Mercurial > x265

changeset 9551:46de85c1be4d draft

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
asm-avx2: filter_vps[8x8]: improve 296c->263c
author Divya Manivannan <divya@multicorewareinc.com>
date Thu, 19 Feb 2015 15:18:12 +0530
parents dbce8036e0c4
children 8575ce28b986
files source/common/x86/asm-primitives.cpp source/common/x86/ipfilter8.asm
diffstat 2 files changed, 31 insertions(+-), 2 deletions(-) [+]
line wrap: on
line diff
--- a/source/common/x86/asm-primitives.cpp	Thu Feb 19 10:01:48 2015 +0530
+++ b/source/common/x86/asm-primitives.cpp	Thu Feb 19 15:18:12 2015 +0530
@@ -179,6 +179,7 @@ extern "C" {
     p.chroma[X265_CSP_I420].pu[CHROMA_420_8x32].prim  = fncdef x265_ ## fname ## _8x32_ ## cpu
 #define ALL_CHROMA_420_4x4_PU(prim, fname, cpu) ALL_CHROMA_420_4x4_PU_TYPED(prim, , fname, cpu)
 
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_8x8].filter_vps = x265_interp_4tap_vert_ps_8x8_avx2;
 
 #define ALL_CHROMA_422_CU_TYPED(prim, fncdef, fname, cpu) \
     p.chroma[X265_CSP_I422].cu[BLOCK_422_4x8].prim   = fncdef x265_ ## fname ## _4x8_ ## cpu; \
--- a/source/common/x86/ipfilter8.asm	Thu Feb 19 10:01:48 2015 +0530
+++ b/source/common/x86/ipfilter8.asm	Thu Feb 19 15:18:12 2015 +0530
@@ -3765,8 +3765,9 @@ FILTER_V4_W8_H8_H16_H32 8, 64
     paddw           m4, m0
 %endmacro
 
+%macro FILTER_VER_CHROMA_AVX2_8x8 1
 INIT_YMM avx2
-cglobal interp_4tap_vert_pp_8x8, 4, 6, 7
+cglobal interp_4tap_vert_%1_8x8, 4, 6, 7
     mov             r4d, r4m
     shl             r4d, 6
 
@@ -3780,6 +3781,7 @@ cglobal interp_4tap_vert_pp_8x8, 4, 6, 7
     lea             r4, [r1 * 3]
     sub             r0, r1
     PROCESS_CHROMA_AVX2_W8_8R
+%ifidn %1,pp
     lea             r4, [r3 * 3]
     mova            m3, [pw_512]
     pmulhrsw        m5, m3                          ; m5 = word: row 0, row 1
@@ -3799,7 +3801,33 @@ cglobal interp_4tap_vert_pp_8x8, 4, 6, 7
     movq            [r2 + r3], xm4
     movhps          [r2 + r3 * 2], xm1
     movhps          [r2 + r4], xm4
-    RET
+%else
+    add             r3d, r3d
+    vbroadcasti128  m3, [pw_2000]
+    lea             r4, [r3 * 3]
+    psubw           m5, m3                          ; m5 = word: row 0, row 1
+    psubw           m2, m3                          ; m2 = word: row 2, row 3
+    psubw           m1, m3                          ; m1 = word: row 4, row 5
+    psubw           m4, m3                          ; m4 = word: row 6, row 7
+    vextracti128    xm6, m5, 1
+    vextracti128    xm3, m2, 1
+    vextracti128    xm0, m1, 1
+    movu            [r2], xm5
+    movu            [r2 + r3], xm6
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r4], xm3
+    lea             r2, [r2 + r3 * 4]
+    movu            [r2], xm1
+    movu            [r2 + r3], xm0
+    movu            [r2 + r3 * 2], xm4
+    vextracti128    xm4, m4, 1
+    movu            [r2 + r4], xm4
+%endif
+    RET
+%endmacro
+
+FILTER_VER_CHROMA_AVX2_8x8 pp
+FILTER_VER_CHROMA_AVX2_8x8 ps
 
 ;-----------------------------------------------------------------------------
 ;void interp_4tap_vert_pp_6x8(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)