Improved mmx/sse2 versions of iwalsh

Removed unnecessary transposes.

Change-Id: I029fbaf8afafee34d54a4f3333c22023c15003c3

diff --git a/vp8/common/x86/iwalsh_mmx.asm b/vp8/common/x86/iwalsh_mmx.asm
index 3ab066b..6582687 100644 (file)
--- a/vp8/common/x86/iwalsh_mmx.asm
+++ b/vp8/common/x86/iwalsh_mmx.asm
@@ -17,160 +17,123 @@ sym(vp8_short_inv_walsh4x4_mmx):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 2
-    push        rsi
-    push        rdi
     ; end prolog
 
-    mov     rax, 3
-    mov     rsi, arg(0)
-    mov     rdi, arg(1)
-    shl     rax, 16
+    mov         rdx, arg(0)
+    mov         rax, 30003h
 
-    movq    mm0, [rsi + 0]        ;ip[0]
-    movq    mm1, [rsi + 8]        ;ip[4]
-    or      rax, 3            ;00030003h
+    movq        mm0, [rdx + 0]    ;ip[0]
+    movq        mm1, [rdx + 8]    ;ip[4]
+    movd        mm7, rax
 
-    movq    mm2, [rsi + 16]       ;ip[8]
-    movq    mm3, [rsi + 24]       ;ip[12]
+    movq        mm2, [rdx + 16]   ;ip[8]
+    movq        mm3, [rdx + 24]   ;ip[12]
+    punpcklwd   mm7, mm7          ;0003000300030003h
+    mov         rdx, arg(1)
 
-    movq    mm7, rax
-    movq    mm4, mm0
+    movq        mm4, mm0
+    movq        mm5, mm1
 
-    punpcklwd mm7, mm7          ;0003000300030003h
-    movq    mm5, mm1
+    paddw       mm4, mm3          ;ip[0] + ip[12] aka al
+    paddw       mm5, mm2          ;ip[4] + ip[8] aka bl
 
-    paddw   mm4, mm3          ;ip[0] + ip[12] aka al
-    paddw   mm5, mm2          ;ip[4] + ip[8] aka bl
+    movq        mm6, mm4          ;temp al
+    paddw       mm4, mm5          ;al + bl
+    psubw       mm6, mm5          ;al - bl
 
-    movq    mm6, mm4          ;temp al
+    psubw       mm0, mm3          ;ip[0] - ip[12] aka d1
+    psubw       mm1, mm2          ;ip[4] - ip[8] aka c1
 
-    paddw   mm4, mm5          ;al + bl
-    psubw   mm6, mm5          ;al - bl
-
-    psubw   mm0, mm3          ;ip[0] - ip[12] aka d1
-    psubw   mm1, mm2          ;ip[4] - ip[8] aka c1
-
-    movq    mm5, mm0          ;temp dl
-
-    paddw   mm0, mm1          ;dl + cl
-    psubw   mm5, mm1          ;dl - cl
+    movq        mm5, mm0          ;temp dl
+    paddw       mm0, mm1          ;dl + cl
+    psubw       mm5, mm1          ;dl - cl
 
     ; 03 02 01 00
     ; 13 12 11 10
     ; 23 22 21 20
     ; 33 32 31 30
 
-    movq    mm3, mm4          ; 03 02 01 00
-    punpcklwd mm4, mm0          ; 11 01 10 00
-    punpckhwd mm3, mm0          ; 13 03 12 02
+    movq        mm3, mm4          ; 03 02 01 00
+    punpcklwd   mm4, mm0          ; 11 01 10 00
+    punpckhwd   mm3, mm0          ; 13 03 12 02
 
-    movq    mm1, mm6          ; 23 22 21 20
-    punpcklwd mm6, mm5          ; 31 21 30 20
-    punpckhwd mm1, mm5          ; 33 23 32 22
+    movq        mm1, mm6          ; 23 22 21 20
+    punpcklwd   mm6, mm5          ; 31 21 30 20
+    punpckhwd   mm1, mm5          ; 33 23 32 22
 
-    movq    mm0, mm4          ; 11 01 10 00
-    movq    mm2, mm3          ; 13 03 12 02
+    movq        mm0, mm4          ; 11 01 10 00
+    movq        mm2, mm3          ; 13 03 12 02
 
