SECTION .text
INIT_MMX
-%macro LOAD_DIFF_4P 4 ; mp, mt, dx, dy
- movd %1, [eax+ebx*%4+%3]
- movd %2, [ecx+edx*%4+%3]
- punpcklbw %1, %2
- punpcklbw %2, %2
- psubw %1, %2
-%endmacro
-
%macro LOAD_DIFF_4x8P 1 ; dx
- LOAD_DIFF_4P m0, m7, %1, 0
- LOAD_DIFF_4P m1, m7, %1, 1
- lea eax, [eax+2*ebx]
- lea ecx, [ecx+2*edx]
- LOAD_DIFF_4P m2, m7, %1, 0
- LOAD_DIFF_4P m3, m7, %1, 1
- lea eax, [eax+2*ebx]
- lea ecx, [ecx+2*edx]
- LOAD_DIFF_4P m4, m7, %1, 0
- LOAD_DIFF_4P m5, m7, %1, 1
- lea eax, [eax+2*ebx]
- lea ecx, [ecx+2*edx]
- LOAD_DIFF_4P m6, m7, %1, 0
- movq [spill], m6
- LOAD_DIFF_4P m7, m6, %1, 1
- movq m6, [spill]
+ LOAD_DIFF m0, m7, none, [r0+%1], [r2+%1]
+ LOAD_DIFF m1, m6, none, [r0+%1+r1], [r2+%1+r3]
+ LOAD_DIFF m2, m7, none, [r0+%1+r1*2], [r2+%1+r3*2]
+ LOAD_DIFF m3, m6, none, [r0+%1+r4], [r2+%1+r5]
+ lea r0, [r0+4*r1]
+ lea r2, [r2+4*r3]
+ LOAD_DIFF m4, m7, none, [r0+%1], [r2+%1]
+ LOAD_DIFF m5, m6, none, [r0+%1+r1], [r2+%1+r3]
+ LOAD_DIFF m6, m7, none, [r0+%1+r1*2], [r2+%1+r3*2]
+ movq [spill], m5
+ LOAD_DIFF m7, m5, none, [r0+%1+r4], [r2+%1+r5]
+ movq m5, [spill]
%endmacro
%macro SUM4x8_MM 0
;-----------------------------------------------------------------------------
; int x264_pixel_sa8d_8x8_mmxext( uint8_t *, int, uint8_t *, int )
;-----------------------------------------------------------------------------
-cglobal x264_pixel_sa8d_8x8_mmxext
- push ebx
- mov eax, [esp+ 8] ; pix1
- mov ebx, [esp+12] ; stride1
- mov ecx, [esp+16] ; pix2
- mov edx, [esp+20] ; stride2
- sub esp, 0x70
+cglobal x264_pixel_sa8d_8x8_internal_mmxext
+ push r0
+ push r2
+ sub esp, 0x74
%define args esp+0x74
%define spill esp+0x60 ; +16
%define trans esp+0 ; +96
LOAD_DIFF_4x8P 0
HADAMARD8_1D m0, m1, m2, m3, m4, m5, m6, m7
- movq [spill], m0
- TRANSPOSE4x4W 4, 5, 6, 7, 0
+ movq [spill], m1
+ TRANSPOSE4x4W 4, 5, 6, 7, 1
movq [trans+0x00], m4
movq [trans+0x08], m5
movq [trans+0x10], m6
movq [trans+0x18], m7
- movq m0, [spill]
+ movq m1, [spill]
TRANSPOSE4x4W 0, 1, 2, 3, 4
movq [trans+0x20], m0
movq [trans+0x28], m1
movq [trans+0x30], m2
movq [trans+0x38], m3
- mov eax, [args+4]
- mov ecx, [args+12]
+ mov r0, [args+4]
+ mov r2, [args]
LOAD_DIFF_4x8P 4
HADAMARD8_1D m0, m1, m2, m3, m4, m5, m6, m7
movq [trans+0x50], m2
