some simplifications to mmx intra pred that should have been done way back when we...

and remove pic spills in functions that have a free caller-saved reg.
patch partly by Fiona Glaser.

git-svn-id: svn://svn.videolan.org/x264/trunk@741 df754926-b1dd-0310-bc7b-ec298dee348c

diff --git a/common/amd64/predict-a.asm b/common/amd64/predict-a.asm
index 91e6b8bccba5dc6b8ea9b95951f1e13954316f18..d2cce53a9b2970fdc166db1cdb3ae7866935ddd4 100644 (file)
--- a/common/amd64/predict-a.asm
+++ b/common/amd64/predict-a.asm
@@ -41,19 +41,18 @@ BITS 64
 
 %macro STORE16x16 2
     mov         eax, 4
-ALIGN 4
 .loop:
+    movq        [parm1q + 0*FDEC_STRIDE], %1
     movq        [parm1q + 1*FDEC_STRIDE], %1
     movq        [parm1q + 2*FDEC_STRIDE], %1
     movq        [parm1q + 3*FDEC_STRIDE], %1
-    movq        [parm1q + 4*FDEC_STRIDE], %1
+    movq        [parm1q + 0*FDEC_STRIDE + 8], %2
     movq        [parm1q + 1*FDEC_STRIDE + 8], %2
     movq        [parm1q + 2*FDEC_STRIDE + 8], %2
     movq        [parm1q + 3*FDEC_STRIDE + 8], %2
-    movq        [parm1q + 4*FDEC_STRIDE + 8], %2
+    add         parm1q, 4*FDEC_STRIDE
     dec         eax
-    lea         parm1q, [parm1q + 4*FDEC_STRIDE]
-    jnz         .loop
+    jg          .loop
     nop
 %endmacro
 
@@ -466,9 +465,8 @@ ALIGN 4
 ; void predict_16x16_v_mmx( uint8_t *src )
 ;-----------------------------------------------------------------------------
 cglobal predict_16x16_v_mmx
-    sub         parm1q, FDEC_STRIDE
-    movq        mm0, [parm1q]
-    movq        mm1, [parm1q + 8]
+    movq        mm0, [parm1q - FDEC_STRIDE]
+    movq        mm1, [parm1q - FDEC_STRIDE + 8]
     STORE16x16  mm0, mm1
     ret
 
@@ -477,18 +475,15 @@ cglobal predict_16x16_v_mmx
 ;-----------------------------------------------------------------------------
 
 %macro PRED16x16_DC 2
-    sub         parm1q, FDEC_STRIDE
-
     pxor        mm0, mm0
     pxor        mm1, mm1
-    psadbw      mm0, [parm1q]
-    psadbw      mm1, [parm1q + 8]
+    psadbw      mm0, [parm1q - FDEC_STRIDE]
+    psadbw      mm1, [parm1q - FDEC_STRIDE + 8]
     paddusw     mm0, mm1
     paddusw     mm0, %1
     psrlw       mm0, %2                       ; dc
     pshufw      mm0, mm0, 0
     packuswb    mm0, mm0                      ; dc in bytes
-
     STORE16x16  mm0, mm0
 %endmacro
 

diff --git a/common/i386/dct-a.asm b/common/i386/dct-a.asm
index 058b477c4816c51466b7aade4862e921d327f887..535327564b5a17290667ea7b15fa088247732542 100644 (file)
--- a/common/i386/dct-a.asm
+++ b/common/i386/dct-a.asm
@@ -141,8 +141,7 @@ cglobal x264_dct4x4dc_mmx
     movq    mm2,        [eax+16]
     movq    mm3,        [eax+24]
 
-    picpush ebx
-    picgetgot ebx
+    picgetgot edx
 
     MMX_SUMSUB_BADC     mm1, mm0, mm3, mm2          ; mm1=s01  mm0=d01  mm3=s23  mm2=d23
     MMX_SUMSUB_BADC     mm3, mm1, mm2, mm0          ; mm3=s01+s23  mm1=s01-s23  mm2=d01+d23  mm0=d01-d23
@@ -152,7 +151,7 @@ cglobal x264_dct4x4dc_mmx
     MMX_SUMSUB_BADC     mm2, mm3, mm0, mm4          ; mm2=s01  mm3=d01  mm0=s23  mm4=d23
     MMX_SUMSUB_BADC     mm0, mm2, mm4, mm3          ; mm0=s01+s23  mm2=s01-s23  mm4=d01+d23  mm3=d01-d23
 
