swscale/x86/swscale: Remove obsolete and harmful MMX(EXT) functions

x64 always has MMX, MMXEXT, SSE and SSE2 and this means
that some functions for MMX, MMXEXT, SSE and 3dnow are always
overridden by other functions (unless one e.g. explicitly
disables SSE2). So given that the only systems that
benefit from these functions are truely ancient 32bit x86s
they are removed.

Moreover, some of the removed code was buggy/not bitexact
and lead to failures involving the f32le and f32be versions of
gray, gbrp and gbrap on x86-32 when SSE2 was not disabled.
See e.g.
https://fate.ffmpeg.org/report.cgi?time=20220609221253&slot=x86_32-debian-kfreebsd-gcc-4.4-cpuflags-mmx

Notice that yuv2yuvX_mmx is not removed, because it is used
by SSE3 and AVX2 as fallback in case of unaligned data and
also for tail processing. I don't know why yuv2yuvX_mmxext
isn't being used for this; an earlier version [1] of
554c2bc7086f49ef5a6a989ad6bc4bc11807eb6f used it, but
the version that was eventually applied does not.

[1]: https://ffmpeg.org/pipermail/ffmpeg-devel/2020-November/272124.html

Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>

5 files changed, 38 insertions, 232 deletions

diff --git a/libswscale/x86/input.asm b/libswscale/x86/input.asm
index fcdfe2fcd8..6de6733faa 100644
--- a/libswscale/x86/input.asm
+++ b/libswscale/x86/input.asm
@@ -133,23 +133,18 @@ SECTION .text
 ; %2 = rgb or bgr
 %macro RGB24_TO_Y_FN 2-3
 cglobal %2 %+ 24ToY, 6, 6, %1, dst, src, u1, u2, w, table
-%if mmsize == 8
-    mova           m5, [%2_Ycoeff_12x4]
-    mova           m6, [%2_Ycoeff_3x56]
-%define coeff1 m5
-%define coeff2 m6
-%elif ARCH_X86_64
+%if ARCH_X86_64
     mova           m8, [%2_Ycoeff_12x4]
     mova           m9, [%2_Ycoeff_3x56]
 %define coeff1 m8
 %define coeff2 m9
-%else ; x86-32 && mmsize == 16
+%else ; x86-32
 %define coeff1 [%2_Ycoeff_12x4]
 %define coeff2 [%2_Ycoeff_3x56]
-%endif ; x86-32/64 && mmsize == 8/16
-%if (ARCH_X86_64 || mmsize == 8) && %0 == 3
+%endif ; x86-32/64
+%if ARCH_X86_64 && %0 == 3
     jmp mangle(private_prefix %+ _ %+ %3 %+ 24ToY %+ SUFFIX).body
-%else ; (ARCH_X86_64 && %0 == 3) || mmsize == 8
+%else ; ARCH_X86_64 && %0 == 3
 .body:
 %if cpuflag(ssse3)
     mova           m7, [shuf_rgb_12x4]
@@ -184,7 +179,6 @@ cglobal %2 %+ 24ToY, 6, 6, %1, dst, src, u1, u2, w, table
     movd           m1, [srcq+2]           ; (byte) { R0, B1, G1, R1 }
     movd           m2, [srcq+6]           ; (byte) { B2, G2, R2, B3 }
     movd           m3, [srcq+8]           ; (byte) { R2, B3, G3, R3 }
-%if mmsize == 16 ; i.e. sse2
     punpckldq      m0, m2                 ; (byte) { B0, G0, R0, B1, B2, G2, R2, B3 }
     punpckldq      m1, m3                 ; (byte) { R0, B1, G1, R1, R2, B3, G3, R3 }
     movd           m2, [srcq+12]          ; (byte) { B4, G4, R4, B5 }
@@ -193,7 +187,6 @@ cglobal %2 %+ 24ToY, 6, 6, %1, dst, src, u1, u2, w, table
     movd           m6, [srcq+20]          ; (byte) { R6, B7, G7, R7 }
     punpckldq      m2, m5                 ; (byte) { B4, G4, R4, B5, B6, G6, R6, B7 }
     punpckldq      m3, m6                 ; (byte) { R4, B5, G5, R5, R6, B7, G7, R7 }
-%endif ; mmsize == 16
     punpcklbw      m0, m7                 ; (word) { B0, G0, R0, B1, B2, G2, R2, B3 }
     punpcklbw      m1, m7                 ; (word) { R0, B1, G1, R1, R2, B3, G3, R3 }
     punpcklbw      m2, m7                 ; (word) { B4, G4, R4, B5, B6, G6, R6, B7 }
@@ -215,7 +208,7 @@ cglobal %2 %+ 24ToY, 6, 6, %1, dst, src, u1, u2, w, table
     add            wq, mmsize
     jl .loop
     REP_RET
-%endif ; (ARCH_X86_64 && %0 == 3) || mmsize == 8
+%endif ; ARCH_X86_64 && %0 == 3
 %endmacro
 