movq [trans+0x58], m3
movq m7, [spill]
- TRANSPOSE4x4W 4, 5, 6, 7, 0
+ TRANSPOSE4x4W 4, 5, 6, 7, 1
movq m0, [trans+0x00]
movq m1, [trans+0x08]
movq m2, [trans+0x10]
SUM4x8_MM
pavgw m0, [trans]
- pshufw m1, m0, 01001110b
- paddw m0, m1
- pshufw m1, m0, 10110001b
- paddw m0, m1
- movd eax, m0
- and eax, 0xffff
- mov ecx, eax ; preserve rounding for 16x16
- add eax, 1
- shr eax, 1
- add esp, 0x70
- pop ebx
+ add esp, 0x7c
ret
%undef args
%undef spill
HADDD %1, %2
%endmacro
+%macro HADDUW 2
+ mova %2, %1
+ pslld %1, 16
+ psrld %2, 16
+ psrld %1, 16
+ paddd %1, %2
+ HADDD %1, %2
+%endmacro
+
;=============================================================================
; SSD
;=============================================================================
HADAMARD4x4_SUM %1
%endmacro
-%macro SATD_8x4_START 1
- SATD_4x4_MMX m0, 0, 0
- SATD_4x4_MMX m1, 4, %1
-%endmacro
-
-%macro SATD_8x4_INC 1
- SATD_4x4_MMX m2, 0, 0
- paddw m0, m1
- SATD_4x4_MMX m1, 4, %1
- paddw m0, m2
-%endmacro
-
-%macro SATD_16x4_START 1
- SATD_4x4_MMX m0, 0, 0
- SATD_4x4_MMX m1, 4, 0
- SATD_4x4_MMX m2, 8, 0
- paddw m0, m1
- SATD_4x4_MMX m1, 12, %1
- paddw m0, m2
-%endmacro
-
-%macro SATD_16x4_INC 1
- SATD_4x4_MMX m2, 0, 0
- paddw m0, m1
- SATD_4x4_MMX m1, 4, 0
- paddw m0, m2
- SATD_4x4_MMX m2, 8, 0
- paddw m0, m1
- SATD_4x4_MMX m1, 12, %1
- paddw m0, m2
-%endmacro
-
-%macro SATD_8x4_SSE2 1
+%macro SATD_8x4_SSE2 2
LOAD_DIFF_8x4P m0, m1, m2, m3, m4, m5
%if %1
lea r0, [r0+4*r1]
lea r2, [r2+4*r3]
%endif
HADAMARD4_1D m0, m1, m2, m3
+%ifidn %2, ssse3_phadd
+ HADAMARD4_ROW_PHADD 0, 1, 2, 3, 4
+%else
TRANSPOSE2x4x4W 0, 1, 2, 3, 4
HADAMARD4_1D m0, m1, m2, m3
- ABS4 m0, m1, m2, m3, m4, m5
- paddusw m0, m1
- paddusw m2, m3
- paddusw m6, m0
- paddusw m6, m2
-%endmacro
-
-%macro SATD_8x4_PHADD 1
- LOAD_DIFF_8x4P m0, m1, m2, m3, m4, m5
-%if %1
- lea r0, [r0+4*r1]
- lea r2, [r2+4*r3]
%endif
- HADAMARD4_1D m0, m1, m2, m3
- HADAMARD4_ROW_PHADD 0, 1, 2, 3, 4
ABS4 m0, m1, m2, m3, m4, m5
paddusw m0, m1
paddusw m2, m3
; int x264_pixel_satd_16x16_mmxext (uint8_t *, int, uint8_t *, int )
;-----------------------------------------------------------------------------
INIT_MMX
+cglobal x264_pixel_satd_16x4_internal_mmxext
+ SATD_4x4_MMX m2, 0, 0
+ SATD_4x4_MMX m1, 4, 0
