[x265] [PATCH] asm: 10bpp code for pixel_sub_64xN
murugan at multicorewareinc.com
murugan at multicorewareinc.com
Fri Dec 6 09:31:20 CET 2013
# HG changeset patch
# User Murugan Vairavel <murugan at multicorewareinc.com>
# Date 1386318564 -19800
# Fri Dec 06 13:59:24 2013 +0530
# Node ID a87aa775087d5fdd3a75d5f3f599178034cf2db1
# Parent b9fcce63ab5cc543b5bdaad68c1aff2e64f8f523
asm: 10bpp code for pixel_sub_64xN
diff -r b9fcce63ab5c -r a87aa775087d source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp Fri Dec 06 13:45:37 2013 +0530
+++ b/source/common/x86/asm-primitives.cpp Fri Dec 06 13:59:24 2013 +0530
@@ -659,6 +659,10 @@
p.luma_sub_ps[LUMA_16x64] = x265_pixel_sub_ps_16x64_sse2;
p.luma_sub_ps[LUMA_32x64] = x265_pixel_sub_ps_32x64_sse2;
p.luma_sub_ps[LUMA_48x64] = x265_pixel_sub_ps_48x64_sse2;
+ p.luma_sub_ps[LUMA_64x16] = x265_pixel_sub_ps_64x16_sse2;
+ p.luma_sub_ps[LUMA_64x32] = x265_pixel_sub_ps_64x32_sse2;
+ p.luma_sub_ps[LUMA_64x48] = x265_pixel_sub_ps_64x48_sse2;
+ p.luma_sub_ps[LUMA_64x64] = x265_pixel_sub_ps_64x64_sse2;
}
if (cpuMask & X265_CPU_SSSE3)
{
diff -r b9fcce63ab5c -r a87aa775087d source/common/x86/pixel-util8.asm
--- a/source/common/x86/pixel-util8.asm Fri Dec 06 13:45:37 2013 +0530
+++ b/source/common/x86/pixel-util8.asm Fri Dec 06 13:59:24 2013 +0530
@@ -3451,15 +3451,92 @@
; void pixel_sub_ps_c_%1x%2(int16_t *dest, intptr_t destride, pixel *src0, pixel *src1, intptr_t srcstride0, intptr_t srcstride1);
;-----------------------------------------------------------------------------
%macro PIXELSUB_PS_W64_H2 2
-INIT_XMM sse4
+%if HIGH_BIT_DEPTH
+cglobal pixel_sub_ps_%1x%2, 6, 7, 6, dest, deststride, src0, src1, srcstride0, srcstride1
+ add r1, r1
+ mov r6d, %2/2
+ add r4, r4
+ add r5, r5
+.loop
+ movu m0, [r2]
+ movu m1, [r3]
+ movu m2, [r2 + 16]
+ movu m3, [r3 + 16]
+ movu m4, [r2 + 32]
+ movu m5, [r3 + 32]
+ psubw m0, m1
+ psubw m2, m3
+ psubw m4, m5
+ movu m3, [r2 + 48]
+ movu m5, [r3 + 48]
+ psubw m3, m5
+
+ movu [r0], m0
+ movu [r0 + 16], m2
+ movu [r0 + 32], m4
+ movu [r0 + 48], m3
+
+ movu m0, [r2 + 64]
+ movu m1, [r3 + 64]
+ movu m2, [r2 + 80]
+ movu m3, [r3 + 80]
+ movu m4, [r2 + 96]
+ movu m5, [r3 + 96]
+ psubw m0, m1
+ psubw m2, m3
+ psubw m4, m5
+ movu m3, [r2 + 112]
+ movu m5, [r3 + 112]
+ psubw m3, m5
+
+ movu [r0 + 64], m0
+ movu [r0 + 80], m2
+ movu [r0 + 96], m4
+ movu [r0 + 112], m3
+
+ movu m0, [r2 + r4]
+ movu m1, [r3 + r5]
+ movu m2, [r2 + r4 + 16]
