Make assembly position-independent

4 files changed, 188 insertions, 29 deletions

diff --git a/libass/x86/blend_bitmaps.asm b/libass/x86/blend_bitmaps.asm
index 7bde78a..63b2344 100644
--- a/libass/x86/blend_bitmaps.asm
+++ b/libass/x86/blend_bitmaps.asm
@@ -18,17 +18,48 @@
 ;* OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
 ;******************************************************************************
 
-%include "x86/x86inc.asm"
+%include "x86/utils.asm"
 
 SECTION_RODATA 32
 
+%if ARCH_X86_64 || !PIC
 times 32 db 0xFF
 edge_mask: times 32 db 0x00
 words_255: times 16 dw 0xFF
+%endif
 
 SECTION .text
 
 ;------------------------------------------------------------------------------
+; LOAD_EDGE_MASK 1:m_dst, 2:n, 3:tmp
+; Set n last bytes of xmm/ymm register to zero and other bytes to 255
+;------------------------------------------------------------------------------
+
+%macro LOAD_EDGE_MASK 3
+%if !PIC
+    movu m%1, [edge_mask + %2 - mmsize]
+%elif ARCH_X86_64
+    lea %3, [rel edge_mask]
+    movu m%1, [%3 + %2 - mmsize]
+%elif mmsize <= STACK_ALIGNMENT
+    %assign %%pad -(stack_offset + gprsize) & (mmsize - 1)
+    pxor m%1, m%1
+    mova [rsp - %%pad - mmsize], m%1
+    pcmpeqb m%1, m%1
+    mova [rsp - %%pad - 2 * mmsize], m%1
+    movu m%1, [rsp + %2 - %%pad - 2 * mmsize]
+%else
+    mov %3, rsp
+    and %3, -mmsize
+    pxor m%1, m%1
+    mova [%3 - mmsize], m%1
+    pcmpeqb m%1, m%1
+    mova [%3 - 2 * mmsize], m%1
+    movu m%1, [%3 + %2 - 2 * mmsize]
+%endif
+%endmacro
+
+;------------------------------------------------------------------------------
 ; BLEND_BITMAPS 1:add/sub
 ; void add_bitmaps(uint8_t *dst, intptr_t dst_stride,
 ;                  uint8_t *src, intptr_t src_stride,
@@ -51,8 +82,7 @@ cglobal %1_bitmaps, 5,7,3
     neg r4
     mov r6, r4
     and r4, mmsize - 1
-    lea t0, [edge_mask]
-    movu m2, [t0 + r4 - mmsize]
+    LOAD_EDGE_MASK 2, r4, t0
 %if !ARCH_X86_64
     mov r5, r5m
 %endif
@@ -113,9 +143,13 @@ cglobal mul_bitmaps, 1,7,7
     neg r6
     mov t0, r6
     and r6, mmsize - 1
-    lea t1, [edge_mask]
-    movu m4, [t1 + r6 - mmsize]
+    LOAD_EDGE_MASK 4, r6, t1
+%if ARCH_X86_64 || !PIC
     mova m5, [words_255]
+%else
+    mov t1d, 255 * 0x10001
+    BCASTD 5, t1d
+%endif
     pxor m6, m6
     mov t1, r7m
     imul t1, r5
diff --git a/libass/x86/blur.asm b/libass/x86/blur.asm
index 20ed455..db9dfe8 100644
--- a/libass/x86/blur.asm
+++ b/libass/x86/blur.asm
@@ -22,15 +22,18 @@
 
 SECTION_RODATA 32
 
+%if ARCH_X86_64 || !PIC
 words_zero: times 16 dw 0
 words_one: times 16 dw 1
 words_dither_init: times 8 dw  8, 40
 words_dither_flip: times 16 dw 48
+%if ARCH_X86_64
 words_sign: times 16 dw 0x8000
-
+%endif
 dwords_two: times 8 dd 2
 dwords_round: times 8 dd 0x8000
 dwords_lomask: times 8 dd 0xFFFF
+%endif
 
 SECTION .text
 
@@ -51,8 +54,13 @@ cglobal stripe_unpack, 5,6,3
     and r5, -mmsize
     sub r3, r4
     sub r2, r5
-    xor r5, r5
+%if ARCH_X86_64 || !PIC
     mova m2, [words_one]
+%else
+    mov r5d, 0x10001
+    BCASTD 2, r5d
+%endif
+    xor r5, r5
     jmp .row_loop
 