-    punpckldq mm0, mm6          ; 30 20 10 00 aka ip[0]
-    punpckhdq mm4, mm6          ; 31 21 11 01 aka ip[4]
+    punpckldq   mm0, mm6          ; 30 20 10 00 aka ip[0]
+    punpckhdq   mm4, mm6          ; 31 21 11 01 aka ip[4]
 
-    punpckldq mm2, mm1          ; 32 22 12 02 aka ip[8]
-    punpckhdq mm3, mm1          ; 33 23 13 03 aka ip[12]
+    punpckldq   mm2, mm1          ; 32 22 12 02 aka ip[8]
+    punpckhdq   mm3, mm1          ; 33 23 13 03 aka ip[12]
 ;~~~~~~~~~~~~~~~~~~~~~
-    movq    mm1, mm0
-    movq    mm5, mm4
-
-    paddw   mm1, mm3          ;ip[0] + ip[12] aka al
-    paddw   mm5, mm2          ;ip[4] + ip[8] aka bl
-
-    movq    mm6, mm1          ;temp al
-
-    paddw   mm1, mm5          ;al + bl
-    psubw   mm6, mm5          ;al - bl
-
-    psubw   mm0, mm3          ;ip[0] - ip[12] aka d1
-    psubw   mm4, mm2          ;ip[4] - ip[8] aka c1
-
-    movq    mm5, mm0          ;temp dl
-
-    paddw   mm0, mm4          ;dl + cl
-    psubw   mm5, mm4          ;dl - cl
+    movq        mm1, mm0
+    movq        mm5, mm4
+    paddw       mm1, mm3          ;ip[0] + ip[12] aka al
+    paddw       mm5, mm2          ;ip[4] + ip[8] aka bl
+
+    movq        mm6, mm1          ;temp al
+    paddw       mm1, mm5          ;al + bl
+    psubw       mm6, mm5          ;al - bl
+    paddw       mm1, mm7
+    paddw       mm6, mm7
+    psraw       mm1, 3
+    psraw       mm6, 3
+
+    psubw       mm0, mm3          ;ip[0] - ip[12] aka d1
+    psubw       mm4, mm2          ;ip[4] - ip[8] aka c1
+
+    movq        mm5, mm0          ;temp dl
+    paddw       mm0, mm4          ;dl + cl
+    psubw       mm5, mm4          ;dl - cl
+    paddw       mm0, mm7
+    paddw       mm5, mm7
+    psraw       mm0, 3
+    psraw       mm5, 3
 ;~~~~~~~~~~~~~~~~~~~~~
-    movq    mm3, mm1          ; 03 02 01 00
-    punpcklwd mm1, mm0          ; 11 01 10 00
-    punpckhwd mm3, mm0          ; 13 03 12 02
-
-    movq    mm4, mm6          ; 23 22 21 20
-    punpcklwd mm6, mm5          ; 31 21 30 20
-    punpckhwd mm4, mm5          ; 33 23 32 22
-
-    movq    mm0, mm1          ; 11 01 10 00
-    movq    mm2, mm3          ; 13 03 12 02
-
-    punpckldq mm0, mm6          ; 30 20 10 00 aka ip[0]
-    punpckhdq mm1, mm6          ; 31 21 11 01 aka ip[4]
-
-    punpckldq mm2, mm4          ; 32 22 12 02 aka ip[8]
-    punpckhdq mm3, mm4          ; 33 23 13 03 aka ip[12]
-
-    paddw   mm0, mm7
-    paddw   mm1, mm7
-    paddw   mm2, mm7
-    paddw   mm3, mm7
-
-    psraw   mm0, 3
-    psraw   mm1, 3
-    psraw   mm2, 3
-    psraw   mm3, 3
-
-;    movq  [rdi + 0], mm0
-;    movq  [rdi + 8], mm1
-;    movq  [rdi + 16], mm2
-;    movq  [rdi + 24], mm3
-
-    movd    eax, mm0
-    psrlq   mm0, 32
-    mov     word ptr[rdi+32*0], ax
-    shr     eax, 16
-    mov     word ptr[rdi+32*1], ax
-    movd    eax, mm0
-    mov     word ptr[rdi+32*2], ax
-    shr     eax, 16
-    mov     word ptr[rdi+32*3], ax
-
-    movd    ecx, mm1
-    psrlq   mm1, 32
-    mov     word ptr[rdi+32*4], cx
-    shr     ecx, 16
-    mov     word ptr[rdi+32*5], cx
-    movd    ecx, mm1
-    mov     word ptr[rdi+32*6], cx
-    shr     ecx, 16
-    mov     word ptr[rdi+32*7], cx
-
-    movd    eax, mm2
-    psrlq   mm2, 32
-    mov     word ptr[rdi+32*8], ax
-    shr     eax, 16
-    mov     word ptr[rdi+32*9], ax
-    movd    eax, mm2
-    mov     word ptr[rdi+32*10], ax
-    shr     eax, 16
-    mov     word ptr[rdi+32*11], ax
-
-    movd    ecx, mm3
-    psrlq   mm3, 32
-    mov     word ptr[rdi+32*12], cx
-    shr     ecx, 16
-    mov     word ptr[rdi+32*13], cx
-    movd    ecx, mm3
-    mov     word ptr[rdi+32*14], cx
-    shr     ecx, 16
-    mov     word ptr[rdi+32*15], cx
+
+    movd        eax, mm1
+    movd        ecx, mm0
+    psrlq       mm0, 32
+    psrlq       mm1, 32
+    mov         word ptr[rdx+32*0], ax
+    mov         word ptr[rdx+32*1], cx
+    shr         eax, 16
+    shr         ecx, 16
+    mov         word ptr[rdx+32*4], ax
+    mov         word ptr[rdx+32*5], cx
+    movd        eax, mm1
+    movd        ecx, mm0
+    mov         word ptr[rdx+32*8], ax
+    mov         word ptr[rdx+32*9], cx
+    shr         eax, 16
+    shr         ecx, 16
+    mov         word ptr[rdx+32*12], ax
+    mov         word ptr[rdx+32*13], cx
+
+    movd        eax, mm6
+    movd        ecx, mm5
+    psrlq       mm5, 32
+    psrlq       mm6, 32
+    mov         word ptr[rdx+32*2], ax
+    mov         word ptr[rdx+32*3], cx
+    shr         eax, 16
+    shr         ecx, 16
+    mov         word ptr[rdx+32*6], ax
+    mov         word ptr[rdx+32*7], cx
+    movd        eax, mm6
+    movd        ecx, mm5
+    mov         word ptr[rdx+32*10], ax
+    mov         word ptr[rdx+32*11], cx
+    shr         eax, 16
+    shr         ecx, 16
+    mov         word ptr[rdx+32*14], ax
+    mov         word ptr[rdx+32*15], cx
 