 ; %1 = nr. of XMM registers
@@ -275,12 +268,10 @@ cglobal %2 %+ 24ToUV, 7, 7, %1, dstU, dstV, u1, src, u2, w, table
     movd           m1, [srcq+2]           ; (byte) { R0, B1, G1, R1 }
     movd           m4, [srcq+6]           ; (byte) { B2, G2, R2, B3 }
     movd           m5, [srcq+8]           ; (byte) { R2, B3, G3, R3 }
-%if mmsize == 16
     punpckldq      m0, m4                 ; (byte) { B0, G0, R0, B1, B2, G2, R2, B3 }
     punpckldq      m1, m5                 ; (byte) { R0, B1, G1, R1, R2, B3, G3, R3 }
     movd           m4, [srcq+12]          ; (byte) { B4, G4, R4, B5 }
     movd           m5, [srcq+14]          ; (byte) { R4, B5, G5, R5 }
-%endif ; mmsize == 16
     punpcklbw      m0, m7                 ; (word) { B0, G0, R0, B1, B2, G2, R2, B3 }
     punpcklbw      m1, m7                 ; (word) { R0, B1, G1, R1, R2, B3, G3, R3 }
 %endif ; cpuflag(ssse3)
@@ -294,12 +285,10 @@ cglobal %2 %+ 24ToUV, 7, 7, %1, dstU, dstV, u1, src, u2, w, table
     pshufb         m5, m4, shuf_rgb2      ; (word) { R4, B5, G5, R5, R6, B7, G7, R7 }
     pshufb         m4, shuf_rgb1          ; (word) { B4, G4, R4, B5, B6, G6, R6, B7 }
 %else ; !cpuflag(ssse3)
-%if mmsize == 16
     movd           m1, [srcq+18]          ; (byte) { B6, G6, R6, B7 }
     movd           m3, [srcq+20]          ; (byte) { R6, B7, G7, R7 }
     punpckldq      m4, m1                 ; (byte) { B4, G4, R4, B5, B6, G6, R6, B7 }
     punpckldq      m5, m3                 ; (byte) { R4, B5, G5, R5, R6, B7, G7, R7 }
-%endif ; mmsize == 16 && !cpuflag(ssse3)
     punpcklbw      m4, m7                 ; (word) { B4, G4, R4, B5, B6, G6, R6, B7 }
     punpcklbw      m5, m7                 ; (word) { R4, B5, G5, R5, R6, B7, G7, R7 }
 %endif ; cpuflag(ssse3)
@@ -320,13 +309,8 @@ cglobal %2 %+ 24ToUV, 7, 7, %1, dstU, dstV, u1, src, u2, w, table
     psrad          m4, 9
     packssdw       m0, m1                 ; (word) { U[0-7] }
     packssdw       m2, m4                 ; (word) { V[0-7] }
-%if mmsize == 8
     mova   [dstUq+wq], m0
     mova   [dstVq+wq], m2
-%else ; mmsize == 16
-    mova   [dstUq+wq], m0
-    mova   [dstVq+wq], m2
-%endif ; mmsize == 8/16
     add            wq, mmsize
     jl .loop
     REP_RET
@@ -342,11 +326,6 @@ RGB24_TO_UV_FN %2, rgb
 RGB24_TO_UV_FN %2, bgr, rgb
 %endmacro
 
-%if ARCH_X86_32
-INIT_MMX mmx
-RGB24_FUNCS 0, 0
-%endif
-
 INIT_XMM sse2
 RGB24_FUNCS 10, 12
 
@@ -483,13 +462,8 @@ cglobal %2%3%4%5 %+ ToUV, 7, 7, %1, dstU, dstV, u1, src, u2, w, table
     psrad          m1, 9
     packssdw       m0, m4                 ; (word) { U[0-7] }
     packssdw       m2, m1                 ; (word) { V[0-7] }
-%if mmsize == 8
     mova   [dstUq+wq], m0
     mova   [dstVq+wq], m2
-%else ; mmsize == 16
-    mova   [dstUq+wq], m0
-    mova   [dstVq+wq], m2
-%endif ; mmsize == 8/16
     add            wq, mmsize
     jl .loop
     sub            wq, mmsize - 1
@@ -535,11 +509,6 @@ RGB32_TO_UV_FN %2, a, r, g, b, rgba
 RGB32_TO_UV_FN %2, a, b, g, r, rgba
 %endmacro
 
