Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(1179)

Issues Search
    My Issues | Starred     Open | Closed | All

Unified Diff: source/libvpx/vp9/common/x86/vp9_subpixel_8t_intrin_ssse3.c

Issue 812033011: libvpx: Pull from upstream (Closed) Base URL: svn://svn.chromium.org/chrome/trunk/deps/third_party/libvpx/
Patch Set: Created 5 years, 11 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View side-by-side diff with in-line comments
Download patch
« no previous file with comments | « source/libvpx/vp9/common/x86/vp9_loopfilter_mmx.asm ('k') | source/libvpx/vp9/common/x86/vp9_subpixel_8t_ssse3.asm » ('j') | no next file with comments »
Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
Index: source/libvpx/vp9/common/x86/vp9_subpixel_8t_intrin_ssse3.c
===================================================================
--- source/libvpx/vp9/common/x86/vp9_subpixel_8t_intrin_ssse3.c (revision 293588)
+++ source/libvpx/vp9/common/x86/vp9_subpixel_8t_intrin_ssse3.c (working copy)
@@ -312,9 +312,11 @@
unsigned int out_pitch,
unsigned int output_height,
int16_t *filter) {
- __m128i addFilterReg64, filtersReg, minReg, srcRegFilt6;
+ __m128i addFilterReg64, filtersReg, minReg;
__m128i firstFilters, secondFilters, thirdFilters, forthFilters;
- __m128i srcRegFilt1, srcRegFilt2, srcRegFilt3, srcRegFilt4, srcRegFilt5;
+ __m128i srcRegFilt1, srcRegFilt2, srcRegFilt3, srcRegFilt5;
+ __m128i srcReg1, srcReg2, srcReg3, srcReg4, srcReg5, srcReg6, srcReg7;
+ __m128i srcReg8;
unsigned int i;
// create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
@@ -333,27 +335,26 @@
// duplicate only the forth 16 bits in the filter
forthFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x706u));
+ // load the first 7 rows of 8 bytes
+ srcReg1 = _mm_loadl_epi64((__m128i *)&src_ptr[0]);
+ srcReg2 = _mm_loadl_epi64((__m128i *)&(src_ptr + src_pitch)[0]);
+ srcReg3 = _mm_loadl_epi64((__m128i *)&(src_ptr + src_pitch * 2)[0]);
+ srcReg4 = _mm_loadl_epi64((__m128i *)&(src_ptr + src_pitch * 3)[0]);
+ srcReg5 = _mm_loadl_epi64((__m128i *)&(src_ptr + src_pitch * 4)[0]);
+ srcReg6 = _mm_loadl_epi64((__m128i *)&(src_ptr + src_pitch * 5)[0]);
+ srcReg7 = _mm_loadl_epi64((__m128i *)&(src_ptr + src_pitch * 6)[0]);
+
for (i = 0; i < output_height; i++) {
- // load the first 8 bytes
- srcRegFilt1 = _mm_loadl_epi64((__m128i *)&src_ptr[0]);
- // load the next 8 bytes in stride of src_pitch
- srcRegFilt2 = _mm_loadl_epi64((__m128i *)&(src_ptr+src_pitch)[0]);
- srcRegFilt3 = _mm_loadl_epi64((__m128i *)&(src_ptr+src_pitch*2)[0]);
- srcRegFilt4 = _mm_loadl_epi64((__m128i *)&(src_ptr+src_pitch*3)[0]);
+ // load the last 8 bytes
+ srcReg8 = _mm_loadl_epi64((__m128i *)&(src_ptr + src_pitch * 7)[0]);
// merge the result together
- srcRegFilt1 = _mm_unpacklo_epi8(srcRegFilt1, srcRegFilt2);
- srcRegFilt3 = _mm_unpacklo_epi8(srcRegFilt3, srcRegFilt4);
+ srcRegFilt1 = _mm_unpacklo_epi8(srcReg1, srcReg2);
+ srcRegFilt3 = _mm_unpacklo_epi8(srcReg3, srcReg4);
