diff options
| -rw-r--r-- | configure.ac | 46 | ||||
| -rw-r--r-- | libass/Makefile.am | 46 | ||||
| -rw-r--r-- | libass/ass_render.c | 33 | ||||
| -rw-r--r-- | libass/ass_utils.c | 36 | ||||
| -rw-r--r-- | libass/ass_utils.h | 6 | ||||
| -rw-r--r-- | libass/x86/be_blur.asm | 239 | ||||
| -rw-r--r-- | libass/x86/be_blur.h | 30 | ||||
| -rw-r--r-- | libass/x86/blend_bitmaps.asm | 290 | ||||
| -rw-r--r-- | libass/x86/blend_bitmaps.h | 56 | ||||
| -rw-r--r-- | libass/x86/cpuid.asm | 32 | ||||
| -rw-r--r-- | libass/x86/cpuid.h | 24 | ||||
| -rw-r--r-- | libass/x86/x86inc.asm | 1450 |
12 files changed, 2284 insertions, 4 deletions
diff --git a/configure.ac b/configure.ac index 11c1f79..f7e6527 100644 --- a/configure.ac +++ b/configure.ac @@ -12,6 +12,7 @@ AC_CONFIG_HEADER([config.h]) AC_PROG_CC AC_PROG_CPP AM_PROG_CC_C_O +AM_PROG_AS # Checks for header files. AC_HEADER_STDC @@ -44,6 +45,51 @@ AC_ARG_ENABLE([fontconfig], AS_HELP_STRING([--disable-fontconfig], [disable fontconfig support @<:@default=enabled@:>@])) AC_ARG_ENABLE([harfbuzz], AS_HELP_STRING([--disable-harfbuzz], [disable HarfBuzz support @<:@default=check@:>@])) +AC_ARG_ENABLE([asm], AS_HELP_STRING([--disable-asm], + [disable compiling with ASM @<:@default=enabled@:>@]), + [asm=false], + [asm=true]) + +AM_CONDITIONAL([ASM], [test x$asm = xtrue]) + +AM_COND_IF([ASM], + [AC_DEFINE(CONFIG_ASM, 1, [ASM enabled])], + [AC_DEFINE(CONFIG_ASM, 0, [ASM disabled])] + ) + +test "${ASFLAGS+set}" = set || ASFLAGS="" + +AC_SUBST([ASFLAGS], ["$ASFLAGS"]) + +case $host in + i?86-*) + INTEL=true + X86=true ;; + x86_64-*) + INTEL=true + X64=true ;; + arm-*) + ARM=true ;; +esac + +case $host in + *darwin*) + MACHO=true ;; + *linux*) + ELF=true ;; + *cygwin*) + WIN32=true ;; + *mingw*) + WIN32=true ;; +esac + +AM_CONDITIONAL([INTEL], [test x$INTEL = xtrue]) +AM_CONDITIONAL([X86], [test x$X86 = xtrue]) +AM_CONDITIONAL([X64], [test x$X64 = xtrue]) +AM_CONDITIONAL([ARM], [test x$ARM = xtrue]) +AM_CONDITIONAL([MACHO], [test x$MACHO = xtrue]) +AM_CONDITIONAL([ELF], [test x$ELF = xtrue]) +AM_CONDITIONAL([WIN32], [test x$WIN32 = xtrue]) PKG_CHECK_MODULES([FREETYPE], freetype2 >= 9.10.3, [ CFLAGS="$CFLAGS $FREETYPE_CFLAGS" diff --git a/libass/Makefile.am b/libass/Makefile.am index 524d9a8..ddd5acf 100644 --- a/libass/Makefile.am +++ b/libass/Makefile.am @@ -6,6 +6,16 @@ LIBASS_LT_CURRENT = 5 LIBASS_LT_REVISION = 0 LIBASS_LT_AGE = 0 +.asm.lo: + $(LIBTOOL) --quiet --mode=compile $(AS) $(ASFLAGS) -o $@ $< -prefer-non-pic + +.S.lo: + $(LIBTOOL) --quiet --mode=compile $(AS) $(ASFLAGS)-o $@ $< -prefer-non-pic + +SRC_INTEL = x86/blend_bitmaps.asm x86/cpuid.asm +SRC_INTEL64 = x86/be_blur.asm +SRC_ARM = arm/blend_bitmaps.S + lib_LTLIBRARIES = libass.la libass_la_SOURCES = ass.c ass_cache.c ass_font.c ass_fontconfig.c ass_render.c \ ass_utils.c ass_bitmap.c ass_library.c ass_bitmap.h \ @@ -14,9 +24,45 @@ libass_la_SOURCES = ass.c ass_cache.c ass_font.c ass_fontconfig.c ass_render.c \ ass_drawing.h ass_cache_template.h ass_render.h \ ass_parse.c ass_parse.h ass_render_api.c ass_shaper.c \ ass_shaper.h ass_strtod.c + libass_la_LDFLAGS = -no-undefined -version-info $(LIBASS_LT_CURRENT):$(LIBASS_LT_REVISION):$(LIBASS_LT_AGE) libass_la_LDFLAGS += -export-symbols $(srcdir)/libass.sym +if ASM +if INTEL +AS = "yasm" +libass_la_SOURCES += $(SRC_INTEL) +if X86 +ASFLAGS += -DARCH_X86_64=0 -m x86 +if MACHO +ASFLAGS += -f macho32 -DPREFIX +endif +if ELF +ASFLAGS += -f elf +endif +if WIN32 +ASFLAGS += -f win32 -DPREFIX +endif +endif +if X64 +libass_la_SOURCES += $(SRC_INTEL64) +ASFLAGS += -DARCH_X86_64=1 -m amd64 +if MACHO +ASFLAGS += -f macho64 -DPREFIX +endif +if ELF +ASFLAGS += -f elf +endif +if WIN32 +ASFLAGS += -f win64 +endif +endif +endif +if ARM +libass_la_SOURCES += $(SRC_ARM) +endif +endif + assheadersdir = $(includedir)/ass dist_assheaders_HEADERS = ass.h ass_types.h diff --git a/libass/ass_render.c b/libass/ass_render.c index e6e6052..0f17404 100644 --- a/libass/ass_render.c +++ b/libass/ass_render.c @@ -33,6 +33,12 @@ #define SUBPIXEL_MASK 63 #define SUBPIXEL_ACCURACY 7 +#if (defined(__i386__) || defined(__x86_64__)) && CONFIG_ASM + +#include "x86/blend_bitmaps.h" +#include "x86/be_blur.h" + +#endif // ASM ASS_Renderer *ass_renderer_init(ASS_Library *library) { @@ -63,10 +69,29 @@ ASS_Renderer *ass_renderer_init(ASS_Library *library) priv->ftlibrary = ft; // images_root and related stuff is zero-filled in calloc - priv->add_bitmaps_func = add_bitmaps_c; - priv->sub_bitmaps_func = sub_bitmaps_c; - priv->mul_bitmaps_func = mul_bitmaps_c; - priv->be_blur_func = be_blur_c; + #if (defined(__i386__) || defined(__x86_64__)) && CONFIG_ASM + int sse2 = has_sse2(); + int avx2 = has_avx2(); + priv->add_bitmaps_func = avx2 ? ass_add_bitmaps_avx2 : + (sse2 ? ass_add_bitmaps_sse2 : ass_add_bitmaps_x86); + #ifdef __x86_64__ + priv->be_blur_func = avx2 ? ass_be_blur_avx2 : + (sse2 ? ass_be_blur_sse2 : be_blur_c); + priv->mul_bitmaps_func = avx2 ? ass_mul_bitmaps_avx2 : + (sse2 ? ass_mul_bitmaps_sse2 : mul_bitmaps_c); + priv->sub_bitmaps_func = avx2 ? ass_sub_bitmaps_avx2 : + (sse2 ? ass_sub_bitmaps_sse2 : ass_sub_bitmaps_x86); + #else + priv->be_blur_func = be_blur_c; + priv->mul_bitmaps_func = mul_bitmaps_c; + priv->sub_bitmaps_func = ass_sub_bitmaps_x86; + #endif + #else + priv->add_bitmaps_func = add_bitmaps_c; + priv->sub_bitmaps_func = sub_bitmaps_c; + priv->mul_bitmaps_func = mul_bitmaps_c; + priv->be_blur_func = be_blur_c; + #endif priv->restride_bitmap_func = restride_bitmap_c; priv->cache.font_cache = ass_font_cache_create(); diff --git a/libass/ass_utils.c b/libass/ass_utils.c index 2a95ddc..549eb9b 100644 --- a/libass/ass_utils.c +++ b/libass/ass_utils.c @@ -29,6 +29,42 @@ #include "ass.h" #include "ass_utils.h" +#if (defined(__i386__) || defined(__x86_64__)) && CONFIG_ASM + +#include "x86/cpuid.h" + +int has_sse2() +{ + uint32_t eax = 1, ebx, ecx, edx; + ass_get_cpuid(&eax, &ebx, &ecx, &edx); + return (!!(edx & (1 << 26))); +} + +int has_avx() +{ + uint32_t eax = 1, ebx, ecx, edx; + ass_get_cpuid(&eax, &ebx, &ecx, &edx); + if(!(ecx & (1 << 27))){ + return 0; + } + uint32_t misc = ecx; + eax = 0; + ass_get_cpuid(&eax, &ebx, &ecx, &edx); + if((ecx & (0x2 | 0x4)) != (0x2 | 0x4)){ + return 0; + } + return (!!(misc & (1 << 28))); +} + +int has_avx2() +{ + uint32_t eax = 7, ebx, ecx, edx; + ass_get_cpuid(&eax, &ebx, &ecx, &edx); + return (!!(ebx & (1 << 5))) && has_avx(); +} + +#endif // ASM + int mystrtoi(char **p, int *res) { double temp_res; diff --git a/libass/ass_utils.h b/libass/ass_utils.h index 39a9da0..6d795f0 100644 --- a/libass/ass_utils.h +++ b/libass/ass_utils.h @@ -43,6 +43,12 @@ #define FFMIN(a,b) ((a) > (b) ? (b) : (a)) #define FFMINMAX(c,a,b) FFMIN(FFMAX(c, a), b) +#if (defined(__i386__) || defined(__x86_64__)) && CONFIG_ASM +int has_sse2(void); +int has_avx(void); +int has_avx2(void); +#endif + int mystrtoi(char **p, int *res); int mystrtoll(char **p, long long *res); int mystrtou32(char **p, int base, uint32_t *res); diff --git a/libass/x86/be_blur.asm b/libass/x86/be_blur.asm new file mode 100644 index 0000000..8acf409 --- /dev/null +++ b/libass/x86/be_blur.asm @@ -0,0 +1,239 @@ +;****************************************************************************** +;* be_blur.asm: SSE2 \be blur +;****************************************************************************** + +%include "x86inc.asm" + +SECTION_RODATA 32 +low_word_zero: dd 0xFFFF0000, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF + +SECTION .text + +;------------------------------------------------------------------------------ +; void be_blur_pass( uint8_t *buf, unsigned width, +; unsigned height, unsigned stride, +; uint16_t *tmp); +;------------------------------------------------------------------------------ + +INIT_XMM sse2 +cglobal be_blur, 5,15 +.skip_prologue: + xor r5, r5 ; int y = 0; + mov r6, 2 ; int x = 2; + pxor xmm6, xmm6 ; __m128i temp3 = 0; + mov r7, r0 ; unsigned char *src=buf; + movzx r8, byte [r7 + 1] ; int old_pix = src[1]; + movzx