-%if ARCH_X86_32
-INIT_MMX mmx
-RGB32_FUNCS 0, 0
-%endif
-
 INIT_XMM sse2
 RGB32_FUNCS 8, 12
 
@@ -588,25 +557,18 @@ cglobal %2ToY, 5, 5, %1, dst, unused0, unused1, src, w
     movsxd         wq, wd
 %endif
     add          dstq, wq
-%if mmsize == 16
     test         srcq, 15
-%endif
     lea          srcq, [srcq+wq*2]
 %ifidn %2, yuyv
     pcmpeqb        m2, m2                 ; (byte) { 0xff } x 16
     psrlw          m2, 8                  ; (word) { 0x00ff } x 8
 %endif ; yuyv
-%if mmsize == 16
     jnz .loop_u_start
     neg            wq
     LOOP_YUYV_TO_Y  a, %2
 .loop_u_start:
     neg            wq
     LOOP_YUYV_TO_Y  u, %2
-%else ; mmsize == 8
-    neg            wq
-    LOOP_YUYV_TO_Y  a, %2
-%endif ; mmsize == 8/16
 %endmacro
 
 ; %1 = a (aligned) or u (unaligned)
@@ -632,16 +594,9 @@ cglobal %2ToY, 5, 5, %1, dst, unused0, unused1, src, w
     packuswb       m0, m1                 ; (byte) { U0, V0, ..., U7, V7 }
     pand           m1, m0, m2             ; (word) { U0, U1, ..., U7 }
     psrlw          m0, 8                  ; (word) { V0, V1, ..., V7 }
-%if mmsize == 16
     packuswb       m1, m0                 ; (byte) { U0, ... U7, V1, ... V7 }
     movh   [dstUq+wq], m1
     movhps [dstVq+wq], m1
-%else ; mmsize == 8
-    packuswb       m1, m1                 ; (byte) { U0, ... U3 }
-    packuswb       m0, m0                 ; (byte) { V0, ... V3 }
-    movh   [dstUq+wq], m1
-    movh   [dstVq+wq], m0
-%endif ; mmsize == 8/16
     add            wq, mmsize / 2
     jl .loop_%1
     REP_RET
@@ -661,24 +616,24 @@ cglobal %2ToUV, 4, 5, %1, dstU, dstV, unused, src, w
 %endif
     add         dstUq, wq
     add         dstVq, wq
-%if mmsize == 16 && %0 == 2
+%if %0 == 2
     test         srcq, 15
 %endif
     lea          srcq, [srcq+wq*4]
     pcmpeqb        m2, m2                 ; (byte) { 0xff } x 16
     psrlw          m2, 8                  ; (word) { 0x00ff } x 8
     ; NOTE: if uyvy+avx, u/a are identical
-%if mmsize == 16 && %0 == 2
+%if %0 == 2
     jnz .loop_u_start
     neg            wq
     LOOP_YUYV_TO_UV a, %2
 .loop_u_start:
     neg            wq
     LOOP_YUYV_TO_UV u, %2
-%else ; mmsize == 8
+%else
     neg            wq
     LOOP_YUYV_TO_UV a, %2
-%endif ; mmsize == 8/16
+%endif
 %endmacro
 
 ; %1 = a (aligned) or u (unaligned)
@@ -716,35 +671,18 @@ cglobal %2ToUV, 4, 5, %1, dstU, dstV, unused, src, w
 %endif
     add         dstUq, wq
     add         dstVq, wq
-%if mmsize == 16
     test         srcq, 15
-%endif
     lea          srcq, [srcq+wq*2]
     pcmpeqb        m5, m5                 ; (byte) { 0xff } x 16
     psrlw          m5, 8                  ; (word) { 0x00ff } x 8
-%if mmsize == 16
     jnz .loop_u_start
     neg            wq
     LOOP_NVXX_TO_UV a, %2
 .loop_u_start:
     neg            wq
     LOOP_NVXX_TO_UV u, %2
-%else ; mmsize == 8
-    neg            wq
-    LOOP_NVXX_TO_UV a, %2
-%endif ; mmsize == 8/16
 %endmacro
 