- // load the next 8 bytes in stride of src_pitch
- srcRegFilt2 = _mm_loadl_epi64((__m128i *)&(src_ptr+src_pitch*4)[0]);
- srcRegFilt4 = _mm_loadl_epi64((__m128i *)&(src_ptr+src_pitch*5)[0]);
- srcRegFilt5 = _mm_loadl_epi64((__m128i *)&(src_ptr+src_pitch*6)[0]);
- srcRegFilt6 = _mm_loadl_epi64((__m128i *)&(src_ptr+src_pitch*7)[0]);
-
// merge the result together
- srcRegFilt2 = _mm_unpacklo_epi8(srcRegFilt2, srcRegFilt4);
- srcRegFilt5 = _mm_unpacklo_epi8(srcRegFilt5, srcRegFilt6);
+ srcRegFilt2 = _mm_unpacklo_epi8(srcReg5, srcReg6);
+ srcRegFilt5 = _mm_unpacklo_epi8(srcReg7, srcReg8);
// multiply 2 adjacent elements with the filter and add the result
srcRegFilt1 = _mm_maddubs_epi16(srcRegFilt1, firstFilters);
@@ -377,6 +378,15 @@
src_ptr+=src_pitch;
+ // shift down a row
+ srcReg1 = srcReg2;
+ srcReg2 = srcReg3;
+ srcReg3 = srcReg4;
+ srcReg4 = srcReg5;
+ srcReg5 = srcReg6;
+ srcReg6 = srcReg7;
+ srcReg7 = srcReg8;
+
// save only 8 bytes convolve result
_mm_storel_epi64((__m128i*)&output_ptr[0], srcRegFilt1);
@@ -390,9 +400,11 @@
unsigned int out_pitch,
unsigned int output_height,
int16_t *filter) {
- __m128i addFilterReg64, filtersReg, srcRegFilt1, srcRegFilt2, srcRegFilt3;
+ __m128i addFilterReg64, filtersReg, srcRegFilt1, srcRegFilt3;
__m128i firstFilters, secondFilters, thirdFilters, forthFilters;
- __m128i srcRegFilt4, srcRegFilt5, srcRegFilt6, srcRegFilt7, srcRegFilt8;
+ __m128i srcRegFilt5, srcRegFilt6, srcRegFilt7, srcRegFilt8;
+ __m128i srcReg1, srcReg2, srcReg3, srcReg4, srcReg5, srcReg6, srcReg7;
+ __m128i srcReg8;
unsigned int i;
// create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
@@ -411,19 +423,24 @@
// duplicate only the forth 16 bits in the filter
forthFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x706u));
+ // load the first 7 rows of 16 bytes
+ srcReg1 = _mm_loadu_si128((__m128i *)(src_ptr));
+ srcReg2 = _mm_loadu_si128((__m128i *)(src_ptr + src_pitch));
+ srcReg3 = _mm_loadu_si128((__m128i *)(src_ptr + src_pitch * 2));
+ srcReg4 = _mm_loadu_si128((__m128i *)(src_ptr + src_pitch * 3));
+ srcReg5 = _mm_loadu_si128((__m128i *)(src_ptr + src_pitch * 4));
+ srcReg6 = _mm_loadu_si128((__m128i *)(src_ptr + src_pitch * 5));
+ srcReg7 = _mm_loadu_si128((__m128i *)(src_ptr + src_pitch * 6));
+
for (i = 0; i < output_height; i++) {
- // load the first 16 bytes
- srcRegFilt1 = _mm_loadu_si128((__m128i *)(src_ptr));
- // load the next 16 bytes in stride of src_pitch
- srcRegFilt2 = _mm_loadu_si128((__m128i *)(src_ptr+src_pitch));
- srcRegFilt3 = _mm_loadu_si128((__m128i *)(src_ptr+src_pitch*6));
- srcRegFilt4 = _mm_loadu_si128((__m128i *)(src_ptr+src_pitch*7));
+ // load the last 16 bytes
+ srcReg8 = _mm_loadu_si128((__m128i *)(src_ptr + src_pitch * 7));
// merge the result together
- srcRegFilt5 = _mm_unpacklo_epi8(srcRegFilt1, srcRegFilt2);
- srcRegFilt6 = _mm_unpacklo_epi8(srcRegFilt3, srcRegFilt4);
- srcRegFilt1 = _mm_unpackhi_epi8(srcRegFilt1, srcRegFilt2);