+ movu m3, [r3 + r5 + 16]
+ movu m4, [r2 + r4 + 32]
+ movu m5, [r3 + r5 + 32]
+ psubw m0, m1
+ psubw m2, m3
+ psubw m4, m5
+ movu m3, [r2 + r4 + 48]
+ movu m5, [r3 + r5 + 48]
+ psubw m3, m5
+
+ movu [r0 + r1], m0
+ movu [r0 + r1 + 16], m2
+ movu [r0 + r1 + 32], m4
+ movu [r0 + r1 + 48], m3
+
+ movu m0, [r2 + r4 + 64]
+ movu m1, [r3 + r5 + 64]
+ movu m2, [r2 + r4 + 80]
+ movu m3, [r3 + r5 + 80]
+ movu m4, [r2 + r4 + 96]
+ movu m5, [r3 + r5 + 96]
+ psubw m0, m1
+ psubw m2, m3
+ psubw m4, m5
+ movu m3, [r2 + r4 + 112]
+ movu m5, [r3 + r5 + 112]
+ psubw m3, m5
+
+ movu [r0 + r1 + 64], m0
+ movu [r0 + r1 + 80], m2
+ movu [r0 + r1 + 96], m4
+ movu [r0 + r1 + 112], m3
+
+%else
+
cglobal pixel_sub_ps_%1x%2, 6, 7, 7, dest, deststride, src0, src1, srcstride0, srcstride1
-
-add r1, r1
-mov r6d, %2/2
-pxor m6, m6
-
+ add r1, r1
+ mov r6d, %2/2
+ pxor m6, m6
.loop
-
movu m1, [r2]
pmovzxbw m0, m1
punpckhbw m1, m6
@@ -3469,7 +3546,6 @@
movu m5, [r2 + 16]
pmovzxbw m4, m5
punpckhbw m5, m6
-
psubw m0, m2
psubw m1, m3
@@ -3482,7 +3558,6 @@
movu m3, [r2 + 32]
pmovzxbw m2, m3
punpckhbw m3, m6
-
psubw m4, m0
psubw m5, m1
@@ -3495,7 +3570,6 @@
movu m1, [r2 + 48]
pmovzxbw m0, m1
punpckhbw m1, m6
-
psubw m2, m4
psubw m3, m5
@@ -3508,7 +3582,6 @@
movu m5, [r2 + r4]
pmovzxbw m4, m5
punpckhbw m5, m6
-
psubw m0, m2
psubw m1, m3
@@ -3521,7 +3594,6 @@
movu m3, [r2 + r4 + 16]
pmovzxbw m2, m3
punpckhbw m3, m6
-
psubw m4, m0
psubw m5, m1
@@ -3534,7 +3606,6 @@
movu m1, [r2 + r4 + 32]
pmovzxbw m0, m1
punpckhbw m1, m6
-
psubw m2, m4
psubw m3, m5
@@ -3547,7 +3618,6 @@
movu m5, [r2 + r4 + 48]
pmovzxbw m4, m5
punpckhbw m5, m6
-
psubw m0, m2
psubw m1, m3
@@ -3557,28 +3627,33 @@
movu m1, [r3 + r5 + 48]
pmovzxbw m0, m1
punpckhbw m1, m6
-
psubw m4, m0
psubw m5, m1
movu [r0 + r1 + 96], m4
movu [r0 + r1 + 112], m5
-
- lea r2, [r2 + 2 * r4]
- lea r3, [r3 + 2 * r5]
- lea r0, [r0 + 2 * r1]
-
+%endif
dec r6d
-
-jnz .loop
-
-RET
+ lea r2, [r2 + 2 * r4]
+ lea r3, [r3 + 2 * r5]
+ lea r0, [r0 + 2 * r1]
+ jnz .loop
+ RET
%endmacro
+%if HIGH_BIT_DEPTH
+INIT_XMM sse2
PIXELSUB_PS_W64_H2 64, 16
PIXELSUB_PS_W64_H2 64, 32
PIXELSUB_PS_W64_H2 64, 48
PIXELSUB_PS_W64_H2 64, 64
+%else
+INIT_XMM sse4
+PIXELSUB_PS_W64_H2 64, 16
+PIXELSUB_PS_W64_H2 64, 32
+PIXELSUB_PS_W64_H2 64, 48
+PIXELSUB_PS_W64_H2 64, 64
+%endif
;=============================================================================
; variance
More information about the x265-devel
mailing list