r9, byte [r7] ; int old_sum = src[0]; + add r9, r8 ; old_sum += old_pix; + lea r12, [r4 + r3 * 2] ; unsigned char *col_sum_buf = tmp + stride * 2; + lea r14, [r1 - 2] ; tmpreg = (stride-2); + and r14, -8 ; tmpreg &= (~7); +.first_loop + movzx r10, byte [r7 + r6] ; int temp1 = src[x]; + lea r11, [r8 + r10] ; int temp2 = old_pix + temp1; + mov r8, r10 ; old_pix = temp1; + lea r10, [r9 + r11] ; temp1 = old_sum + temp2; + mov r9, r11 ; old_sum = temp2; + mov word [r4 + r6 * 2], r10w ; col_pix_buf[x] = temp1; + inc r6 ; x++ + cmp r6, r1 ; x < w + jl .first_loop + mov r5, 1 ; int y = 1; + mov r6, 2 ; int x = 2; + lea r7, [r0 + r3] ; unsigned char *src=buf+stride; + movzx r8, byte [r7 + 1] ; int old_pix = src[1]; + movzx r9, byte [r7] ; int old_sum = src[0]; + add r9, r8 ; old_sum += old_pix +.second_loop + movzx r10, byte [r7 + r6] ; int temp1 = src[x]; + lea r11, [r8 + r10] ; int temp2 = old_pix + temp1; + mov r8, r10 ; old_pix = temp1; + lea r10, [r9 + r11] ; temp1 = old_sum + temp2; + mov r9, r11 ; old_sum = temp2; + movzx r11, word [r4 + r6 * 2] ; temp2 = col_pix_buf[x]; + add r11, r10 ; temp2 += temp1; + mov word [r4 + r6 * 2], r10w ; col_pix_buf[x] = temp1; + mov word [r12 + r6 * 2], r11w ; col_sum_buf[x] = temp2; + inc r6 ; x++ + cmp r6, r1 ; x < w + jl .second_loop +.height_loop + mov r10, r5; int tmpreg = y; + imul r10, r3; tmpreg *= stride; + lea r7, [r0 + r10] ; unsigned char *src=buf+y*stride; + sub r10, r3 ; tmpreg -= stride; + lea r13, [r0 + r10]; unsigned char *dst=buf+(y-1)*stride; + mov r6, 2 ; int x = 2; + movzx r10, byte [r7] ; temp1 = src[0]; + movzx r11, byte [r7 + 1] ; temp2 = src[1]; + add r10, r11; temp1 += temp2 + movd xmm0, r10; __m128i old_pix_128 = temp2; + movd xmm1, r11; __m128i old_sum_128 = temp1; +.width_loop + movq xmm2, [r7 + r6]; __m128i new_pix = (src+x); + punpcklbw xmm2, xmm6 ; new_pix = _mm_unpacklo_epi8(new_pix, temp3); + movdqa xmm3, xmm2 ; __m128i temp = new_pix; + pslldq xmm3, 2 ; temp = temp << 2 * 8; + paddw xmm3, xmm0 ; temp = _mm_add_epi16(temp, old_pix_128); + paddw xmm3, xmm2 ; temp = _mm_add_epi16(temp, new_pix); + movdqa xmm0, xmm2 ; old_pix_128 = new_pix; + psrldq xmm0, 14 ; old_pix_128 = old_pix_128 >> 14 * 8; + movdqa xmm2, xmm3 ; new_pix = temp; + pslldq xmm2, 2 ; new_pix = new_pix << 2 * 8; + paddw xmm2, xmm1 ; new_pix = _mm_add_epi16(new_pix, old_sum_128); + paddw xmm2, xmm3 ; new_pix = _mm_add_epi16(new_pix, temp); + movdqa xmm1, xmm3 ; old_sum_128 = temp; + psrldq xmm1, 14 ; old_sum_128 = old_sum_128 >> 14 * 8; + movdqu xmm4, [r4 + r6 * 2] ; __m128i old_col_pix = *(col_pix_buf+x); + movdqu [r4 + r6 * 2], xmm2 ; *(col_pix_buf+x) = new_pix ; + movdqu xmm5, [r12 + r6 * 2] ; __m128i old_col_sum = *(col_pix_sum+x); + movdqa xmm3, xmm2 ; temp = new_pix; + paddw xmm3, xmm4 ; temp = _mm_add_epi16(temp, old_col_pix); + movdqu [r12 + r6 * 2], xmm3 ; *(col_sum_buf+x) = temp; + paddw xmm5, xmm3 ; old_col_sum = _mm_add_epi16(old_col_sum, temp); + psrlw xmm5, 4 ; old_col_sum = old_col_sum >> 4; + packuswb xmm5, xmm5 ; old_col_sum = _mm_packus_epi16(old_col_sum, old_col_sum); + movq qword [r13 + r6 - 1], xmm5 ; *(dst+x-1) = old_col_sum; + add r6, 8; x += 8; + cmp r6, r14; x < ((w - 2) & (~7)); + jl .width_loop + movzx r8, byte [r7 + r6 - 1] ; old_pix = src[x-1]; + movzx r9, byte [r7 + r6 - 2] ; old_sum = old_pix + src[x-2]; + add r9, r8 +.final_width_loop + movzx r10, byte [r7 + r6] ; temp1 = src[x]; + lea r11, [r8 + r10] ; temp2 = old_pix + temp1; + mov r8, r10 ; old_pix = temp1; + lea r10, [r9 + r11] ; temp1 = old_sum + temp2; + mov r9, r11 ; old_sum = temp2; + movzx r11, word [r4 + r6 * 2] ; temp2 = col_pix_buf[x]; + add r11, r10 ; temp2 += temp1; + mov word [r4 + r6 * 2], r10w ; col_pix_buf[x] = temp1; + movzx r10, word [r12 + r6 * 2] ; temp1 = col_sum_buf[x]; + add r10, r11 ; temp1 += temp2; + shr r10, 4 ; temp1 >>= 4; + mov byte [r13 + r6 - 1], r10b ; dst[x-1] = temp1 + mov [r12 + r6 * 2], r11w ; col_sum_buf[x] = temp2; + inc r6 ; x++ + cmp r6, r1 ; x < w + jl .final_width_loop + inc r5 ; y++; + cmp r5, r2 ; y < h; + jl .height_loop + RET + +INIT_YMM avx2 +cglobal be_blur, 5,15 + %if mmsize == 32 + vzeroupper + %endif + cmp r1, 32 + jl be_blur_sse2.skip_prologue + xor r5, r5 ; int y = 0; + mov r6, 2 ; int x = 2; + vpxor ymm6, ymm6 ; __m128i temp3 = 0; + mov r7, r0 ; unsigned char *src=buf; + movzx r8, byte [r7 + 1] ; int old_pix = src[1]; + movzx r9, byte [r7] ; int old_sum = src[0]; + add r9, r8 ; old_sum += old_pix; + lea r12, [r4 + r3 * 2] ; unsigned char *col_sum_buf = tmp + stride * 2; + lea r14, [r1 - 2] ; tmpreg = (stride-2); + and r14, -16 ; tmpreg &= (~15); + vmovdqa ymm8, [low_word_zero wrt rip] +.first_loop + movzx r10, byte [r7 + r6] ; int temp1 = src[x]; + lea r11, [r8 + r10] ; int temp2 = old_pix + temp1; + mov r8, r10 ; old_pix = temp1; + lea r10, [r9 + r11] ; temp1 = old_sum + temp2; + mov r9, r11 ; old_sum = temp2; + mov word [r4 + r6 * 2], r10w ; col_pix_buf[x] = temp1; + inc r6 ; x++ + cmp r6, r1 ; x < w + jl .first_loop + mov r5, 1 ; int y = 1; + mov r6, 2 ; int x = 2; + lea r7, [r0 + r3] ; unsigned char *src=buf+stride; + movzx r8, byte [r7 + 1] ; int old_pix = src[1]; + movzx r9, byte [r7] ; int old_sum = src[0]; + add r9, r8 ; old_sum += old_pix +.second_loop + movzx r10, byte [r7 + r6] ; int temp1 = src[x]; + lea r11, [r8 + r10] ; int temp2 = old_pix + temp1; + mov