-    movq    mm6,        [pw_1 GOT_ebx]
+    movq    mm6,        [pw_1 GOT_edx]
     paddw   mm0,        mm6
     paddw   mm2,        mm6
     psraw   mm0,        1
@@ -165,7 +164,6 @@ cglobal x264_dct4x4dc_mmx
     movq    [eax+16],   mm3
     psraw   mm4,        1
     movq    [eax+24],   mm4
-    picpop  ebx
     ret
 
 ;-----------------------------------------------------------------------------
@@ -241,8 +239,7 @@ cglobal x264_add4x4_idct_mmx
     
     mov     eax, [esp+ 4]   ; p_dst
 
-    picpush ebx
-    picgetgot ebx
+    picgetgot edx
 
     MMX_SUMSUB_BA       mm2, mm0                        ; mm2=s02  mm0=d02
     MMX_SUMSUBD2_AB     mm1, mm3, mm5, mm4              ; mm1=s13  mm4=d13 ( well 1 + 3>>1 and 1>>1 + 3)
@@ -258,14 +255,13 @@ cglobal x264_add4x4_idct_mmx
     MMX_SUMSUB_BADC     mm2, mm3, mm4, mm1              ; mm2=s02+s13  mm3=s02-s13  mm4=d02+d13  mm1=d02-d13
 
     MMX_ZERO            mm7
-    movq                mm6, [pw_32 GOT_ebx]
+    movq                mm6, [pw_32 GOT_edx]
     
     MMX_STORE_DIFF_4P   mm2, mm0, mm6, mm7, [eax+0*FDEC_STRIDE]
     MMX_STORE_DIFF_4P   mm4, mm0, mm6, mm7, [eax+1*FDEC_STRIDE]
     MMX_STORE_DIFF_4P   mm1, mm0, mm6, mm7, [eax+2*FDEC_STRIDE]
     MMX_STORE_DIFF_4P   mm3, mm0, mm6, mm7, [eax+3*FDEC_STRIDE]
 
-    picpop  ebx
     ret
 
 

diff --git a/common/i386/mc-a.asm b/common/i386/mc-a.asm
index 991b09665042197386d4244b1b208a73bb54d634..2622b64337052a7df3a3f17795ab56c254a71286 100644 (file)
--- a/common/i386/mc-a.asm
+++ b/common/i386/mc-a.asm
@@ -253,21 +253,19 @@ AVG2_END
 %macro BIWEIGHT_START_MMX 0
     push    edi
     push    esi
-    picpush ebx
-    picgetgot ebx
-    mov     edi, [picesp+12]    ; dst
-    mov     esi, [picesp+16]    ; i_dst
-    mov     edx, [picesp+20]    ; src
-    mov     ecx, [picesp+24]    ; i_src
-
-    pshufw  mm4, [picesp+28], 0  ; weight_dst
-    movq    mm5, [pw_64 GOT_ebx]
-    psubw   mm5, mm4             ; weight_src
-    movq    mm6, [pw_32 GOT_ebx] ; rounding
+    picgetgot ecx
+    movq    mm5, [pw_64 GOT_ecx]
+    movq    mm6, [pw_32 GOT_ecx] ; rounding
+    mov     edi, [esp+12]        ; dst
+    mov     esi, [esp+16]        ; i_dst
+    mov     edx, [esp+20]        ; src
+    mov     ecx, [esp+24]        ; i_src
+    pshufw  mm4, [esp+28], 0     ; weight_dst
     pxor    mm7, mm7
+    psubw   mm5, mm4             ; weight_src
 %endmacro
+
 %macro BIWEIGHT_END_MMX 0
-    picpop  ebx
     pop     esi
     pop     edi
     ret
@@ -278,7 +276,7 @@ AVG2_END
 ;-----------------------------------------------------------------------------
 cglobal x264_pixel_avg_weight_w16_mmxext
     BIWEIGHT_START_MMX
-    mov     eax, [picesp+32] ; i_height
+    mov     eax, [esp+32] ; i_height
     ALIGN 4
     .height_loop
 