-%if ARCH_X86_32
-INIT_MMX mmx
-YUYV_TO_Y_FN  0, yuyv
-YUYV_TO_Y_FN  0, uyvy
-YUYV_TO_UV_FN 0, yuyv
-YUYV_TO_UV_FN 0, uyvy
-NVXX_TO_UV_FN 0, nv12
-NVXX_TO_UV_FN 0, nv21
-%endif
-
 INIT_XMM sse2
 YUYV_TO_Y_FN  3, yuyv
 YUYV_TO_Y_FN  2, uyvy
diff --git a/libswscale/x86/output.asm b/libswscale/x86/output.asm
index 1e498fddf6..84e94baaf6 100644
--- a/libswscale/x86/output.asm
+++ b/libswscale/x86/output.asm
@@ -312,11 +312,9 @@ cglobal yuv2planeX_%1, %3, 8, %2, filter, fltsize, src, dst, w, dither, offset
 %endif ; %1 == 8/9/10/16
 %endmacro
 
-%if ARCH_X86_32
+%if ARCH_X86_32 && HAVE_ALIGNED_STACK == 0
 INIT_MMX mmxext
 yuv2planeX_fn  8,  0, 7
-yuv2planeX_fn  9,  0, 5
-yuv2planeX_fn 10,  0, 5
 %endif
 
 INIT_XMM sse2
@@ -407,19 +405,11 @@ cglobal yuv2plane1_%1, %3, %3, %2, src, dst, w, dither, offset
     movq            m3, [ditherq]        ; dither
     test       offsetd, offsetd
     jz              .no_rot
-%if mmsize == 16
     punpcklqdq      m3, m3
-%endif ; mmsize == 16
     PALIGNR         m3, m3, 3, m2
 .no_rot:
-%if mmsize == 8
-    mova            m2, m3
-    punpckhbw       m3, m4               ; byte->word
-    punpcklbw       m2, m4               ; byte->word
-%else
     punpcklbw       m3, m4
     mova            m2, m3
-%endif
 %elif %1 == 9
     pxor            m4, m4
     mova            m3, [pw_512]
@@ -431,36 +421,22 @@ cglobal yuv2plane1_%1, %3, %3, %2, src, dst, w, dither, offset
 %else ; %1 == 16
 %if cpuflag(sse4) ; sse4/avx
     mova            m4, [pd_4]
-%else ; mmx/sse2
+%else ; sse2
     mova            m4, [pd_4min0x40000]
     mova            m5, [minshort]
-%endif ; mmx/sse2/sse4/avx
+%endif ; sse2/sse4/avx
 %endif ; %1 == ..
 
     ; actual pixel scaling
-%if mmsize == 8
-    yuv2plane1_mainloop %1, a
-%else ; mmsize == 16
     test          dstq, 15
     jnz .unaligned
     yuv2plane1_mainloop %1, a
     REP_RET
 .unaligned:
     yuv2plane1_mainloop %1, u
-%endif ; mmsize == 8/16
     REP_RET
 %endmacro
 
-%if ARCH_X86_32
-INIT_MMX mmx
-yuv2plane1_fn  8, 0, 5
-yuv2plane1_fn 16, 0, 3
-
-INIT_MMX mmxext
-yuv2plane1_fn  9, 0, 3
-yuv2plane1_fn 10, 0, 3
-%endif
-
 INIT_XMM sse2
 yuv2plane1_fn  8, 5, 5
 yuv2plane1_fn  9, 5, 3
diff --git a/libswscale/x86/scale.asm b/libswscale/x86/scale.asm
index 83cabff722..c62ae3dcc2 100644
--- a/libswscale/x86/scale.asm
+++ b/libswscale/x86/scale.asm
@@ -61,13 +61,11 @@ cglobal hscale%1to%2_%4, %5, 10, %6, pos0, dst, w, srcmem, filter, fltpos, fltsi
 %define mov32 mov
 %endif ; x86-64
 %if %2 == 19
-%if mmsize == 8 ; mmx
-    mova          m2, [max_19bit_int]
-%elif cpuflag(sse4)
+%if cpuflag(sse4)
     mova          m2, [max_19bit_int]
 %else ; ssse3/sse2
     mova          m2, [max_19bit_flt]
-%endif ; mmx/sse2/ssse3/sse4
+%endif ; sse2/ssse3/sse4
 %endif ; %2 == 19
 %if %1 == 16
     mova          m6, [minshort]
@@ -144,12 +142,7 @@ cglobal hscale%1to%2_%4, %5, 10, %6, pos0, dst, w, srcmem, filter, fltpos, fltsi
     pmaddwd       m1, [filterq+wq*8+mmsize*1]   ; *= filter[{8,9,..,14,15}]
 
