[x265] [PATCH Review Only] asm code for blockcopy_sp, 8x16 partition

praveen at multicorewareinc.com praveen at multicorewareinc.com
Tue Nov 5 16:52:39 CET 2013


# HG changeset patch
# User Praveen Tiwari
# Date 1383666746 -19800
# Node ID c10e41531bbd8837a3153be41885a303ee8ce405
# Parent  8f3ef0e6ff6385c39e1be336f8b54e46e5c67ddc
asm code for blockcopy_sp, 8x16 partition

diff -r 8f3ef0e6ff63 -r c10e41531bbd source/common/x86/blockcopy8.asm
--- a/source/common/x86/blockcopy8.asm	Tue Nov 05 21:02:58 2013 +0530
+++ b/source/common/x86/blockcopy8.asm	Tue Nov 05 21:22:26 2013 +0530
@@ -931,3 +931,62 @@
 movh       [r4 + r1],       m1
 
 RET
+
+;-----------------------------------------------------------------------------
+; void blockcopy_sp_%1x%2(pixel *dest, intptr_t destStride, int16_t *src, intptr_t srcStride)
+;-----------------------------------------------------------------------------
+%macro BLOCKCOPY_SP_W8_H8 2
+INIT_XMM sse2
+cglobal blockcopy_sp_%1x%2, 4, 7, 8, dest, destStride, src, srcStride
+
+mov         r6d,       %2
+
+add         r3,      r3
+
+mova        m0,     [tab_Vm]
+
+.loop
+      movu        m1,     [r2]
+      movu        m2,     [r2 + r3]
+      movu        m3,     [r2 + 2 * r3]
+      lea         r4,     [r2 + 2 * r3]
+      movu        m4,     [r4 + r3]
+      movu        m5,     [r4 + 2 * r3]
+      lea         r4,     [r4 + 2 * r3]
+      movu        m6,     [r4 + r3]
+      movu        m7,     [r4 + 2 * r3]
+      lea         r5,     [r4 + 2 * r3]
+
+      pshufb     m1,      m0
+      pshufb     m2,      m0
+      pshufb     m3,      m0
+      pshufb     m4,      m0
+      pshufb     m5,      m0
+      pshufb     m6,      m0
+      pshufb     m7,      m0
+
+      movh       [r0],            m1
+      movh       [r0 + r1],       m2
+      movh       [r0 + 2 * r1],   m3
+      lea        r4,              [r0 + 2 * r1]
+      movh       [r4 + r1],       m4
+      movh       [r4 + 2 * r1],   m5
+      lea        r4,              [r4 + 2 * r1]
+      movh       [r4 + r1],       m6
+      movh       [r4 + 2 * r1],   m7
+
+      movu       m1,              [r5 + r3]
+      pshufb     m1,              m0
+      lea        r4,              [r4 + 2 * r1]
+      movh       [r4 + r1],       m1
+
+     lea         r0,              [r0 + 8 * r1]
+     lea         r2,              [r2 + 8 * r3]
+
+     sub         r6d,             8
+     jnz         .loop
+
+RET
+%endmacro
+
+BLOCKCOPY_SP_W8_H8 8, 16


More information about the x265-devel mailing list