 .col_loop:
@@ -122,7 +130,12 @@ cglobal stripe_pack, 5,7,5
     add r3, r2
     MUL r4, mmsize
     sub r5, r6
+%if ARCH_X86_64 || !PIC
     mova m4, [words_dither_flip]
+%else
+    mov r6d, 48 * 0x10001
+    BCASTD 4, r6d
+%endif
     jmp .row_loop
 
 .col_loop:
@@ -151,7 +164,12 @@ cglobal stripe_pack, 5,7,5
     add r0, r5
     add r2, r4
 .row_loop:
+%if ARCH_X86_64 || !PIC
     mova m3, [words_dither_init]
+%else
+    mov r6d, 8 | 40 << 16
+    BCASTD 3, r6d
+%endif
     lea r6, [r2 + r4]
     cmp r6, r3
     jb .col_loop
@@ -209,7 +227,11 @@ STRIPE_PACK
 %endmacro
 
 %macro LOAD_LINE_COMPACT 5-6
+%if ARCH_X86_64 || !PIC
     lea %5, [words_zero]
+%else
+    mov %5, rsp
+%endif
     sub %5, %2
     cmp %4, %3
     cmovb %5, %4
@@ -235,7 +257,13 @@ STRIPE_PACK
 cglobal shrink_horz, 4,9,9
     DECLARE_REG_TMP 8
 %else
+%if !PIC
 cglobal shrink_horz, 4,7,8
+%else
+cglobal shrink_horz, 4,7,8, -mmsize
+    pxor m0, m0
+    mova [rsp], m0
+%endif
     DECLARE_REG_TMP 6
 %endif
     lea t0, [r2 + mmsize + 3]
@@ -248,13 +276,18 @@ cglobal shrink_horz, 4,7,8
     xor r4, r4
     MUL r3, mmsize
     sub r4, r3
+%if ARCH_X86_64 || !PIC
     mova m7, [dwords_lomask]
+%else
+    mov r5d, 0xFFFF
+    BCASTD 7, r5d
+%endif
 %if ARCH_X86_64
     mova m8, [dwords_two]
     lea r7, [words_zero]
     sub r7, r1
 %else
-    PUSH t0
+    mov [rsp - 4], t0
 %endif
 
     lea r5, [r0 + r3]
@@ -327,7 +360,12 @@ cglobal shrink_horz, 4,7,8
     paddd m0, m8
     paddd m1, m8
 %else
+%if !PIC
     mova m6, [dwords_two]
+%else
+    mov r6d, 2
+    BCASTD 6, r6d
+%endif
     paddd m0, m6
     paddd m1, m6
 %endif
@@ -348,12 +386,9 @@ cglobal shrink_horz, 4,7,8
 %if ARCH_X86_64
     cmp r0, t0
 %else
-    cmp r0, [rsp]
+    cmp r0, [rsp - 4]
 %endif
     jb .main_loop
-%if !ARCH_X86_64
-    ADD rsp, 4
-%endif
     RET
 %endmacro
 
@@ -371,8 +406,12 @@ SHRINK_HORZ
 %macro SHRINK_VERT 0
 %if ARCH_X86_64
 cglobal shrink_vert, 4,7,9
-%else
+%elif !PIC
 cglobal shrink_vert, 4,7,8
+%else
+cglobal shrink_vert, 4,7,8, -mmsize
+    pxor m0, m0
+    mova [rsp], m0
 %endif
     lea r2, [2 * r2 + mmsize - 1]
     lea r5, [r3 + 5]
@@ -381,11 +420,17 @@ cglobal shrink_vert, 4,7,8
     imul r2, r5
     MUL r3, mmsize
     add r2, r0
+%if ARCH_X86_64 || !PIC
     mova m7, [words_one]
 %if ARCH_X86_64
     mova m8, [words_sign]
 %endif
     lea r6, [words_zero]
+%else
+    mov r4d, 0x10001
+    BCASTD 7, r4d
+    mov r6, rsp
+%endif
     sub r6, r1
 