r8, r10 ; old_pix = temp1; + lea r10, [r9 + r11] ; temp1 = old_sum + temp2; + mov r9, r11 ; old_sum = temp2; + movzx r11, word [r4 + r6 * 2] ; temp2 = col_pix_buf[x]; + add r11, r10 ; temp2 += temp1; + mov word [r4 + r6 * 2], r10w ; col_pix_buf[x] = temp1; + mov word [r12 + r6 * 2], r11w ; col_sum_buf[x] = temp2; + inc r6 ; x++ + cmp r6, r1 ; x < w + jl .second_loop +.height_loop + mov r10, r5; int tmpreg = y; + imul r10, r3; tmpreg *= stride; + lea r7, [r0 + r10] ; unsigned char *src=buf+y*stride; + sub r10, r3 ; tmpreg -= stride; + lea r13, [r0 + r10]; unsigned char *dst=buf+(y-1)*stride; + mov r6, 2 ; int x = 2; + movzx r10, byte [r7] ; temp1 = src[0]; + movzx r11, byte [r7 + 1] ; temp2 = src[1]; + add r10, r11; temp1 += temp2 + vmovd xmm0, r10d; __m128i old_pix_128 = temp2; + vmovd xmm1, r11d; __m128i old_sum_128 = temp1; +.width_loop + vpermq ymm2, [r7 + r6], 0x10 + vpunpcklbw ymm2, ymm2, ymm6 ; new_pix = _mm_unpacklo_epi8(new_pix, temp3); + vpermq ymm11, ymm2, 0x4e + vpalignr ymm3, ymm2, ymm11, 14 + vpand ymm3, ymm3, ymm8 + vpaddw ymm3, ymm0 ; temp = _mm_add_epi16(temp, old_pix_128); + vpaddw ymm3, ymm2 ; temp = _mm_add_epi16(temp, new_pix); + vperm2i128 ymm0, ymm2, ymm6, 0x21 + vpsrldq ymm0, ymm0, 14; temp = temp >> 14 * 8; + vpermq ymm11, ymm3, 0x4e + vpand ymm11, ymm11, ymm8; + vpalignr ymm2, ymm3, ymm11, 14 + vpand ymm2, ymm2, ymm8 + vpaddw ymm2, ymm1 ; new_pix = _mm_add_epi16(new_pix, old_sum_128); + vpaddw ymm2, ymm3 ; new_pix = _mm_add_epi16(new_pix, temp); + vperm2i128 ymm1, ymm3, ymm6, 0x21 + vpsrldq ymm1, ymm1, 14; temp = temp << 2 * 8; + vmovdqu ymm4, [r4 + r6 * 2] ; __m128i old_col_pix = *(col_pix_buf+x); + vmovdqu [r4 + r6 * 2], ymm2 ; *(col_pix_buf+x) = new_pix ; + vmovdqu ymm5, [r12 + r6 * 2] ; __m128i old_col_sum = *(col_pix_sum+x); + vpaddw ymm3, ymm2, ymm4 + vmovdqu [r12 + r6 * 2], ymm3 ; *(col_sum_buf+x) = temp; + vpaddw ymm5, ymm3 ; old_col_sum = _mm_add_epi16(old_col_sum, temp); + vpsrlw ymm5, 4 ; old_col_sum = old_col_sum >> 4; + vpackuswb ymm5, ymm5 ; old_col_sum = _mm_packus_epi16(old_col_sum, old_col_sum); + vpermq ymm5, ymm5, 11_01_10_00b + vmovdqu [r13 + r6 - 1], xmm5 ; *(dst+x-1) = old_col_sum; + add r6, 16; x += 16; + cmp r6, r14; x < ((w - 2) & (~15)); + jl .width_loop + movzx r8, byte [r7 + r6 - 1] ; old_pix = src[x-1]; + movzx r9, byte [r7 + r6 - 2] ; old_sum = old_pix + src[x-2]; + add r9, r8 +.final_width_loop + movzx r10, byte [r7 + r6] ; temp1 = src[x]; + lea r11, [r8 + r10] ; temp2 = old_pix + temp1; + mov r8, r10 ; old_pix = temp1; + lea r10, [r9 + r11] ; temp1 = old_sum + temp2; + mov r9, r11 ; old_sum = temp2; + movzx r11, word [r4 + r6 * 2] ; temp2 = col_pix_buf[x]; + add