+ paddw m0, m2
+ SATD_4x4_MMX m2, 8, 0
+ paddw m0, m1
+ SATD_4x4_MMX m1, 12, 0
+ paddw m0, m2
+ paddw m0, m1
+ ret
+
+cglobal x264_pixel_satd_8x8_internal_mmxext
+ SATD_4x4_MMX m2, 0, 0
+ SATD_4x4_MMX m1, 4, 1
+ paddw m0, m2
+ paddw m0, m1
+x264_pixel_satd_8x4_internal_mmxext:
+ SATD_4x4_MMX m2, 0, 0
+ SATD_4x4_MMX m1, 4, 0
+ paddw m0, m2
+ paddw m0, m1
+ ret
+
cglobal x264_pixel_satd_16x16_mmxext, 4,6
SATD_START_MMX
- SATD_16x4_START 1
- SATD_16x4_INC 1
- SATD_16x4_INC 1
- SATD_16x4_INC 0
- paddw m0, m1
- pxor m3, m3
- pshufw m1, m0, 01001110b
- paddw m0, m1
- punpcklwd m0, m3
- pshufw m1, m0, 01001110b
- paddd m0, m1
- movd eax, m0
+ pxor m0, m0
+%rep 3
+ call x264_pixel_satd_16x4_internal_mmxext
+ lea r0, [r0+4*r1]
+ lea r2, [r2+4*r3]
+%endrep
+ call x264_pixel_satd_16x4_internal_mmxext
+ HADDUW m0, m1
+ movd eax, m0
RET
cglobal x264_pixel_satd_16x8_mmxext, 4,6
SATD_START_MMX
- SATD_16x4_START 1
- SATD_16x4_INC 0
- paddw m0, m1
+ pxor m0, m0
+ call x264_pixel_satd_16x4_internal_mmxext
+ lea r0, [r0+4*r1]
+ lea r2, [r2+4*r3]
+ call x264_pixel_satd_16x4_internal_mmxext
SATD_END_MMX
cglobal x264_pixel_satd_8x16_mmxext, 4,6
SATD_START_MMX
- SATD_8x4_START 1
- SATD_8x4_INC 1
- SATD_8x4_INC 1
- SATD_8x4_INC 0
- paddw m0, m1
+ pxor m0, m0
+ call x264_pixel_satd_8x8_internal_mmxext
+ lea r0, [r0+4*r1]
+ lea r2, [r2+4*r3]
+ call x264_pixel_satd_8x8_internal_mmxext
SATD_END_MMX
cglobal x264_pixel_satd_8x8_mmxext, 4,6
SATD_START_MMX
- SATD_8x4_START 1
- SATD_8x4_INC 0
- paddw m0, m1
+ pxor m0, m0
+ call x264_pixel_satd_8x8_internal_mmxext
SATD_END_MMX
cglobal x264_pixel_satd_8x4_mmxext, 4,6
SATD_START_MMX
- SATD_8x4_START 0
- paddw m0, m1
+ pxor m0, m0
+ call x264_pixel_satd_8x4_internal_mmxext
SATD_END_MMX
%macro SATD_W4 1
;-----------------------------------------------------------------------------
%macro SATDS_SSE2 1
INIT_XMM
+cglobal x264_pixel_satd_8x8_internal_%1
+ SATD_8x4_SSE2 1, %1
+x264_pixel_satd_8x4_internal_%1:
+ SATD_8x4_SSE2 0, %1
+ ret
+
cglobal x264_pixel_satd_16x16_%1, 4,6
SATD_START_SSE2
BACKUP_POINTERS
- SATD_8x4_SSE2 1
- SATD_8x4_SSE2 1
- SATD_8x4_SSE2 1
- SATD_8x4_SSE2 0
+ call x264_pixel_satd_8x8_internal_%1
+ lea r0, [r0+4*r1]
+ lea r2, [r2+4*r3]
+ call x264_pixel_satd_8x8_internal_%1