     ; add up horizontally (4 srcpix * 4 coefficients -> 1 dstpix)
-%if mmsize == 8 ; mmx
-    movq          m4, m0
-    punpckldq     m0, m1
-    punpckhdq     m4, m1
-    paddd         m0, m4
-%elif notcpuflag(ssse3) ; sse2
+%if notcpuflag(ssse3) ; sse2
     mova          m4, m0
     shufps        m0, m1, 10001000b
     shufps        m4, m1, 11011101b
@@ -159,7 +152,7 @@ cglobal hscale%1to%2_%4, %5, 10, %6, pos0, dst, w, srcmem, filter, fltpos, fltsi
                                                 ; filter[{ 4, 5, 6, 7}]*src[filterPos[1]+{0,1,2,3}],
                                                 ; filter[{ 8, 9,10,11}]*src[filterPos[2]+{0,1,2,3}],
                                                 ; filter[{12,13,14,15}]*src[filterPos[3]+{0,1,2,3}]
-%endif ; mmx/sse2/ssse3/sse4
+%endif ; sse2/ssse3/sse4
 %else ; %3 == 8, i.e. filterSize == 8 scaling
     ; load 2x8 or 4x8 source pixels into m0, m1, m4 and m5
     mov32      pos0q, dword [fltposq+wq*2+0]    ; filterPos[0]
@@ -197,14 +190,7 @@ cglobal hscale%1to%2_%4, %5, 10, %6, pos0, dst, w, srcmem, filter, fltpos, fltsi
     pmaddwd       m5, [filterq+wq*8+mmsize*3]   ; *= filter[{24,25,..,30,31}]
 
     ; add up horizontally (8 srcpix * 8 coefficients -> 1 dstpix)
-%if mmsize == 8
-    paddd         m0, m1
-    paddd         m4, m5
-    movq          m1, m0
-    punpckldq     m0, m4
-    punpckhdq     m1, m4
-    paddd         m0, m1
-%elif notcpuflag(ssse3) ; sse2
+%if notcpuflag(ssse3) ; sse2
 %if %1 == 8
 %define mex m6
 %else
@@ -233,7 +219,7 @@ cglobal hscale%1to%2_%4, %5, 10, %6, pos0, dst, w, srcmem, filter, fltpos, fltsi
                                                 ; filter[{ 8, 9,...,14,15}]*src[filterPos[1]+{0,1,...,6,7}],
                                                 ; filter[{16,17,...,22,23}]*src[filterPos[2]+{0,1,...,6,7}],
                                                 ; filter[{24,25,...,30,31}]*src[filterPos[3]+{0,1,...,6,7}]
-%endif ; mmx/sse2/ssse3/sse4
+%endif ; sse2/ssse3/sse4
 %endif ; %3 == 4/8
 
 %else ; %3 == X, i.e. any filterSize scaling
@@ -274,7 +260,7 @@ cglobal hscale%1to%2_%4, %5, 10, %6, pos0, dst, w, srcmem, filter, fltpos, fltsi
     mov         srcq, srcmemmp
 
 .innerloop:
-    ; load 2x4 (mmx) or 2x8 (sse) source pixels into m0/m1 -> m4/m5
+    ; load 2x8 (sse) source pixels into m0/m1 -> m4/m5
     movbh         m0, [srcq+ pos0q     *srcmul] ; src[filterPos[0] + {0,1,2,3(,4,5,6,7)}]
     movbh         m1, [srcq+(pos1q+dlt)*srcmul] ; src[filterPos[1] + {0,1,2,3(,4,5,6,7)}]
 %if %1 == 8
@@ -319,12 +305,6 @@ cglobal hscale%1to%2_%4, %5, 10, %6, pos0, dst, w, srcmem, filter, fltpos, fltsi
 
     lea      filterq, [filterq+(fltsizeq+dlt)*2]
 
-%if mmsize == 8 ; mmx
-    movq          m0, m4
-    punpckldq     m4, m5
-    punpckhdq     m0, m5
-    paddd         m0, m4
-%else ; mmsize == 16
 %if notcpuflag(ssse3) ; sse2
     mova          m1, m4
     punpcklqdq    m4, m5
@@ -344,7 +324,6 @@ cglobal hscale%1to%2_%4, %5, 10, %6, pos0, dst, w, srcmem, filter, fltpos, fltsi
     phaddd        m4, m4
     SWAP           0, 4
 %endif ; sse2/ssse3/sse4
-%endif ; mmsize == 8/16
 %endif ; %3 ==/!= X
 