     ; begin epilog
-    pop rdi
-    pop rsi
     UNSHADOW_ARGS
     pop         rbp
     ret

diff --git a/vp8/common/x86/iwalsh_sse2.asm b/vp8/common/x86/iwalsh_sse2.asm
index 5a7133d..51cb5e2 100644 (file)
--- a/vp8/common/x86/iwalsh_sse2.asm
+++ b/vp8/common/x86/iwalsh_sse2.asm
@@ -17,145 +17,105 @@ sym(vp8_short_inv_walsh4x4_sse2):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 2
-    SAVE_XMM 6
-    push        rsi
-    push        rdi
     ; end prolog
 
-    mov     rsi, arg(0)
-    mov     rdi, arg(1)
-    mov     rax, 3
+    mov         rcx, arg(0)
+    mov         rdx, arg(1)
+    mov         rax, 30003h
 
-    movdqa    xmm0, [rsi + 0]       ;ip[4] ip[0]
-    movdqa    xmm1, [rsi + 16]      ;ip[12] ip[8]
+    movdqa      xmm0, [rcx + 0]     ;ip[4] ip[0]
+    movdqa      xmm1, [rcx + 16]    ;ip[12] ip[8]
 
-    shl     rax, 16
-    or      rax, 3            ;00030003h
 
-    pshufd    xmm2, xmm1, 4eh       ;ip[8] ip[12]
-    movdqa    xmm3, xmm0          ;ip[4] ip[0]
+    pshufd      xmm2, xmm1, 4eh     ;ip[8] ip[12]
+    movdqa      xmm3, xmm0          ;ip[4] ip[0]
 
-    paddw   xmm0, xmm2          ;ip[4]+ip[8] ip[0]+ip[12] aka b1 a1
-    psubw   xmm3, xmm2          ;ip[4]-ip[8] ip[0]-ip[12] aka c1 d1
+    paddw       xmm0, xmm2          ;ip[4]+ip[8] ip[0]+ip[12] aka b1 a1
+    psubw       xmm3, xmm2          ;ip[4]-ip[8] ip[0]-ip[12] aka c1 d1
 