@@ -298,7 +296,7 @@ cglobal x264_pixel_avg_weight_w16_mmxext
 ;-----------------------------------------------------------------------------
 cglobal x264_pixel_avg_weight_w8_mmxext
     BIWEIGHT_START_MMX
-    mov     eax, [picesp+32]
+    mov     eax, [esp+32]
     ALIGN 4
     .height_loop
 

diff --git a/common/i386/predict-a.asm b/common/i386/predict-a.asm
index 01156cc8d4d233b13904feedd6170d64c6f49591..baafd68e4a778765023d0374170a9bc555dc0d29 100644 (file)
--- a/common/i386/predict-a.asm
+++ b/common/i386/predict-a.asm
@@ -39,17 +39,23 @@ BITS 32
     movq        [edx + 7*FDEC_STRIDE], %2
 %endmacro
 
-%macro SAVE_0_1 1
-    movq        [%1]         , mm0
-    movq        [%1 + 8]     , mm1
-%endmacro
-
-%macro SAVE_0_0 1
-    movq        [%1]         , mm0
-    movq        [%1 + 8]     , mm0
+%macro STORE16x16 2
+    mov         eax, 4
+.loop:
+    movq        [edx + 0*FDEC_STRIDE], %1
+    movq        [edx + 1*FDEC_STRIDE], %1
+    movq        [edx + 2*FDEC_STRIDE], %1
+    movq        [edx + 3*FDEC_STRIDE], %1
+    movq        [edx + 0*FDEC_STRIDE + 8], %2
+    movq        [edx + 1*FDEC_STRIDE + 8], %2
+    movq        [edx + 2*FDEC_STRIDE + 8], %2
+    movq        [edx + 3*FDEC_STRIDE + 8], %2
+    add         edx, 4*FDEC_STRIDE
+    dec         eax
+    jg          .loop
+    nop
 %endmacro
 
-
 SECTION_RODATA
 
 ALIGN 8
@@ -76,7 +82,7 @@ SECTION .text
     pavgb       %2, %3
     pxor        %3, %5
     movq        %1, %4
-    pand        %3, [pb_1 GOT_ebx]
+    pand        %3, [pb_1 GOT_ecx]
     psubusb     %2, %3
     pavgb       %1, %2
 %endmacro
@@ -96,21 +102,19 @@ cglobal predict_8x8_v_mmxext
 ; void predict_8x8_dc_mmxext( uint8_t *src, uint8_t *edge )
 ;-----------------------------------------------------------------------------
 cglobal predict_8x8_dc_mmxext
-    picpush     ebx
-    picgetgot   ebx
-    mov         eax, [picesp + 8]
-    mov         edx, [picesp + 4]
+    picgetgot   ecx
+    mov         eax, [esp + 8]
+    mov         edx, [esp + 4]
     pxor        mm0, mm0
     pxor        mm1, mm1
     psadbw      mm0, [eax+7]
     psadbw      mm1, [eax+16]
-    paddw       mm0, [pw_8 GOT_ebx]
+    paddw       mm0, [pw_8 GOT_ecx]
     paddw       mm0, mm1
     psrlw       mm0, 4
     pshufw      mm0, mm0, 0
     packuswb    mm0, mm0
     STORE8x8    mm0, mm0
-    picpop      ebx
     ret
 
 ;-----------------------------------------------------------------------------
@@ -118,18 +122,16 @@ cglobal predict_8x8_dc_mmxext
 ;-----------------------------------------------------------------------------
 %macro PRED8x8_DC 2
 cglobal %1
-    picpush     ebx
-    picgetgot   ebx
-    mov         eax, [picesp + 8]
-    mov         edx, [picesp + 4]
+    picgetgot   ecx
+    mov         eax, [esp + 8]
+    mov         edx, [esp + 4]
     pxor        mm0, mm0
     psadbw      mm0, [eax+%2]
-    paddw       mm0, [pw_4 GOT_ebx]
+    paddw       mm0, [pw_4 GOT_ecx]
     psrlw       mm0, 3
     pshufw      mm0, mm0, 0
     packuswb    mm0, mm0
     STORE8x8    mm0, mm0
-    picpop      ebx
     ret
 %endmacro
 