 %if %1 == 16 ; add 0x8000 * sum(coeffs), i.e. back from signed -> unsigned
@@ -372,7 +351,7 @@ cglobal hscale%1to%2_%4, %5, 10, %6, pos0, dst, w, srcmem, filter, fltpos, fltsi
 %endif ; %3 ==/!= X
 %endif ; %2 == 15/19
 %ifnidn %3, X
-    add           wq, (mmsize<<wshr)/4          ; both 8tap and 4tap really only do 4 pixels (or for mmx: 2 pixels)
+    add           wq, (mmsize<<wshr)/4          ; both 8tap and 4tap really only do 4 pixels
                                                 ; per iteration. see "shl wq,1" above as for why we do this
 %else ; %3 == X
     add           wq, 2
@@ -385,12 +364,8 @@ cglobal hscale%1to%2_%4, %5, 10, %6, pos0, dst, w, srcmem, filter, fltpos, fltsi
 %macro SCALE_FUNCS 3
 SCALE_FUNC %1, %2, 4, 4,  6, %3
 SCALE_FUNC %1, %2, 8, 8,  6, %3
-%if mmsize == 8
-SCALE_FUNC %1, %2, X, X,  7, %3
-%else
 SCALE_FUNC %1, %2, X, X4, 7, %3
 SCALE_FUNC %1, %2, X, X8, 7, %3
-%endif
 %endmacro
 
 ; SCALE_FUNCS2 8_xmm_args, 9to10_xmm_args, 16_xmm_args
@@ -411,10 +386,6 @@ SCALE_FUNCS 14, 19, %2
 SCALE_FUNCS 16, 19, %3
 %endmacro
 
-%if ARCH_X86_32
-INIT_MMX mmx
-SCALE_FUNCS2 0, 0, 0
-%endif
 INIT_XMM sse2
 SCALE_FUNCS2 7, 6, 8
 INIT_XMM ssse3
diff --git a/libswscale/x86/swscale.c b/libswscale/x86/swscale.c
index 73869355b8..97bbc4f2d0 100644
--- a/libswscale/x86/swscale.c
+++ b/libswscale/x86/swscale.c
@@ -54,14 +54,6 @@ DECLARE_ASM_ALIGNED(8, const uint64_t, ff_bgr2UVOffset) = 0x8080808080808080ULL;
 DECLARE_ASM_ALIGNED(8, const uint64_t, ff_w1111)        = 0x0001000100010001ULL;
 
 
-//MMX versions
-#if HAVE_MMX_INLINE
-#undef RENAME
-#define COMPILE_TEMPLATE_MMXEXT 0
-#define RENAME(a) a ## _mmx
-#include "swscale_template.c"
-#endif
-
 // MMXEXT versions
 #if HAVE_MMXEXT_INLINE
 #undef RENAME
@@ -269,9 +261,6 @@ void ff_hscale ## from_bpc ## to ## to_bpc ## _ ## filter_n ## _ ## opt( \
     SCALE_FUNCS(X4, opt); \
     SCALE_FUNCS(X8, opt)
 
-#if ARCH_X86_32
-SCALE_FUNCS_MMX(mmx);
-#endif
 SCALE_FUNCS_SSE(sse2);
 SCALE_FUNCS_SSE(ssse3);
 SCALE_FUNCS_SSE(sse4);
@@ -288,9 +277,7 @@ void ff_yuv2planeX_ ## size ## _ ## opt(const int16_t *filter, int filterSize, \
     VSCALEX_FUNC(9,  opt); \
     VSCALEX_FUNC(10, opt)
 
-#if ARCH_X86_32
-VSCALEX_FUNCS(mmxext);
-#endif
+VSCALEX_FUNC(8, mmxext);
 VSCALEX_FUNCS(sse2);
 VSCALEX_FUNCS(sse4);
 VSCALEX_FUNC(16, sse4);
@@ -305,9 +292,6 @@ void ff_yuv2plane1_ ## size ## _ ## opt(const int16_t *src, uint8_t *dst, int ds
     VSCALE_FUNC(10, opt2); \
     VSCALE_FUNC(16, opt1)
 