 .col_loop:
@@ -456,7 +501,13 @@ SHRINK_VERT
 cglobal expand_horz, 4,9,5
     DECLARE_REG_TMP 8
 %else
+%if !PIC
 cglobal expand_horz, 4,7,5
+%else
+cglobal expand_horz, 4,7,5, -mmsize
+    pxor m0, m0
+    mova [rsp], m0
+%endif
     DECLARE_REG_TMP 6
 %endif
     lea t0, [4 * r2 + 7]
@@ -469,7 +520,12 @@ cglobal expand_horz, 4,7,5
     xor r4, r4
     MUL r3, mmsize
     sub r4, r3
+%if ARCH_X86_64 || !PIC
     mova m4, [words_one]
+%else
+    mov r5d, 0x10001
+    BCASTD 4, r5d
+%endif
 %if ARCH_X86_64
     lea r7, [words_zero]
     sub r7, r1
@@ -479,7 +535,7 @@ cglobal expand_horz, 4,7,5
     cmp r0, t0
     jae .odd_stripe
 %if !ARCH_X86_64
-    PUSH t0
+    mov [rsp - 4], t0
 %endif
 .main_loop:
 %if ARCH_X86_64
@@ -527,14 +583,11 @@ cglobal expand_horz, 4,7,5
     add r0, r3
     lea r5, [r0 + r3]
 %if !ARCH_X86_64
-    mov t0, [rsp]
+    mov t0, [rsp - 4]
 %endif
     cmp r0, t0
     jb .main_loop
     add t0, r3
-%if !ARCH_X86_64
-    ADD rsp, 4
-%endif
     cmp r0, t0
     jb .odd_stripe
     RET
@@ -592,15 +645,27 @@ EXPAND_HORZ
 ;------------------------------------------------------------------------------
 
 %macro EXPAND_VERT 0
+%if ARCH_X86_64 || !PIC
 cglobal expand_vert, 4,7,5
+%else
+cglobal expand_vert, 4,7,5, -mmsize
+    pxor m0, m0
+    mova [rsp], m0
+%endif
     lea r2, [2 * r2 + mmsize - 1]
     lea r5, [2 * r3 + 4]
     and r2, -mmsize
     imul r2, r5
     MUL r3, mmsize
     add r2, r0
+%if ARCH_X86_64 || !PIC
     mova m4, [words_one]
     lea r6, [words_zero]
+%else
+    mov r4d, 0x10001
+    BCASTD 4, r4d
+    mov r6, rsp
+%endif
     sub r6, r1
 
 .col_loop:
@@ -744,7 +809,13 @@ EXPAND_VERT
     %assign %%narg 9 + (%1 + 1) / 2
 cglobal blur%1_horz, 5,8,%%narg
 %else
+%if !PIC
 cglobal blur%1_horz, 5,7,8
+%else
+cglobal blur%1_horz, 5,7,8, -mmsize
+    pxor m0, m0
+    mova [rsp], m0
+%endif
     SWAP 7, 9
 %endif
     LOAD_MULTIPLIER %1, 9, r4, r5
@@ -821,7 +892,12 @@ cglobal blur%1_horz, 5,7,8
 %if ARCH_X86_64
     mova m6, m7
 %else
+%if !PIC
     mova m6, [dwords_round]
+%else
+    mov r6d, 0x8000
+    BCASTD 6, r6d
+%endif
     mova [r0], m1
     SWAP 1, 8
 %endif
@@ -911,8 +987,12 @@ BLUR_HORZ 8
 %if ARCH_X86_64
     %assign %%narg 7 + (%1 + 1) / 2
 cglobal blur%1_vert, 5,7,%%narg
-%else
+%elif !PIC
 cglobal blur%1_vert, 5,7,8
+%else
+cglobal blur%1_vert, 5,7,8, -mmsize
+    pxor m0, m0
+    mova [rsp], m0
 %endif
     LOAD_MULTIPLIER %1, 7, r4, r5
     lea r2, [2 * r2 + mmsize - 1]
@@ -921,8 +1001,14 @@ cglobal blur%1_vert, 5,7,8
     imul r2, r5
     MUL r3, mmsize
     add r2, r0
+%if ARCH_X86_64 || !PIC
     mova m4, [dwords_round]
     lea r6, [words_zero]
+%else
+    mov r5d, 0x8000
+    BCASTD 4, r5d
+    mov r6, rsp
+%endif
     sub r6, r1
 
 .col_loop:
diff --git a/libass/x86/rasterizer.asm b/libass/x86/rasterizer.asm
index dc59a7b..0b53c10 100644
--- a/libass/x86/rasterizer.asm
+++ b/libass/x86/rasterizer.asm
@@ -23,8 +23,10 @@
 SECTION_RODATA 32
 
 words_index: dw 0x00,0x01,0x02,0x03,0x04,0x05,0x06,0x07,0x08,0x09,0x0A,0x0B,0x0C,0x0D,0x0E,0x0F
+%if ARCH_X86_64 || !PIC
 words_tile16: times 16 dw 1024
 words_tile32: times 16 dw 512
+%endif
 
