[x265] [PATCH 4 of 4] asm: fix Main12 fault on saoCuOrgB0_avx2
Min Chen
chenm003 at 163.com
Wed Jul 22 01:20:05 CEST 2015
# HG changeset patch
# User Min Chen <chenm003 at 163.com>
# Date 1437514219 25200
# Node ID c0e98ac3f7edef57744edcd7d090380f4a70798f
# Parent 486e77fdd864304b5016411daf507e3fdde5b618
asm: fix Main12 fault on saoCuOrgB0_avx2
---
source/common/x86/asm-primitives.cpp | 2 --
source/common/x86/loopfilter.asm | 6 +++---
2 files changed, 3 insertions(+), 5 deletions(-)
diff -r 486e77fdd864 -r c0e98ac3f7ed source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp Tue Jul 21 14:30:16 2015 -0700
+++ b/source/common/x86/asm-primitives.cpp Tue Jul 21 14:30:19 2015 -0700
@@ -1323,9 +1323,7 @@
p.saoCuOrgE2[1] = PFX(saoCuOrgE2_32_avx2);
p.saoCuOrgE3[0] = PFX(saoCuOrgE3_avx2);
p.saoCuOrgE3[1] = PFX(saoCuOrgE3_32_avx2);
-#if X265_DEPTH <= 10
p.saoCuOrgB0 = PFX(saoCuOrgB0_avx2);
-#endif
p.cu[BLOCK_16x16].intra_pred[2] = PFX(intra_pred_ang16_2_avx2);
p.cu[BLOCK_16x16].intra_pred[3] = PFX(intra_pred_ang16_3_avx2);
diff -r 486e77fdd864 -r c0e98ac3f7ed source/common/x86/loopfilter.asm
--- a/source/common/x86/loopfilter.asm Tue Jul 21 14:30:16 2015 -0700
+++ b/source/common/x86/loopfilter.asm Tue Jul 21 14:30:19 2015 -0700
@@ -1666,8 +1666,8 @@
.loopW
movu m2, [r0]
movu m5, [r0 + r4]
- psrlw m0, m2, 5
- psrlw m6, m5, 5
+ psrlw m0, m2, (BIT_DEPTH - 5)
+ psrlw m6, m5, (BIT_DEPTH - 5)
packuswb m0, m6
vpermq m0, m0, 11011000b
pand m0, [pb_31] ; m0 = [index]
@@ -1706,7 +1706,7 @@
xor r1, r1
.loopW1:
movu m2, [r0 + r1]
- psrlw m0, m2, 5
+ psrlw m0, m2, (BIT_DEPTH - 5)
packuswb m0, m0
vpermq m0, m0, 10001000b
pand m0, [pb_31] ; m0 = [index]
More information about the x265-devel
mailing list