-    movdqa    xmm4, xmm0
+    movdqa      xmm4, xmm0
     punpcklqdq  xmm0, xmm3          ;d1 a1
     punpckhqdq  xmm4, xmm3          ;c1 b1
-    movd    xmm6, eax
 
-    movdqa    xmm1, xmm4          ;c1 b1
-    paddw   xmm4, xmm0          ;dl+cl a1+b1 aka op[4] op[0]
-    psubw   xmm0, xmm1          ;d1-c1 a1-b1 aka op[12] op[8]
+    movdqa      xmm1, xmm4          ;c1 b1
+    paddw       xmm4, xmm0          ;dl+cl a1+b1 aka op[4] op[0]
+    psubw       xmm0, xmm1          ;d1-c1 a1-b1 aka op[12] op[8]
 
-;;;temp output
-;;  movdqu  [rdi + 0], xmm4
-;;  movdqu  [rdi + 16], xmm3
-
-;~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+    ;~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
     ; 13 12 11 10 03 02 01 00
     ;
     ; 33 32 31 30 23 22 21 20
     ;
-    movdqa    xmm3, xmm4          ; 13 12 11 10 03 02 01 00
-    punpcklwd xmm4, xmm0          ; 23 03 22 02 21 01 20 00
-    punpckhwd xmm3, xmm0          ; 33 13 32 12 31 11 30 10
-    movdqa    xmm1, xmm4          ; 23 03 22 02 21 01 20 00
-    punpcklwd xmm4, xmm3          ; 31 21 11 01 30 20 10 00
-    punpckhwd xmm1, xmm3          ; 33 23 13 03 32 22 12 02
+    movdqa      xmm3, xmm4          ; 13 12 11 10 03 02 01 00
+    punpcklwd   xmm4, xmm0          ; 23 03 22 02 21 01 20 00
+    punpckhwd   xmm3, xmm0          ; 33 13 32 12 31 11 30 10
+    movdqa      xmm1, xmm4          ; 23 03 22 02 21 01 20 00
+    punpcklwd   xmm4, xmm3          ; 31 21 11 01 30 20 10 00
+    punpckhwd   xmm1, xmm3          ; 33 23 13 03 32 22 12 02
     ;~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-    pshufd    xmm2, xmm1, 4eh       ;ip[8] ip[12]
-    movdqa    xmm3, xmm4          ;ip[4] ip[0]
+    movd        xmm0, eax
+    pshufd      xmm2, xmm1, 4eh     ;ip[8] ip[12]
+    movdqa      xmm3, xmm4          ;ip[4] ip[0]
 
-    pshufd    xmm6, xmm6, 0       ;03 03 03 03 03 03 03 03
+    pshufd      xmm0, xmm0, 0       ;03 03 03 03 03 03 03 03
 
-    paddw   xmm4, xmm2          ;ip[4]+ip[8] ip[0]+ip[12] aka b1 a1
-    psubw   xmm3, xmm2          ;ip[4]-ip[8] ip[0]-ip[12] aka c1 d1
+    paddw       xmm4, xmm2          ;ip[4]+ip[8] ip[0]+ip[12] aka b1 a1
+    psubw       xmm3, xmm2          ;ip[4]-ip[8] ip[0]-ip[12] aka c1 d1
 
-    movdqa    xmm5, xmm4
+    movdqa      xmm5, xmm4
     punpcklqdq  xmm4, xmm3          ;d1 a1
     punpckhqdq  xmm5, xmm3          ;c1 b1
 
