x86inc: support stack mem allocation and re-alignment in PROLOGUE

Use this in VP8/H264-8bit loopfilter functions so they can be used if
there is no aligned stack (e.g. MSVC 32bit or ICC 10.x).

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

4 files changed, 42 insertions, 65 deletions

diff --git a/libavcodec/x86/h264_deblock.asm b/libavcodec/x86/h264_deblock.asm
index c124c4daa0..1f1dbc6a2f 100644
--- a/libavcodec/x86/h264_deblock.asm
+++ b/libavcodec/x86/h264_deblock.asm
@@ -398,14 +398,12 @@ DEBLOCK_LUMA
 ;-----------------------------------------------------------------------------
 ; void deblock_v8_luma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
 ;-----------------------------------------------------------------------------
-cglobal deblock_%1_luma_8, 5,5
+cglobal deblock_%1_luma_8, 5,5,8,2*%2
     lea     r4, [r1*3]
     dec     r2     ; alpha-1
     neg     r4
     dec     r3     ; beta-1
     add     r4, r0 ; pix-3*stride
-    %assign pad 2*%2+12-(stack_offset&15)
-    SUB     esp, pad
 
     mova    m0, [r4+r1]   ; p1
     mova    m1, [r4+2*r1] ; p0
@@ -443,22 +441,19 @@ cglobal deblock_%1_luma_8, 5,5
     DEBLOCK_P0_Q0
     mova    [r4+2*r1], m1
     mova    [r0], m2
-    ADD     esp, pad
     RET
 
 ;-----------------------------------------------------------------------------
 ; void deblock_h_luma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
 ;-----------------------------------------------------------------------------
 INIT_MMX cpuname
-cglobal deblock_h_luma_8, 0,5
+cglobal deblock_h_luma_8, 0,5,8,0x60+HAVE_ALIGNED_STACK*12
     mov    r0, r0mp
     mov    r3, r1m
     lea    r4, [r3*3]
     sub    r0, 4
     lea    r1, [r0+r4]
-    %assign pad 0x78-(stack_offset&15)
-    SUB    esp, pad
-%define pix_tmp esp+12
+%define pix_tmp esp+12*HAVE_ALIGNED_STACK
 
     ; transpose 6x16 -> tmp space
     TRANSPOSE6x8_MEM  PASS8ROWS(r0, r1, r3, r4), pix_tmp
@@ -500,7 +495,6 @@ cglobal deblock_h_luma_8, 0,5
     movq   m3, [pix_tmp+0x48]
     TRANSPOSE8x4B_STORE  PASS8ROWS(r0, r1, r3, r4)
 
-    ADD    esp, pad
     RET
 %endmacro ; DEBLOCK_LUMA
 
@@ -631,7 +625,7 @@ DEBLOCK_LUMA v, 16
     %define mpb_0 m14
     %define mpb_1 m15
 %else
-    %define spill(x) [esp+16*x+((stack_offset+4)&15)]
+    %define spill(x) [esp+16*x]
     %define p2 [r4+r1]
     %define q2 [r0+2*r1]
     %define t4 spill(0)
@@ -646,10 +640,7 @@ DEBLOCK_LUMA v, 16
 ;-----------------------------------------------------------------------------
 ; void deblock_v_luma_intra( uint8_t *pix, int stride, int alpha, int beta )
 ;-----------------------------------------------------------------------------
-cglobal deblock_%1_luma_intra_8, 4,6,16
-%if ARCH_X86_64 == 0
-    sub     esp, 0x60
-%endif
+cglobal deblock_%1_luma_intra_8, 4,6,16,ARCH_X86_64*0x50-0x50
     lea     r4, [r1*4]
     lea     r5, [r1*3] ; 3*stride
     dec     r2d        ; alpha-1
@@ -698,9 +689,6 @@ cglobal deblock_%1_luma_intra_8, 4,6,16
     LUMA_INTRA_SWAP_PQ
     LUMA_INTRA_P012 [r0], [r0+r1], [r0+2*r1], [r0+r5]
 .end:
-%if ARCH_X86_64 == 0
-    add     esp, 0x60
-%endif
     RET
 
 INIT_MMX cpuname
@@ -737,12 +725,10 @@ cglobal deblock_h_luma_intra_8, 4,9
     add    rsp, 0x88
     RET
 %else
-cglobal deblock_h_luma_intra_8, 2,4
+cglobal deblock_h_luma_intra_8, 2,4,8,0x80
     lea    r3,  [r1*3]
     sub    r0,  4
     lea    r2,  [r0+r3]
-%assign pad 0x8c-(stack_offset&15)
-    SUB    rsp, pad
     %define pix_tmp rsp
 