@@ -140,10 +142,9 @@ PRED8x8_DC predict_8x8_dc_left_mmxext, 7
 ; void predict_8x8_ddl_mmxext( uint8_t *src, uint8_t *edge )
 ;-----------------------------------------------------------------------------
 cglobal predict_8x8_ddl_mmxext
-    picpush     ebx
-    picgetgot   ebx
-    mov         eax, [picesp + 8]
-    mov         edx, [picesp + 4]
+    picgetgot   ecx
+    mov         eax, [esp + 8]
+    mov         edx, [esp + 4]
     movq        mm1, [eax + 15]
     movq        mm2, [eax + 17]
     movq        mm3, [eax + 23]
@@ -168,17 +169,15 @@ cglobal predict_8x8_ddl_mmxext
 %assign Y (Y-1)
     movq        [edx + Y*FDEC_STRIDE], mm1
 
-    picpop      ebx
     ret
 
 ;-----------------------------------------------------------------------------
 ; void predict_8x8_ddr_mmxext( uint8_t *src, uint8_t *edge )
 ;-----------------------------------------------------------------------------
 cglobal predict_8x8_ddr_mmxext
-    picpush     ebx
-    picgetgot   ebx
-    mov         eax, [picesp + 8]
-    mov         edx, [picesp + 4]
+    picgetgot   ecx
+    mov         eax, [esp + 8]
+    mov         edx, [esp + 4]
     movq        mm1, [eax + 7]
     movq        mm2, [eax + 9]
     movq        mm3, [eax + 15]
@@ -203,7 +202,6 @@ cglobal predict_8x8_ddr_mmxext
 %assign Y (Y-1)
     movq        [edx + Y*FDEC_STRIDE], mm0
 
-    picpop      ebx
     ret
 
 ;-----------------------------------------------------------------------------
@@ -222,10 +220,9 @@ cglobal predict_8x8_ddr_mmxext
 ; 7   ,,,,,
 
 cglobal predict_8x8_vr_core_mmxext
-    picpush     ebx
-    picgetgot   ebx
-    mov         eax, [picesp + 8]
-    mov         edx, [picesp + 4]
+    picgetgot   ecx
+    mov         eax, [esp + 8]
+    mov         edx, [esp + 4]
     movq        mm2, [eax + 16]
     movq        mm3, [eax + 15]
     movq        mm1, [eax + 14]
@@ -244,13 +241,12 @@ cglobal predict_8x8_vr_core_mmxext
     movq        [edx +  Y   *FDEC_STRIDE], mm3
     movq        [edx + (Y+1)*FDEC_STRIDE], mm0
 
-    picpop      ebx
     ret
 
 ;-----------------------------------------------------------------------------
 ; void predict_8x8c_v_mmx( uint8_t *src )
 ;-----------------------------------------------------------------------------
-cglobal predict_8x8c_v_mmx 
+cglobal predict_8x8c_v_mmx
     mov         edx, [esp + 4]
     movq        mm0, [edx - FDEC_STRIDE]
     STORE8x8    mm0, mm0
@@ -260,10 +256,9 @@ cglobal predict_8x8c_v_mmx
 ; void predict_8x8c_dc_core_mmxext( uint8_t *src, int s2, int s3 )
 ;-----------------------------------------------------------------------------
 cglobal predict_8x8c_dc_core_mmxext
-    picpush     ebx
-    picgetgot   ebx
+    picgetgot   ecx
 
-    mov         edx, [picesp + 4]
+    mov         edx, [esp + 4]
 
     movq        mm0, [edx - FDEC_STRIDE]
     pxor        mm1, mm1
@@ -273,10 +268,10 @@ cglobal predict_8x8c_dc_core_mmxext
     psadbw      mm1, mm2        ; s1
     psadbw      mm0, mm2        ; s0
 