RESTORE_AND_INC_POINTERS
- SATD_8x4_SSE2 1
- SATD_8x4_SSE2 1
- SATD_8x4_SSE2 1
- SATD_8x4_SSE2 0
+ call x264_pixel_satd_8x8_internal_%1
+ lea r0, [r0+4*r1]
+ lea r2, [r2+4*r3]
+ call x264_pixel_satd_8x8_internal_%1
SATD_END_SSE2
cglobal x264_pixel_satd_16x8_%1, 4,6
SATD_START_SSE2
BACKUP_POINTERS
- SATD_8x4_SSE2 1
- SATD_8x4_SSE2 0
+ call x264_pixel_satd_8x8_internal_%1
RESTORE_AND_INC_POINTERS
- SATD_8x4_SSE2 1
- SATD_8x4_SSE2 0
+ call x264_pixel_satd_8x8_internal_%1
SATD_END_SSE2
cglobal x264_pixel_satd_8x16_%1, 4,6
SATD_START_SSE2
- SATD_8x4_SSE2 1
- SATD_8x4_SSE2 1
- SATD_8x4_SSE2 1
- SATD_8x4_SSE2 0
+ call x264_pixel_satd_8x8_internal_%1
+ lea r0, [r0+4*r1]
+ lea r2, [r2+4*r3]
+ call x264_pixel_satd_8x8_internal_%1
SATD_END_SSE2
cglobal x264_pixel_satd_8x8_%1, 4,6
SATD_START_SSE2
- SATD_8x4_SSE2 1
- SATD_8x4_SSE2 0
+ call x264_pixel_satd_8x8_internal_%1
SATD_END_SSE2
cglobal x264_pixel_satd_8x4_%1, 4,6
SATD_START_SSE2
- SATD_8x4_SSE2 0
+ call x264_pixel_satd_8x4_internal_%1
SATD_END_SSE2
%ifdef ARCH_X86_64
;-----------------------------------------------------------------------------
; int x264_pixel_sa8d_8x8_sse2( uint8_t *, int, uint8_t *, int )
;-----------------------------------------------------------------------------
-cglobal x264_pixel_sa8d_8x8_%1
- lea r4, [3*r1]
- lea r5, [3*r3]
-.skip_lea:
- LOAD_DIFF_8x4P m0, m1, m2, m3, m8, m9
- lea r0, [r0+4*r1]
- lea r2, [r2+4*r3]
- LOAD_DIFF_8x4P m4, m5, m6, m7, m8, m9
+cglobal x264_pixel_sa8d_8x8_internal_%1
+ lea r10, [r0+4*r1]
+ lea r11, [r2+4*r3]
+ LOAD_DIFF_8x4P m0, m1, m2, m3, m8, m9, r0, r2
+ LOAD_DIFF_8x4P m4, m5, m6, m7, m8, m9, r10, r11
HADAMARD8_1D m0, m1, m2, m3, m4, m5, m6, m7
TRANSPOSE8x8W 0, 1, 2, 3, 4, 5, 6, 7, 8
paddusw m0, m2
paddusw m4, m6
pavgw m0, m4
- HADDW m0, m1
+ ret
+
+cglobal x264_pixel_sa8d_8x8_%1, 4,6
+ lea r4, [3*r1]
+ lea r5, [3*r3]
+ call x264_pixel_sa8d_8x8_internal_%1
+ HADDW m0, m1
movd eax, m0
- add r10d, eax ; preserve rounding for 16x16
add eax, 1
shr eax, 1
ret
-cglobal x264_pixel_sa8d_16x16_%1
- xor r10d, r10d
- call x264_pixel_sa8d_8x8_%1 ; pix[0]
- lea r0, [r0+4*r1]
- lea r2, [r2+4*r3]
- call x264_pixel_sa8d_8x8_%1.skip_lea ; pix[8*stride]
- neg r4 ; it's already r1*3
- neg r5