-#if ARCH_X86_32
-VSCALE_FUNCS(mmx, mmxext);
-#endif
 VSCALE_FUNCS(sse2, sse2);
 VSCALE_FUNC(16, sse4);
 VSCALE_FUNCS(avx, avx);
@@ -337,9 +321,6 @@ void ff_ ## fmt ## ToUV_ ## opt(uint8_t *dstU, uint8_t *dstV, \
     INPUT_FUNC(rgb24, opt); \
     INPUT_FUNC(bgr24, opt)
 
-#if ARCH_X86_32
-INPUT_FUNCS(mmx);
-#endif
 INPUT_FUNCS(sse2);
 INPUT_FUNCS(ssse3);
 INPUT_FUNCS(avx);
@@ -470,19 +451,11 @@ av_cold void ff_sws_init_swscale_x86(SwsContext *c)
 {
     int cpu_flags = av_get_cpu_flags();
 
-#if HAVE_MMX_INLINE
-    if (INLINE_MMX(cpu_flags))
-        sws_init_swscale_mmx(c);
-#endif
 #if HAVE_MMXEXT_INLINE
     if (INLINE_MMXEXT(cpu_flags))
         sws_init_swscale_mmxext(c);
 #endif
     if(c->use_mmx_vfilter && !(c->flags & SWS_ACCURATE_RND)) {
-#if HAVE_MMX_EXTERNAL
-        if (EXTERNAL_MMX(cpu_flags))
-            c->yuv2planeX = yuv2yuvX_mmx;
-#endif
 #if HAVE_MMXEXT_EXTERNAL
         if (EXTERNAL_MMXEXT(cpu_flags))
             c->yuv2planeX = yuv2yuvX_mmxext;
@@ -496,6 +469,14 @@ av_cold void ff_sws_init_swscale_x86(SwsContext *c)
             c->yuv2planeX = yuv2yuvX_avx2;
 #endif
     }
+#if ARCH_X86_32 && !HAVE_ALIGNED_STACK
+    // The better yuv2planeX_8 functions need aligned stack on x86-32,
+    // so we use MMXEXT in this case if they are not available.
+    if (EXTERNAL_MMXEXT(cpu_flags)) {
+        if (c->dstBpc == 8 && !c->use_mmx_vfilter)
+            c->yuv2planeX = ff_yuv2planeX_8_mmxext;
+    }
+#endif /* ARCH_X86_32 && !HAVE_ALIGNED_STACK */
 
 #define ASSIGN_SCALE_FUNC2(hscalefn, filtersize, opt1, opt2) do { \
     if (c->srcBpc == 8) { \
@@ -519,12 +500,6 @@ av_cold void ff_sws_init_swscale_x86(SwsContext *c)
                                      ff_hscale16to19_ ## filtersize ## _ ## opt1; \
     } \
 } while (0)
-#define ASSIGN_MMX_SCALE_FUNC(hscalefn, filtersize, opt1, opt2) \
-    switch (filtersize) { \
-    case 4:  ASSIGN_SCALE_FUNC2(hscalefn, 4, opt1, opt2); break; \
-    case 8:  ASSIGN_SCALE_FUNC2(hscalefn, 8, opt1, opt2); break; \
-    default: ASSIGN_SCALE_FUNC2(hscalefn, X, opt1, opt2); break; \
-    }
 #define ASSIGN_VSCALEX_FUNC(vscalefn, opt, do_16_case, condition_8bit) \
 switch(c->dstBpc){ \
     case 16:                          do_16_case;                          break; \
@@ -546,46 +521,6 @@ switch(c->dstBpc){ \
             if (!c->chrSrcHSubSample) \
                 c->chrToYV12 = ff_ ## x ## ToUV_ ## opt; \
             break
-#if ARCH_X86_32
-    if (EXTERNAL_MMX(cpu_flags)) {
-        ASSIGN_MMX_SCALE_FUNC(c->hyScale, c->hLumFilterSize, mmx, mmx);
-        ASSIGN_MMX_SCALE_FUNC(c->hcScale, c->hChrFilterSize, mmx, mmx);
-        ASSIGN_VSCALE_FUNC(c->yuv2plane1, mmx, mmxext, cpu_flags & AV_CPU_FLAG_MMXEXT);
-
-        switch (c->srcFormat) {
-        case AV_PIX_FMT_YA8:
-            c->lumToYV12 = ff_yuyvToY_mmx;
-            if (c->needAlpha)
-                c->alpToYV12 = ff_uyvyToY_mmx;
-            break;
-        case AV_PIX_FMT_YUYV422:
-            c->lumToYV12 = ff_yuyvToY_mmx;
-            c->chrToYV12 = ff_yuyvToUV_mmx;
-            break;
-        case AV_PIX_FMT_UYVY422:
-            c->lumToYV12 = ff_uyvyToY_mmx;
-            c->chrToYV12 = ff_uyvyToUV_mmx;
-            break;
-        case AV_PIX_FMT_NV12:
-            c->chrToYV12 = ff_nv12ToUV_mmx;
-            break;
-        case AV_PIX_FMT_NV21:
-            c->chrToYV12 = ff_nv21ToUV_mmx;
-            break;
-        case_rgb(rgb24, RGB24, mmx);
-        case_rgb(bgr24, BGR24, mmx);
-        case_rgb(bgra,  BGRA,  mmx);
-        case_rgb(rgba,  RGBA,  mmx);
-        case_rgb(abgr,  ABGR,  mmx);
-        case_rgb(argb,  ARGB,  mmx);
-        default:
-            break;
-        }
-    }
-    if (EXTERNAL_MMXEXT(cpu_flags)) {
-        ASSIGN_VSCALEX_FUNC(c->yuv2planeX, mmxext, , 1);
-    }
-#endif /* ARCH_X86_32 */
 #define ASSIGN_SSE_SCALE_FUNC(hscalefn, filtersize, opt1, opt2) \
     switch (filtersize) { \
     case 4:  ASSIGN_SCALE_FUNC2(hscalefn, 4, opt1, opt2); break; \
diff --git a/libswscale/x86/swscale_template.c b/libswscale/x86/swscale_template.c
index 97d8cae613..6190fcb4fe 100644
--- a/libswscale/x86/swscale_template.c
+++ b/libswscale/x86/swscale_template.c
@@ -29,13 +29,8 @@
 #undef PREFETCH
 