-    paddw       mm0, [picesp +  8]
-    pshufw      mm2, [picesp + 12], 0
+    paddw       mm0, [esp +  8]
+    pshufw      mm2, [esp + 12], 0
     psrlw       mm0, 3
-    paddw       mm1, [pw_2 GOT_ebx]
+    paddw       mm1, [pw_2 GOT_ecx]
     movq        mm3, mm2
     pshufw      mm1, mm1, 0
     pshufw      mm0, mm0, 0     ; dc0 (w)
@@ -289,24 +284,20 @@ cglobal predict_8x8c_dc_core_mmxext
     packuswb    mm2, mm3        ; dc2,dc3 (b)
 
     STORE8x8    mm0, mm2
-
-    picpop      ebx
     ret
 
 ;-----------------------------------------------------------------------------
 ; void predict_8x8c_p_core_mmxext( uint8_t *src, int i00, int b, int c )
 ;-----------------------------------------------------------------------------
 cglobal predict_8x8c_p_core_mmxext
-    picpush     ebx
-    picgetgot   ebx
-
-    mov         edx, [picesp + 4]
-    mov         ecx, FDEC_STRIDE
-    pshufw      mm0, [picesp + 8], 0
-    pshufw      mm2, [picesp +12], 0
-    pshufw      mm4, [picesp +16], 0
+    picgetgot   ecx
+
+    mov         edx, [esp + 4]
+    pshufw      mm0, [esp + 8], 0
+    pshufw      mm2, [esp +12], 0
+    pshufw      mm4, [esp +16], 0
     movq        mm1, mm2
-    pmullw      mm2, [pw_3210 GOT_ebx]
+    pmullw      mm2, [pw_3210 GOT_ecx]
     psllw       mm1, 2
     paddsw      mm0, mm2        ; mm0 = {i+0*b, i+1*b, i+2*b, i+3*b}
     paddsw      mm1, mm0        ; mm1 = {i+4*b, i+5*b, i+6*b, i+7*b}
@@ -323,29 +314,26 @@ ALIGN 4
 
     paddsw      mm0, mm4
     paddsw      mm1, mm4
-    add         edx, ecx
+    add         edx, FDEC_STRIDE
     dec         eax
     jg          .loop
 
     nop
-    picpop      ebx
     ret
 
 ;-----------------------------------------------------------------------------
 ; void predict_16x16_p_core_mmxext( uint8_t *src, int i00, int b, int c )
 ;-----------------------------------------------------------------------------
 cglobal predict_16x16_p_core_mmxext
-    picpush     ebx
-    picgetgot   ebx
-
-    mov         edx, [picesp + 4]
-    mov         ecx, FDEC_STRIDE
-    pshufw      mm0, [picesp + 8], 0
-    pshufw      mm2, [picesp +12], 0
-    pshufw      mm4, [picesp +16], 0
+    picgetgot   ecx
+
+    mov         edx, [esp + 4]
+    pshufw      mm0, [esp + 8], 0
+    pshufw      mm2, [esp +12], 0
+    pshufw      mm4, [esp +16], 0
     movq        mm5, mm2
     movq        mm1, mm2
-    pmullw      mm5, [pw_3210 GOT_ebx]
+    pmullw      mm5, [pw_3210 GOT_ecx]
     psllw       mm2, 3
     psllw       mm1, 2
     movq        mm3, mm2
@@ -375,91 +363,47 @@ ALIGN 4
     paddsw      mm1, mm4
     paddsw      mm2, mm4
     paddsw      mm3, mm4
-    add         edx, ecx
+    add         edx, FDEC_STRIDE
     dec         eax
     jg          .loop
 