- lea r0, [r0+4*r4+8]
- lea r2, [r2+4*r5+8]
- call x264_pixel_sa8d_8x8_%1 ; pix[8]
- lea r0, [r0+4*r1]
- lea r2, [r2+4*r3]
- call x264_pixel_sa8d_8x8_%1.skip_lea ; pix[8*stride+8]
- mov eax, r10d
+cglobal x264_pixel_sa8d_16x16_%1, 4,6
+ lea r4, [3*r1]
+ lea r5, [3*r3]
+ call x264_pixel_sa8d_8x8_internal_%1 ; pix[0]
+ add r0, 8
+ add r2, 8
+ mova m10, m0
+ call x264_pixel_sa8d_8x8_internal_%1 ; pix[8]
+ lea r0, [r0+8*r1]
+ lea r2, [r2+8*r3]
+ paddusw m10, m0
+ call x264_pixel_sa8d_8x8_internal_%1 ; pix[8*stride+8]
+ sub r0, 8
+ sub r2, 8
+ paddusw m10, m0
+ call x264_pixel_sa8d_8x8_internal_%1 ; pix[8*stride]
+ paddusw m0, m10
+ HADDUW m0, m1
+ movd eax, m0
add eax, 1
shr eax, 1
ret
+
%else ; ARCH_X86_32
-cglobal x264_pixel_sa8d_8x8_%1, 4,7
- mov r6, esp
- and esp, ~15
- sub esp, 32
- lea r4, [3*r1]
- lea r5, [3*r3]
+cglobal x264_pixel_sa8d_8x8_internal_%1
LOAD_DIFF_8x4P m0, m1, m2, m3, m6, m7
- movdqa [esp], m2
+ movdqa [esp+4], m2
lea r0, [r0+4*r1]
lea r2, [r2+4*r3]
LOAD_DIFF_8x4P m4, m5, m6, m7, m2, m2
- movdqa m2, [esp]
+ movdqa m2, [esp+4]
HADAMARD8_1D m0, m1, m2, m3, m4, m5, m6, m7
- TRANSPOSE8x8W 0, 1, 2, 3, 4, 5, 6, 7, [esp], [esp+16]
+ TRANSPOSE8x8W 0, 1, 2, 3, 4, 5, 6, 7, [esp+4], [esp+20]
HADAMARD8_1D m0, m1, m2, m3, m4, m5, m6, m7
%ifidn %1, sse2
- movdqa [esp], m4
- movdqa [esp+16], m2
+ movdqa [esp+4], m4
+ movdqa [esp+20], m2
%endif
ABS2 m6, m3, m4, m2
ABS2 m0, m7, m4, m2
paddusw m0, m6
paddusw m7, m3
%ifidn %1, sse2
- movdqa m4, [esp]
- movdqa m2, [esp+16]
+ movdqa m4, [esp+4]
+ movdqa m2, [esp+20]
%endif
ABS2 m5, m1, m6, m3
ABS2 m4, m2, m6, m3
paddusw m0, m7
paddusw m5, m4
pavgw m0, m5
- HADDW m0, m7
+ ret
+%endif ; ARCH
+%endmacro ; SATDS_SSE2
+
+%macro SA8D_16x16_32 1
+%ifndef ARCH_X86_64
+cglobal x264_pixel_sa8d_8x8_%1, 4,7
+ mov r6, esp
+ and esp, ~15
+ sub esp, 32
+ lea r4, [3*r1]
+ lea r5, [3*r3]
+ call x264_pixel_sa8d_8x8_internal_%1
+ HADDW m0, m1
movd eax, m0
- mov ecx, eax ; preserve rounding for 16x16
add eax, 1
shr eax, 1
mov esp, r6
RET
-%endif ; ARCH
-%endmacro ; SATDS_SSE2
-%macro SA8D_16x16_32 1
-%ifndef ARCH_X86_64
-cglobal x264_pixel_sa8d_16x16_%1
- push ebp
- push dword [esp+20] ; stride2
- push dword [esp+20] ; pix2
- push dword [esp+20] ; stride1
- push dword [esp+20] ; pix1
