diff options
Diffstat (limited to 'libavcodec/x86/h264_deblock_10bit.asm')
-rw-r--r-- | libavcodec/x86/h264_deblock_10bit.asm | 20 |
1 files changed, 14 insertions, 6 deletions
diff --git a/libavcodec/x86/h264_deblock_10bit.asm b/libavcodec/x86/h264_deblock_10bit.asm index baac725..ee31625 100644 --- a/libavcodec/x86/h264_deblock_10bit.asm +++ b/libavcodec/x86/h264_deblock_10bit.asm @@ -165,7 +165,7 @@ cglobal deblock_v_luma_10_%1, 5,5,8*(mmsize/16) SUB rsp, pad shl r2d, 2 shl r3d, 2 - LOAD_AB m4, m5, r2, r3 + LOAD_AB m4, m5, r2d, r3d mov r3, 32/mmsize mov r2, r0 sub r0, r1 @@ -222,7 +222,7 @@ cglobal deblock_h_luma_10_%1, 5,6,8*(mmsize/16) SUB rsp, pad shl r2d, 2 shl r3d, 2 - LOAD_AB m4, m5, r2, r3 + LOAD_AB m4, m5, r2d, r3d mov r3, r1 mova am, m4 add r3, r1 @@ -352,7 +352,7 @@ cglobal deblock_v_luma_10_%1, 5,5,15 %define mask2 m11 shl r2d, 2 shl r3d, 2 - LOAD_AB m12, m13, r2, r3 + LOAD_AB m12, m13, r2d, r3d mov r2, r0 sub r0, r1 sub r0, r1 @@ -380,7 +380,7 @@ cglobal deblock_v_luma_10_%1, 5,5,15 cglobal deblock_h_luma_10_%1, 5,7,15 shl r2d, 2 shl r3d, 2 - LOAD_AB m12, m13, r2, r3 + LOAD_AB m12, m13, r2d, r3d mov r2, r1 add r2, r1 add r2, r1 @@ -419,9 +419,11 @@ cglobal deblock_h_luma_10_%1, 5,7,15 INIT_XMM DEBLOCK_LUMA_64 sse2 +%ifdef HAVE_AVX INIT_AVX DEBLOCK_LUMA_64 avx %endif +%endif %macro SWAPMOVA 2 %ifid %1 @@ -714,8 +716,10 @@ cglobal deblock_h_luma_intra_10_%1, 4,7,16 INIT_XMM DEBLOCK_LUMA_INTRA_64 sse2 +%ifdef HAVE_AVX INIT_AVX DEBLOCK_LUMA_INTRA_64 avx +%endif %endif @@ -799,10 +803,12 @@ DEBLOCK_LUMA_INTRA mmxext INIT_XMM DEBLOCK_LUMA sse2 DEBLOCK_LUMA_INTRA sse2 +%ifdef HAVE_AVX INIT_AVX DEBLOCK_LUMA avx DEBLOCK_LUMA_INTRA avx %endif +%endif ; in: %1=p0, %2=q0, %3=p1, %4=q1, %5=mask, %6=tmp, %7=tmp ; out: %1=p0', %2=q0' @@ -858,7 +864,7 @@ cglobal deblock_v_chroma_10_%1, 5,7-(mmsize/16),8*(mmsize/16) .loop: %endif CHROMA_V_LOAD r5 - LOAD_AB m4, m5, r2, r3 + LOAD_AB m4, m5, r2d, r3d LOAD_MASK m0, m1, m2, m3, m4, m5, m7, m6, m4 pxor m4, m4 CHROMA_V_LOAD_TC m6, r4 @@ -892,7 +898,7 @@ cglobal deblock_v_chroma_intra_10_%1, 4,6-(mmsize/16),8*(mmsize/16) .loop: %endif CHROMA_V_LOAD r4 - LOAD_AB m4, m5, r2, r3 + LOAD_AB m4, m5, r2d, r3d LOAD_MASK m0, m1, m2, m3, m4, m5, m7, m6, m4 CHROMA_DEBLOCK_P0_Q0_INTRA m1, m2, m0, m3, m7, m5, m6 CHROMA_V_STORE @@ -913,5 +919,7 @@ DEBLOCK_CHROMA mmxext %endif INIT_XMM DEBLOCK_CHROMA sse2 +%ifdef HAVE_AVX INIT_AVX DEBLOCK_CHROMA avx +%endif |