     ; transpose 8x16 -> tmp space
@@ -773,7 +759,6 @@ cglobal deblock_h_luma_intra_8, 2,4
     lea    r0,  [r0+r1*8]
     lea    r2,  [r2+r1*8]
     TRANSPOSE8x8_MEM  PASS8ROWS(pix_tmp+8, pix_tmp+0x38, 0x10, 0x30), PASS8ROWS(r0, r2, r1, r3)
-    ADD    rsp, pad
     RET
 %endif ; ARCH_X86_64
 %endmacro ; DEBLOCK_LUMA_INTRA
diff --git a/libavcodec/x86/h264dsp_init.c b/libavcodec/x86/h264dsp_init.c
index ac231cbb08..73d4990899 100644
--- a/libavcodec/x86/h264dsp_init.c
+++ b/libavcodec/x86/h264dsp_init.c
@@ -275,18 +275,16 @@ void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth,
                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_sse2;
                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_sse2;
 
-#if HAVE_ALIGNED_STACK
                     c->h264_v_loop_filter_luma       = ff_deblock_v_luma_8_sse2;
                     c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_sse2;
                     c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_sse2;
                     c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_sse2;
-#endif /* HAVE_ALIGNED_STACK */
                 }
                 if (EXTERNAL_SSSE3(mm_flags)) {
                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_ssse3;
                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_ssse3;
                 }
-                if (EXTERNAL_AVX(mm_flags) && HAVE_ALIGNED_STACK) {
+                if (EXTERNAL_AVX(mm_flags)) {
                     c->h264_v_loop_filter_luma       = ff_deblock_v_luma_8_avx;
                     c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_avx;
                     c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_avx;
diff --git a/libavcodec/x86/vp8dsp.asm b/libavcodec/x86/vp8dsp.asm
index d732bf45eb..af8403e991 100644
--- a/libavcodec/x86/vp8dsp.asm
+++ b/libavcodec/x86/vp8dsp.asm
@@ -1631,28 +1631,31 @@ SIMPLE_LOOPFILTER h, 5
 ;-----------------------------------------------------------------------------
 
 %macro INNER_LOOPFILTER 2
+%define stack_size 0
+%ifndef m8   ; stack layout: [0]=E, [1]=I, [2]=hev_thr
+%ifidn %1, v ;               [3]=hev() result
+%define stack_size mmsize * -4
+%else ; h    ; extra storage space for transposes
+%define stack_size mmsize * -5
+%endif
+%endif
+
 %if %2 == 8 ; chroma
-cglobal vp8_%1_loop_filter8uv_inner, 6, 6, 13, dst, dst8, stride, flimE, flimI, hevthr
+cglobal vp8_%1_loop_filter8uv_inner, 6, 6, 13, stack_size, dst, dst8, stride, flimE, flimI, hevthr
 %else ; luma
-cglobal vp8_%1_loop_filter16y_inner, 5, 5, 13, dst, stride, flimE, flimI, hevthr
+cglobal vp8_%1_loop_filter16y_inner, 5, 5, 13, stack_size, dst, stride, flimE, flimI, hevthr
 %endif
 
 %if cpuflag(ssse3)
     pxor             m7, m7
 %endif
-%ifndef m8   ; stack layout: [0]=E, [1]=I, [2]=hev_thr
-%ifidn %1, v ;               [3]=hev() result
-%assign pad 16 + mmsize * 4 - gprsize - (stack_offset & 15)
-%else ; h    ; extra storage space for transposes
-%assign pad 16 + mmsize * 5 - gprsize - (stack_offset & 15)
-%endif
+
+%ifndef m8
     ; splat function arguments
     SPLATB_REG       m0, flimEq, m7   ; E
     SPLATB_REG       m1, flimIq, m7   ; I
     SPLATB_REG       m2, hevthrq, m7  ; hev_thresh
 
-    SUB             rsp, pad
-
 %define m_flimE    [rsp]
 %define m_flimI    [rsp+mmsize]
 %define m_hevthr   [rsp+mmsize*2]
@@ -2082,12 +2085,10 @@ cglobal vp8_%1_loop_filter16y_inner, 5, 5, 13, dst, stride, flimE, flimI, hevthr
     dec           cntrq
     jg .next8px
 %endif
-%endif
-
-%ifndef m8 ; sse2 on x86-32 or mmx/mmxext
-    ADD             rsp, pad
-%endif
+    REP_RET
+%else ; mmsize == 16
     RET
+%endif
 %endmacro
 
 %if ARCH_X86_32
@@ -2122,31 +2123,34 @@ INNER_LOOPFILTER h,  8
 ;-----------------------------------------------------------------------------
 