-    movdqa    xmm1, xmm5          ;c1 b1
-    paddw   xmm5, xmm4          ;dl+cl a1+b1 aka op[4] op[0]
-    psubw   xmm4, xmm1          ;d1-c1 a1-b1 aka op[12] op[8]
-;~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-    ; 13 12 11 10 03 02 01 00
-    ;
-    ; 33 32 31 30 23 22 21 20
-    ;
-    movdqa    xmm0, xmm5          ; 13 12 11 10 03 02 01 00
-    punpcklwd xmm5, xmm4          ; 23 03 22 02 21 01 20 00
-    punpckhwd xmm0, xmm4          ; 33 13 32 12 31 11 30 10
-    movdqa    xmm1, xmm5          ; 23 03 22 02 21 01 20 00
-    punpcklwd xmm5, xmm0          ; 31 21 11 01 30 20 10 00
-    punpckhwd xmm1, xmm0          ; 33 23 13 03 32 22 12 02
-;~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-    paddw   xmm5, xmm6
-    paddw   xmm1, xmm6
-
-    psraw   xmm5, 3
-    psraw   xmm1, 3
-
-;;    movdqa  [rdi + 0], xmm5
-;;    movdqa  [rdi + 16], xmm1
-
-    movd    eax, xmm5
-    psrldq   xmm5, 4
-    mov     word ptr[rdi+32*0], ax
-    shr     eax, 16
-    mov     word ptr[rdi+32*1], ax
-    movd    eax, xmm5
-    psrldq   xmm5, 4
-    mov     word ptr[rdi+32*2], ax
-    shr     eax, 16
-    mov     word ptr[rdi+32*3], ax
-
-    movd    eax, xmm5
-    psrldq   xmm5, 4
-    mov     word ptr[rdi+32*4], ax
-    shr     eax, 16
-    mov     word ptr[rdi+32*5], ax
-    movd    eax, xmm5
-    mov     word ptr[rdi+32*6], ax
-    shr     eax, 16
-    mov     word ptr[rdi+32*7], ax
-
-    movd    eax, xmm1
-    psrldq   xmm1, 4
-    mov     word ptr[rdi+32*8], ax
-    shr     eax, 16
-    mov     word ptr[rdi+32*9], ax
-    movd    eax, xmm1
-    psrldq   xmm1, 4
-    mov     word ptr[rdi+32*10], ax
-    shr     eax, 16
-    mov     word ptr[rdi+32*11], ax
-
-    movd    eax, xmm1
-    psrldq   xmm1, 4
-    mov     word ptr[rdi+32*12], ax
-    shr     eax, 16
-    mov     word ptr[rdi+32*13], ax
-    movd    eax, xmm1
-    mov     word ptr[rdi+32*14], ax
-    shr     eax, 16
-    mov     word ptr[rdi+32*15], ax
+    movdqa      xmm1, xmm5          ;c1 b1
+    paddw       xmm5, xmm4          ;dl+cl a1+b1 aka op[4] op[0]
+    psubw       xmm4, xmm1          ;d1-c1 a1-b1 aka op[12] op[8]
+
+    paddw       xmm5, xmm0
+    paddw       xmm4, xmm0
+    psraw       xmm5, 3
+    psraw       xmm4, 3
+
+    movd        eax, xmm5
+    movd        ecx, xmm4
+    psrldq      xmm5, 4
+    psrldq      xmm4, 4
+    mov         word ptr[rdx+32*0], ax
+    mov         word ptr[rdx+32*2], cx
+    shr         eax, 16
+    shr         ecx, 16
+    mov         word ptr[rdx+32*4], ax
+    mov         word ptr[rdx+32*6], cx
+    movd        eax, xmm5
+    movd        ecx, xmm4
+    psrldq      xmm5, 4
+    psrldq      xmm4, 4
+    mov         word ptr[rdx+32*8], ax
+    mov         word ptr[rdx+32*10], cx
+    shr         eax, 16
+    shr         ecx, 16
+    mov         word ptr[rdx+32*12], ax
+    mov         word ptr[rdx+32*14], cx
+
+    movd        eax, xmm5
+    movd        ecx, xmm4
+    psrldq      xmm5, 4
+    psrldq      xmm4, 4
+    mov         word ptr[rdx+32*1], ax
+    mov         word ptr[rdx+32*3], cx
+    shr         eax, 16
+    shr         ecx, 16
+    mov         word ptr[rdx+32*5], ax
+    mov         word ptr[rdx+32*7], cx
+    movd        eax, xmm5
+    movd        ecx, xmm4
+    mov         word ptr[rdx+32*9], ax
+    mov         word ptr[rdx+32*11], cx
+    shr         eax, 16
+    shr         ecx, 16
+    mov         word ptr[rdx+32*13], ax
+    mov         word ptr[rdx+32*15], cx
 
     ; begin epilog
-    pop rdi
-    pop rsi
-    RESTORE_XMM
     UNSHADOW_ARGS
     pop         rbp
     ret
-
-SECTION_RODATA
-align 16
-x_s1sqr2:
-    times 4 dw 0x8A8C
-align 16
-x_c1sqr2less1:
-    times 4 dw 0x4E7B
-align 16
-fours:
-    times 4 dw 0x0004