- call x264_pixel_sa8d_8x8_%1
- mov ebp, ecx
- add dword [esp+0], 8 ; pix1+8
- add dword [esp+8], 8 ; pix2+8
- call x264_pixel_sa8d_8x8_%1
- add ebp, ecx
- mov eax, [esp+4]
- mov edx, [esp+12]
- shl eax, 3
- shl edx, 3
- add [esp+0], eax ; pix1+8*stride1+8
- add [esp+8], edx ; pix2+8*stride2+8
- call x264_pixel_sa8d_8x8_%1
- add ebp, ecx
- sub dword [esp+0], 8 ; pix1+8*stride1
- sub dword [esp+8], 8 ; pix2+8*stride2
- call x264_pixel_sa8d_8x8_%1
- lea eax, [ebp+ecx+1]
- shr eax, 1
- add esp, 16
- pop ebp
- ret
+cglobal x264_pixel_sa8d_16x16_%1, 4,7
+ mov r6, esp
+ and esp, ~15
+ sub esp, 48
+ lea r4, [3*r1]
+ lea r5, [3*r3]
+ call x264_pixel_sa8d_8x8_internal_%1
+ lea r0, [r0+4*r1]
+ lea r2, [r2+4*r3]
+ mova [esp+32], m0
+ call x264_pixel_sa8d_8x8_internal_%1
+ mov r0, [r6+20]
+ mov r2, [r6+28]
+ add r0, 8
+ add r2, 8
+ paddusw m0, [esp+32]
+ mova [esp+32], m0
+ call x264_pixel_sa8d_8x8_internal_%1
+ lea r0, [r0+4*r1]
+ lea r2, [r2+4*r3]
+%if mmsize == 16
+ paddusw m0, [esp+32]
+%endif
+ mova [esp+48-mmsize], m0
+ call x264_pixel_sa8d_8x8_internal_%1
+ paddusw m0, [esp+48-mmsize]
+%if mmsize == 16
+ HADDUW m0, m1
+%else
+ mova m2, [esp+32]
+ pxor m7, m7
+ mova m1, m0
+ mova m3, m2
+ punpcklwd m0, m7
+ punpckhwd m1, m7
+ punpcklwd m2, m7
+ punpckhwd m3, m7
+ paddd m0, m1
+ paddd m2, m3
+ paddd m0, m2
+ HADDD m0, m1
+%endif
+ movd eax, m0
+ add eax, 1
+ shr eax, 1
+ mov esp, r6
+ RET
%endif ; !ARCH_X86_64
%endmacro ; SA8D_16x16_32
paddusw m1, [rsp+0x18]
%endif
%if %1*%2 == 256
- paddusw m0, [rsp+0x20]
+ mova m2, [rsp+0x20]
paddusw m1, [rsp+0x28]
- paddusw m0, [rsp+0x30]
+ paddusw m2, [rsp+0x30]
+ mova m3, m0
paddusw m1, [rsp+0x38]
-%endif
+ pxor m3, m2
+ pand m3, [pw_1 GLOBAL]
+ pavgw m0, m2
+ psubusw m0, m3
+ HADDUW m0, m2
+%else
psrlw m0, 1
- psrlw m1, 1
HADDW m0, m2
+%endif
+ psrlw m1, 1
HADDW m1, m3
movd edx, m0
movd eax, m1
; instantiate satds
%ifndef ARCH_X86_64
-cextern x264_pixel_sa8d_8x8_mmxext
+cextern x264_pixel_sa8d_8x8_internal_mmxext
SA8D_16x16_32 mmxext
%endif
INTRA_SA8D_SSE2 ssse3
INTRA_SATDS_MMX ssse3
HADAMARD_AC_SSE2 ssse3
-%define SATD_8x4_SSE2 SATD_8x4_PHADD
SATDS_SSE2 ssse3_phadd
x264_predict_4x4_init( 0, predict_4x4 );
x264_predict_8x8_filter( buf2+40, edge, ALL_NEIGHBORS, ALL_NEIGHBORS );