 %macro MBEDGE_LOOPFILTER 2
-%if %2 == 8 ; chroma
-cglobal vp8_%1_loop_filter8uv_mbedge, 6, 6, 15, dst1, dst8, stride, flimE, flimI, hevthr
-%else ; luma
-cglobal vp8_%1_loop_filter16y_mbedge, 5, 5, 15, dst1, stride, flimE, flimI, hevthr
-%endif
-
-%if cpuflag(ssse3)
-    pxor             m7, m7
-%endif
+%define stack_size 0
 %ifndef m8       ; stack layout: [0]=E, [1]=I, [2]=hev_thr
 %if mmsize == 16 ;               [3]=hev() result
                  ;               [4]=filter tmp result
                  ;               [5]/[6] = p2/q2 backup
                  ;               [7]=lim_res sign result
-%assign pad 16 + mmsize * 7 - gprsize - (stack_offset & 15)
+%define stack_size mmsize * -7
 %else ; 8        ; extra storage space for transposes
-%assign pad 16 + mmsize * 8 - gprsize - (stack_offset & 15)
+%define stack_size mmsize * -8
+%endif
 %endif
+
+%if %2 == 8 ; chroma
+cglobal vp8_%1_loop_filter8uv_mbedge, 6, 6, 15, stack_size, dst1, dst8, stride, flimE, flimI, hevthr
+%else ; luma
+cglobal vp8_%1_loop_filter16y_mbedge, 5, 5, 15, stack_size, dst1, stride, flimE, flimI, hevthr
+%endif
+
+%if cpuflag(ssse3)
+    pxor             m7, m7
+%endif
+
+%ifndef m8
     ; splat function arguments
     SPLATB_REG       m0, flimEq, m7   ; E
     SPLATB_REG       m1, flimIq, m7   ; I
     SPLATB_REG       m2, hevthrq, m7  ; hev_thresh
 
-    SUB             rsp, pad
-
 %define m_flimE    [rsp]
 %define m_flimI    [rsp+mmsize]
 %define m_hevthr   [rsp+mmsize*2]
@@ -2740,12 +2744,10 @@ cglobal vp8_%1_loop_filter16y_mbedge, 5, 5, 15, dst1, stride, flimE, flimI, hevt
     dec          cntrq
     jg .next8px
 %endif
-%endif
-
-%ifndef m8 ; sse2 on x86-32 or mmx/mmxext
-    ADD            rsp, pad
-%endif
+    REP_RET
+%else ; mmsize == 16
     RET
+%endif
 %endmacro
 
 %if ARCH_X86_32
diff --git a/libavcodec/x86/vp8dsp_init.c b/libavcodec/x86/vp8dsp_init.c
index bf5463f080..a9d3541807 100644
--- a/libavcodec/x86/vp8dsp_init.c
+++ b/libavcodec/x86/vp8dsp_init.c
@@ -390,13 +390,11 @@ av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c)
 
         c->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_sse2;
 
-#if ARCH_X86_64 || HAVE_ALIGNED_STACK
         c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_sse2;
         c->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_sse2;
 
         c->vp8_v_loop_filter16y       = ff_vp8_v_loop_filter16y_mbedge_sse2;
         c->vp8_v_loop_filter8uv       = ff_vp8_v_loop_filter8uv_mbedge_sse2;
-#endif
     }
 
     if (mm_flags & AV_CPU_FLAG_SSE2) {
@@ -404,13 +402,11 @@ av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c)
 
         c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_sse2;
 
-#if ARCH_X86_64 || HAVE_ALIGNED_STACK
         c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_sse2;
         c->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_sse2;
 
         c->vp8_h_loop_filter16y       = ff_vp8_h_loop_filter16y_mbedge_sse2;
         c->vp8_h_loop_filter8uv       = ff_vp8_h_loop_filter8uv_mbedge_sse2;
-#endif
     }
 
     if (mm_flags & AV_CPU_FLAG_SSSE3) {
@@ -424,7 +420,6 @@ av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c)
         c->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_ssse3;
         c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_ssse3;
 
-#if ARCH_X86_64 || HAVE_ALIGNED_STACK
         c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_ssse3;
         c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_ssse3;
         c->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_ssse3;
@@ -434,17 +429,14 @@ av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c)
         c->vp8_h_loop_filter16y       = ff_vp8_h_loop_filter16y_mbedge_ssse3;
         c->vp8_v_loop_filter8uv       = ff_vp8_v_loop_filter8uv_mbedge_ssse3;
         c->vp8_h_loop_filter8uv       = ff_vp8_h_loop_filter8uv_mbedge_ssse3;
-#endif
     }
 
     if (mm_flags & AV_CPU_FLAG_SSE4) {
         c->vp8_idct_dc_add                  = ff_vp8_idct_dc_add_sse4;
 
         c->vp8_h_loop_filter_simple   = ff_vp8_h_loop_filter_simple_sse4;
-#if ARCH_X86_64 || HAVE_ALIGNED_STACK
         c->vp8_h_loop_filter16y       = ff_vp8_h_loop_filter16y_mbedge_sse4;
         c->vp8_h_loop_filter8uv       = ff_vp8_h_loop_filter8uv_mbedge_sse4;
-#endif
     }
 #endif /* HAVE_YASM */
 }