 
-#if COMPILE_TEMPLATE_MMXEXT
 #define REAL_MOVNTQ(a,b) "movntq " #a ", " #b " \n\t"
 #define MOVNTQ2 "movntq "
-#else
-#define REAL_MOVNTQ(a,b) "movq " #a ", " #b " \n\t"
-#define MOVNTQ2 "movq "
-#endif
 #define MOVNTQ(a,b)  REAL_MOVNTQ(a,b)
 
 #define YSCALEYUV2PACKEDX_UV \
@@ -600,13 +595,8 @@ static void RENAME(yuv2rgb555_X)(SwsContext *c, const int16_t *lumFilter,
     "cmp   "dstw", "#index"     \n\t"\
     " jb       1b               \n\t"
 
-#if COMPILE_TEMPLATE_MMXEXT
 #undef WRITEBGR24
 #define WRITEBGR24(dst, dstw, index)  WRITEBGR24MMXEXT(dst, dstw, index)
-#else
-#undef WRITEBGR24
-#define WRITEBGR24(dst, dstw, index)  WRITEBGR24MMX(dst, dstw, index)
-#endif
 
 #if HAVE_6REGS
 static void RENAME(yuv2bgr24_X_ar)(SwsContext *c, const int16_t *lumFilter,
@@ -1478,17 +1468,13 @@ static av_cold void RENAME(sws_init_swscale)(SwsContext *c)
     }
 
     if (c->srcBpc == 8 && c->dstBpc <= 14) {
-    // Use the new MMX scaler if the MMXEXT one can't be used (it is faster than the x86 ASM one).
-#if COMPILE_TEMPLATE_MMXEXT
-    if (c->flags & SWS_FAST_BILINEAR && c->canMMXEXTBeUsed) {
-        c->hyscale_fast = ff_hyscale_fast_mmxext;
-        c->hcscale_fast = ff_hcscale_fast_mmxext;
-    } else {
-#endif /* COMPILE_TEMPLATE_MMXEXT */
-        c->hyscale_fast = NULL;
-        c->hcscale_fast = NULL;
-#if COMPILE_TEMPLATE_MMXEXT
-    }
-#endif /* COMPILE_TEMPLATE_MMXEXT */
+        // Use the new MMX scaler if the MMXEXT one can't be used (it is faster than the x86 ASM one).
+        if (c->flags & SWS_FAST_BILINEAR && c->canMMXEXTBeUsed) {
+            c->hyscale_fast = ff_hyscale_fast_mmxext;
+            c->hcscale_fast = ff_hcscale_fast_mmxext;
+        } else {
+            c->hyscale_fast = NULL;
+            c->hcscale_fast = NULL;
+        }
     }
 }