+ // maximize sum
+ for( i=0; i<256; i++ )
+ {
+ int z = i|(i>>4);
+ z ^= z>>2;
+ z ^= z>>1;
+ buf3[i] = ~(buf4[i] = -(z&1));
+ }
+ // random pattern made of maxed pixel differences, in case an intermediate value overflows
+ for( ; i<0x1000; i++ )
+ buf3[i] = ~(buf4[i] = -(buf1[i&~0x88]&1));
+
#define TEST_PIXEL( name, align ) \
for( i = 0, ok = 1, used_asm = 0; i < 7; i++ ) \
{ \
if( pixel_asm.name[i] != pixel_ref.name[i] ) \
{ \
set_func_name( "%s_%s", #name, pixel_names[i] ); \
+ used_asm = 1; \
for( j=0; j<64; j++ ) \
{ \
- used_asm = 1; \
res_c = call_c( pixel_c.name[i], buf1, 16, buf2+j*!align, 64 ); \
res_asm = call_a( pixel_asm.name[i], buf1, 16, buf2+j*!align, 64 ); \
if( res_c != res_asm ) \
break; \
} \
} \
+ for( j=0; j<0x1000 && ok; j+=256 ) \
+ { \
+ res_c = pixel_c .name[i]( buf3+j, 16, buf4+j, 16 ); \
+ res_asm = pixel_asm.name[i]( buf3+j, 16, buf4+j, 16 ); \
+ if( res_c != res_asm ) \
+ { \
+ ok = 0; \
+ fprintf( stderr, #name "[%d]: overflow %d != %d\n", i, res_c, res_asm ); \
+ } \
+ } \
} \
} \
report( "pixel " #name " :" );
if( pixel_asm.sad_x##N[i] && pixel_asm.sad_x##N[i] != pixel_ref.sad_x##N[i] ) \
{ \
set_func_name( "sad_x%d_%s", N, pixel_names[i] ); \
+ used_asm = 1; \
for( j=0; j<64; j++) \
{ \
uint8_t *pix2 = buf2+j; \
- used_asm = 1; \
res_c[0] = pixel_c.sad[i]( buf1, 16, pix2, 64 ); \
res_c[1] = pixel_c.sad[i]( buf1, 16, pix2+6, 64 ); \
res_c[2] = pixel_c.sad[i]( buf1, 16, pix2+1, 64 ); \
{
set_func_name( "hadamard_ac_%s", pixel_names[i] );
used_asm = 1;
- uint64_t rc = pixel_c.hadamard_ac[i]( buf1, 16 );
- uint64_t ra = pixel_asm.hadamard_ac[i]( buf1, 16 );
- if( rc != ra )
+ for( j=0; j<32; j++ )
{
- ok = 0;
- fprintf( stderr, "hadamard_ac[%d]: %d,%d != %d,%d\n", i, (int)rc, (int)(rc>>32), (int)ra, (int)(ra>>32) );
+ uint8_t *pix = (j&16 ? buf1 : buf3) + (j&15)*256;
+ uint64_t rc = pixel_c.hadamard_ac[i]( pix, 16 );
+ uint64_t ra = pixel_asm.hadamard_ac[i]( pix, 16 );
+ if( rc != ra )
+ {
+ ok = 0;
+ fprintf( stderr, "hadamard_ac[%d]: %d,%d != %d,%d\n", i, (int)rc, (int)(rc>>32), (int)ra, (int)(ra>>32) );
+ break;
+ }
}
call_c2( pixel_c.hadamard_ac[i], buf1, 16 );
call_a2( pixel_asm.hadamard_ac[i], buf1, 16 );
projects / libx264 / commitdiff