- srcRegFilt3 = _mm_unpackhi_epi8(srcRegFilt3, srcRegFilt4);
+ srcRegFilt5 = _mm_unpacklo_epi8(srcReg1, srcReg2);
+ srcRegFilt6 = _mm_unpacklo_epi8(srcReg7, srcReg8);
+ srcRegFilt1 = _mm_unpackhi_epi8(srcReg1, srcReg2);
+ srcRegFilt3 = _mm_unpackhi_epi8(srcReg7, srcReg8);
// multiply 2 adjacent elements with the filter and add the result
srcRegFilt5 = _mm_maddubs_epi16(srcRegFilt5, firstFilters);
@@ -435,25 +452,17 @@
srcRegFilt5 = _mm_adds_epi16(srcRegFilt5, srcRegFilt6);
srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt3);
- // load the next 16 bytes in stride of two/three src_pitch
- srcRegFilt2 = _mm_loadu_si128((__m128i *)(src_ptr+src_pitch*2));
- srcRegFilt3 = _mm_loadu_si128((__m128i *)(src_ptr+src_pitch*3));
-
// merge the result together
- srcRegFilt4 = _mm_unpacklo_epi8(srcRegFilt2, srcRegFilt3);
- srcRegFilt6 = _mm_unpackhi_epi8(srcRegFilt2, srcRegFilt3);
+ srcRegFilt3 = _mm_unpacklo_epi8(srcReg3, srcReg4);
+ srcRegFilt6 = _mm_unpackhi_epi8(srcReg3, srcReg4);
// multiply 2 adjacent elements with the filter and add the result
- srcRegFilt4 = _mm_maddubs_epi16(srcRegFilt4, secondFilters);
+ srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, secondFilters);
srcRegFilt6 = _mm_maddubs_epi16(srcRegFilt6, secondFilters);
- // load the next 16 bytes in stride of four/five src_pitch
- srcRegFilt2 = _mm_loadu_si128((__m128i *)(src_ptr+src_pitch*4));
- srcRegFilt3 = _mm_loadu_si128((__m128i *)(src_ptr+src_pitch*5));
-
// merge the result together
- srcRegFilt7 = _mm_unpacklo_epi8(srcRegFilt2, srcRegFilt3);
- srcRegFilt8 = _mm_unpackhi_epi8(srcRegFilt2, srcRegFilt3);
+ srcRegFilt7 = _mm_unpacklo_epi8(srcReg5, srcReg6);
+ srcRegFilt8 = _mm_unpackhi_epi8(srcReg5, srcReg6);
// multiply 2 adjacent elements with the filter and add the result
srcRegFilt7 = _mm_maddubs_epi16(srcRegFilt7, thirdFilters);
@@ -461,13 +470,13 @@
// add and saturate the results together
srcRegFilt5 = _mm_adds_epi16(srcRegFilt5,
- _mm_min_epi16(srcRegFilt4, srcRegFilt7));
+ _mm_min_epi16(srcRegFilt3, srcRegFilt7));
srcRegFilt1 = _mm_adds_epi16(srcRegFilt1,
_mm_min_epi16(srcRegFilt6, srcRegFilt8));
// add and saturate the results together
srcRegFilt5 = _mm_adds_epi16(srcRegFilt5,
- _mm_max_epi16(srcRegFilt4, srcRegFilt7));
+ _mm_max_epi16(srcRegFilt3, srcRegFilt7));
srcRegFilt1 = _mm_adds_epi16(srcRegFilt1,
_mm_max_epi16(srcRegFilt6, srcRegFilt8));
srcRegFilt5 = _mm_adds_epi16(srcRegFilt5, addFilterReg64);
@@ -484,6 +493,15 @@
src_ptr+=src_pitch;
+ // shift down a row
+ srcReg1 = srcReg2;
+ srcReg2 = srcReg3;
+ srcReg3 = srcReg4;
+ srcReg4 = srcReg5;
+ srcReg5 = srcReg6;
+ srcReg6 = srcReg7;
+ srcReg7 = srcReg8;
+
// save 16 bytes convolve result
_mm_store_si128((__m128i*)output_ptr, srcRegFilt1);
« no previous file with comments | « source/libvpx/vp9/common/x86/vp9_loopfilter_mmx.asm ('k') | source/libvpx/vp9/common/x86/vp9_subpixel_8t_ssse3.asm » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698