[x265] [PATCH 2 of 4] asm: intra_filter8x8 sse4 code, improved 990c->201c over C code

dnyaneshwar at multicorewareinc.com dnyaneshwar at multicorewareinc.com
Fri Jun 26 15:22:50 CEST 2015


# HG changeset patch
# User Dnyaneshwar G <dnyaneshwar at multicorewareinc.com>
# Date 1435323520 -19800
#      Fri Jun 26 18:28:40 2015 +0530
# Node ID 93c31f8b404708cd39d00b85a07b2418794fc103
# Parent  44b574b61b29a3cfba99e8f0d06622e44a86df17
asm: intra_filter8x8 sse4 code, improved 990c->201c over C code

diff -r 44b574b61b29 -r 93c31f8b4047 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Fri Jun 26 18:21:07 2015 +0530
+++ b/source/common/x86/asm-primitives.cpp	Fri Jun 26 18:28:40 2015 +0530
@@ -2454,6 +2454,7 @@
         p.weight_sp = PFX(weight_sp_sse4);
 
         p.cu[BLOCK_4x4].intra_filter = PFX(intra_filter_4x4_sse4);
+        p.cu[BLOCK_8x8].intra_filter = PFX(intra_filter_8x8_sse4);
 
         ALL_LUMA_TU_S(intra_pred[PLANAR_IDX], intra_pred_planar, sse4);
         ALL_LUMA_TU_S(intra_pred[DC_IDX], intra_pred_dc, sse4);
diff -r 44b574b61b29 -r 93c31f8b4047 source/common/x86/intrapred8.asm
--- a/source/common/x86/intrapred8.asm	Fri Jun 26 18:21:07 2015 +0530
+++ b/source/common/x86/intrapred8.asm	Fri Jun 26 18:28:40 2015 +0530
@@ -18320,3 +18320,63 @@
     mov             [r1 +  8], r2b                  ; topLast
     mov             [r1 + 16], r3b                  ; LeftLast
     RET
+
+INIT_XMM sse4
+cglobal intra_filter_8x8, 2,4,6
+    mov             r2b, byte [r0 + 16]             ; topLast
+    mov             r3b, byte [r0 + 32]             ; LeftLast
+
+    ; filtering top
+    pmovzxbw        m0, [r0 +  0]
+    pmovzxbw        m1, [r0 +  8]
+    pmovzxbw        m2, [r0 + 16]
+
+    pshufb          m4, m0, [intra_filter4_shuf0]   ; [6 5 4 3 2 1 0 1] samples[i - 1]
+    palignr         m5, m1, m0, 2
+    pinsrb          m5, [r0 + 17], 0                ; [8 7 6 5 4 3 2 9] samples[i + 1]
+
+    palignr         m3, m1, m0, 14
+    psllw           m0, 1
+    paddw           m4, m5
+    paddw           m0, m4
+    paddw           m0, [pw_2]
+    psrlw           m0, 2
+
+    palignr         m4, m2, m1, 2
+    psllw           m1, 1
+    paddw           m4, m3
+    paddw           m1, m4
+    paddw           m1, [pw_2]
+    psrlw           m1, 2
+
+    packuswb        m0, m1
+    movu            [r1], m0
+
+    ; filtering left
+    pmovzxbw        m1, [r0 + 24]
+    pmovzxbw        m0, [r0 + 32]
+
+    palignr         m4, m2, m2, 14
+    pinsrb          m4, [r0], 2
+    palignr         m5, m1, m2, 2
+
+    palignr         m3, m1, m2, 14
+    palignr         m0, m1, 2
+
+    psllw           m2, 1
+    paddw           m4, m5
+    paddw           m2, m4
+    paddw           m2, [pw_2]
+    psrlw           m2, 2
+
+    psllw           m1, 1
+    paddw           m0, m3
+    paddw           m1, m0
+    paddw           m1, [pw_2]
+    psrlw           m1, 2
+
+    packuswb        m2, m1
+    movu            [r1 + 16], m2
+    mov             [r1 + 16], r2b                  ; topLast
+    mov             [r1 + 32], r3b                  ; LeftLast
+    RET


More information about the x265-devel mailing list