5);
s5 = _mm_unpackhi_epi8(s4, s5);
t3 = _mm_unpacklo_epi8(s6, s7);
s7 = _mm_unpackhi_epi8(s6, s7);
s0 = _mm_unpacklo_epi8(t0, s1);
s2 = _mm_unpacklo_epi8(t1, s3);
s4 = _mm_unpacklo_epi8(t2, s5);
s6 = _mm_unpacklo_epi8(t3, s7);
s0 = _mm_unpacklo_epi32(s0, s2);
s4 = _mm_unpacklo_epi32(s4, s6);
s0 = _mm_unpacklo_epi8(s0, zero);
s4 = _mm_unpacklo_epi8(s4, zero);
p1 = _mm_unpackhi_epi8(p0, zero);
p0 = _mm_unpacklo_epi8(p0, zero);
p0 = _mm_adds_epu16(s0, p0);
p1 = _mm_adds_epu16(s4, p1);
p0 = _mm_adds_epu16(p0, one);
p1 = _mm_adds_epu16(p1, one);
p0 = _mm_srli_epi16(p0, 1);
p1 = _mm_srli_epi16(p1, 1);
p0 = _mm_packus_epi16(p0, p1);
_mm_storeu_si128((__m128i *)(comp_pred), p0);
comp_pred += 16;
pred += 16;
ref += 16 * 8;
}
ref += stride - (width << 3);
}
} else if (width >= 8) {
// read 8 points at one time
for (i = 0; i < height; i++) {
for (j = 0; j < width; j+= 8) {
__m128i s0 = _mm_loadu_si128((const __m128i *)ref);
__m128i s1 = _mm_loadu_si128((const __m128i *)(ref + 16));
__m128i s2 = _mm_loadu_si128((const __m128i *)(ref + 32));
__m128i s3 = _mm_loadu_si128((const __m128i *)(ref + 48));
__m128i p0 = _mm_loadl_epi64((const __m128i *)pred);
__m128i t0, t1;
t0 = _mm_unpacklo_epi8(s0, s1);
s1 = _mm_unpackhi_epi8(s0, s1);
t1 = _mm_unpacklo_epi8(s2, s3);
s3 = _mm_unpackhi_epi8(s2, s3);
s0 = _mm_unpacklo_epi8(t0, s1);
s2 = _mm_unpacklo_epi8(t1, s3);
s0 = _mm_unpacklo_epi32(s0, s2);
s0 = _mm_unpacklo_epi8(s0, zero);
p0 = _mm_unpacklo_epi8(p0, zero);
p0 = _mm_adds_epu16(s0, p0);
p0 = _mm_adds_epu16(p0, one);
p0 = _mm_srli_epi16(p0, 1);
p0 = _mm_packus_epi16(p0, zero);
_mm_storel_epi64((__m128i *)(comp_pred), p0);
comp_pred += 8;
pred += 8;
ref += 8 * 8;
}
ref += stride - (width << 3);
}
} else {
// read 4 points at one time
for (i = 0; i < height; i++) {
for (j = 0; j < width; j+= 4) {
__m128i s0 = _mm_loadu_si128((const __m128i *)ref);
__m128i s1 = _mm_loadu_si128((const __m128i *)(ref + 16));
__m128i p0 = _mm_cvtsi32_si128(*(const uint32_t *)pred);
__m128i t0;
t0 = _mm_unpacklo_epi8(s0, s1);
s1 = _mm_unpackhi_epi8(s0, s1);
s0 = _mm_unpacklo_epi8(t0, s1);
s0 = _mm_unpacklo_epi8(s0, zero);
p0 = _mm_unpacklo_epi8(p0, zero);
p0 = _mm_adds_epu16(s0, p0);
p0 = _mm_adds_epu16(p0, one);
p0 = _mm_srli_epi16(p0, 1);
p0 = _mm_packus_epi16(p0, zero);
*(int *)comp_pred = _mm_cvtsi128_si32(p0);
comp_pred += 4;
pred += 4;
ref += 4 * 8;
}
ref += stride - (width << 3);
}
}
}
Reference in New Issue View Git Blame Copy Permalink