 SECTION .text
 
@@ -56,12 +58,7 @@ cglobal fill_solid_tile%2, 3,4,1
     mov r3d, -1
     test r2d, r2d
     cmovnz r2d, r3d
-    movd xm0, r2d
-%if mmsize == 32
-    vpbroadcastd m0, xm0
-%else
-    pshufd m0, m0, q0000
-%endif
+    BCASTD 0, r2d
 
 %rep (1 << %1) - 1
     FILL_LINE r0, 0, 1 << %1
@@ -123,8 +120,13 @@ FILL_SOLID_TILE 5,32
     DEF_A_SHIFT %1
 %if ARCH_X86_64 && a_shift
 cglobal fill_halfplane_tile%2, 6,7,9
-%else
+%elif ARCH_X86_64 || !PIC
 cglobal fill_halfplane_tile%2, 0,7,8
+%else
+cglobal fill_halfplane_tile%2, 0,7,8, -mmsize
+    LEA r0, words_index
+    mova m0, [r0]
+    mova [rsp], m0
 %endif
 %if !a_shift
     SWAP 3, 8
@@ -182,7 +184,11 @@ cglobal fill_halfplane_tile%2, 0,7,8
 %if a_shift
     psllw m3, m2, a_shift  ; aa * (mmsize / 2)
 %endif
+%if ARCH_X86_64 || !PIC
     pmullw m2, [words_index]
+%else
+    pmullw m2, [rsp]
+%endif
     psubw m1, m2  ; cc - aa * i
 
     mov r4d, r2d  ; aa
@@ -212,7 +218,12 @@ cglobal fill_halfplane_tile%2, 0,7,8
 %endif
 
     pxor m0, m0
+%if ARCH_X86_64 || !PIC
     mova m4, [words_tile%2]
+%else
+    mov r4d, 0x10001 << (14 - %1)
+    BCASTD 4, r4d
+%endif
     mov r2d, (1 << %1)
     jmp .loop_entry
 
@@ -510,13 +521,15 @@ endstruc
     %assign alloc_size buf_size + 32
 %endif
     %assign alloc_size (alloc_size + mmsize - 1) & -mmsize
+%if !ARCH_X86_64 && PIC
+    %assign alloc_size alloc_size + mmsize
+%endif
     %xdefine delta (rsp + delta_offs)
     DEF_A_SHIFT %1
 
     %define zero 5
     %define vc   7
 %if ARCH_X86_64
-    %define m_index m8
     %define full 9
     %define vba 10
     %define van 11
@@ -525,9 +538,9 @@ cglobal fill_generic_tile%2, 5,14,12, -alloc_size
 %else
 cglobal fill_generic_tile%2, 5,14,11, -alloc_size
 %endif
+    %define m_index m8
 
 %else
-    %define m_index [words_index]
     %define full 4
 %if a_shift
     %define vba  3
@@ -544,6 +557,14 @@ cglobal fill_generic_tile%2, 0,7,8, -alloc_size
     movu xm0, r0m
     mova [args], xm0
 %endif
+%if ARCH_X86_64 || !PIC
+    %define m_index [words_index]
+%else
+    %define m_index [rsp + alloc_size - mmsize]
+    LEA r0, words_index
+    mova m1, [r0]
+    mova m_index, m1
+%endif
     mov r4d, r4m
 %endif
 
@@ -747,7 +768,12 @@ cglobal fill_generic_tile%2, 0,7,8, -alloc_size
     BCASTW 0, t2d
     paddw m%+vc, m0
 
+%if !PIC
     mova m%+full, [words_tile%2]
+%else
+    mov t0d, 0x10001 << (14 - %1)
+    BCASTD full, t0d
+%endif
 %endif
 .internal_loop:
 %assign i 0
diff --git a/libass/x86/utils.asm b/libass/x86/utils.asm
index 9d0ecb9..9102472 100644
--- a/libass/x86/utils.asm
+++ b/libass/x86/utils.asm
@@ -64,13 +64,26 @@
     movd xm%1, %2
 %if mmsize == 32
     vpbroadcastw m%1, xm%1
-%elif mmsize == 16
+%else
     punpcklwd m%1, m%1
     pshufd m%1, m%1, q0000
 %endif
 %endmacro
 
 ;------------------------------------------------------------------------------
+; BCASTD 1:m_dst, 2:r_src
+;------------------------------------------------------------------------------
+
+%macro BCASTD 2
+    movd xm%1, %2
+%if mmsize == 32
+    vpbroadcastd m%1, xm%1
+%else
+    pshufd m%1, m%1, q0000
+%endif
+%endmacro
+
+;------------------------------------------------------------------------------
 ; PABSW 1:m_reg, 2:m_tmp
 ;------------------------------------------------------------------------------