     nop
-    picpop      ebx
     ret
 
 ;-----------------------------------------------------------------------------
 ; void predict_16x16_v_mmx( uint8_t *src )
 ;-----------------------------------------------------------------------------
-cglobal predict_16x16_v_mmx 
+cglobal predict_16x16_v_mmx
     mov         edx, [esp + 4]
-    mov         ecx, FDEC_STRIDE
-    sub         edx, ecx                ; edx <-- line -1
-
-    movq        mm0, [edx]
-    movq        mm1, [edx + 8]
-    lea         eax, [ecx + 2*ecx]      ; eax <-- 3* stride
-
-    SAVE_0_1    (edx + ecx)             ; 0
-    SAVE_0_1    (edx + 2 * ecx)         ; 1
-    SAVE_0_1    (edx + eax)             ; 2
-    SAVE_0_1    (edx + 4 * ecx)         ; 3
-    SAVE_0_1    (edx + 2 * eax)         ; 5
-    SAVE_0_1    (edx + 8 * ecx)         ; 7
-    SAVE_0_1    (edx + 4 * eax)         ; 11
-    add         edx, ecx                ; edx <-- line 0
-    SAVE_0_1    (edx + 4 * ecx)         ; 4
-    SAVE_0_1    (edx + 2 * eax)         ; 6
-    SAVE_0_1    (edx + 8 * ecx)         ; 8
-    SAVE_0_1    (edx + 4 * eax)         ; 12
-    lea         edx, [edx + 8 * ecx]    ; edx <-- line 8
-    SAVE_0_1    (edx + ecx)             ; 9
-    SAVE_0_1    (edx + 2 * ecx)         ; 10
-    lea         edx, [edx + 4 * ecx]    ; edx <-- line 12
-    SAVE_0_1    (edx + ecx)             ; 13
-    SAVE_0_1    (edx + 2 * ecx)         ; 14
-    SAVE_0_1    (edx + eax)             ; 15
-
+    movq        mm0, [edx - FDEC_STRIDE]
+    movq        mm1, [edx + 8 - FDEC_STRIDE]
+    STORE16x16  mm0, mm1
     ret
 
 ;-----------------------------------------------------------------------------
 ; void predict_16x16_dc_core_mmxext( uint8_t *src, int i_dc_left )
 ;-----------------------------------------------------------------------------
 
-%macro PRED16x16_DC 3
-    mov         edx, [%3 + 4]
-    mov         ecx, FDEC_STRIDE
-    sub         edx, ecx                ; edx <-- line -1
-
+%macro PRED16x16_DC 2
+    mov         edx, [esp+4]
     pxor        mm0, mm0
     pxor        mm1, mm1
-    psadbw      mm0, [edx]
-    psadbw      mm1, [edx + 8]
+    psadbw      mm0, [edx - FDEC_STRIDE]
+    psadbw      mm1, [edx - FDEC_STRIDE + 8]
     paddusw     mm0, mm1
-    paddusw     mm0, %1                 ; FIXME is stack alignment guaranteed?
+    paddusw     mm0, %1
     psrlw       mm0, %2                 ; dc
-    push        edi
     pshufw      mm0, mm0, 0
-    lea         eax, [ecx + 2*ecx]      ; eax <-- 3* stride
     packuswb    mm0, mm0                ; dc in bytes
-
-    mov         edi, 4
-ALIGN 4
-.loop:
-    SAVE_0_0    (edx + ecx)             ; 0
-    SAVE_0_0    (edx + 2 * ecx)         ; 1
-    SAVE_0_0    (edx + eax)             ; 2
-    SAVE_0_0    (edx + 4 * ecx)         ; 3
-    dec         edi
-    lea         edx, [edx + 4 * ecx]
-    jg          .loop
-
-    pop         edi
+    STORE16x16  mm0, mm0
 %endmacro
 
 cglobal predict_16x16_dc_core_mmxext
-    PRED16x16_DC [esp+8], 5, esp
+    PRED16x16_DC [esp+8], 5
     ret
 
 cglobal predict_16x16_dc_top_mmxext
-    picpush ebx
-    picgetgot ebx
-    PRED16x16_DC [pw_8 GOT_ebx], 4, picesp
-    picpop ebx
+    picgetgot ecx
+    PRED16x16_DC [pw_8 GOT_ecx], 4
     ret
 

diff --git a/common/i386/quant-a.asm b/common/i386/quant-a.asm
index 40f9df9373cabc52c66a3a8756c2a3dede68c897..dc7611ae080450650fff39c08c9484314ce09a8d 100644 (file)
--- a/common/i386/quant-a.asm
+++ b/common/i386/quant-a.asm
@@ -275,11 +275,9 @@ cglobal %1
 
 .rshift32:
     neg   eax
-    picpush ebx
-    picgetgot ebx
-    movq  mm6, [pd_1 GOT_ebx]
-    picpop ebx
     movd  mm5, eax
+    picgetgot eax
+    movq  mm6, [pd_1 GOT_eax]
     pxor  mm7, mm7
     pslld mm6, mm5
     psrld mm6, 1