r11, r10 ; temp2 += temp1; + mov word [r4 + r6 * 2], r10w ; col_pix_buf[x] = temp1; + movzx r10, word [r12 + r6 * 2] ; temp1 = col_sum_buf[x]; + add r10, r11 ; temp1 += temp2; + shr r10, 4 ; temp1 >>= 4; + mov byte [r13 + r6 - 1], r10b ; dst[x-1] = temp1 + mov [r12 + r6 * 2], r11w ; col_sum_buf[x] = temp2; + inc r6 ; x++ + cmp r6, r1 ; x < w + jl .final_width_loop + inc r5 ; y++; + cmp r5, r2 ; y < h; + jl .height_loop + RET + diff --git a/libass/x86/be_blur.h b/libass/x86/be_blur.h new file mode 100644 index 0000000..2f53ec9 --- /dev/null +++ b/libass/x86/be_blur.h @@ -0,0 +1,30 @@ +/* + * Copyright (C) 2013 Rodger Combs <rcombs@rcombs.me> + * + * This file is part of libass. + * + * Permission to use, copy, modify, and distribute this software for any + * purpose with or without fee is hereby granted, provided that the above + * copyright notice and this permission notice appear in all copies. + * + * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES + * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF + * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR + * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES + * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN + * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF + * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. + */ + +#ifndef INTEL_BE_BLUR_H +#define INTEL_BE_BLUR_H + +void ass_be_blur_sse2( uint8_t *buf, intptr_t width, + intptr_t height, intptr_t stride, + uint16_t *tmp); + +void ass_be_blur_avx2( uint8_t *buf, intptr_t width, + intptr_t height, intptr_t stride, + uint16_t *tmp); + +#endif diff --git a/libass/x86/blend_bitmaps.asm b/libass/x86/blend_bitmaps.asm new file mode 100644 index 0000000..bd8358a --- /dev/null +++ b/libass/x86/blend_bitmaps.asm @@ -0,0 +1,290 @@ +;****************************************************************************** +;* add_bitmaps.asm: SSE2 and x86 add_bitmaps +;****************************************************************************** + +%define HAVE_ALIGNED_STACK 1 +%include "x86inc.asm" + +SECTION_RODATA 32 + +words_255: dw 0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF + +SECTION .text + +;------------------------------------------------------------------------------ +; void add_bitmaps( uint8_t *dst, intptr_t dst_stride, +; uint8_t *src, intptr_t src_stride, +; intptr_t height, intptr_t width ); +;------------------------------------------------------------------------------ + +INIT_XMM +cglobal add_bitmaps_x86, 6,7 +.skip_prologue: + imul r4, r3 + add r4, r2 + PUSH r4 + mov r4, r3 +.height_loop: + xor r6, r6 ; x offset +.stride_loop: + movzx r3, byte [r0 + r6] + add r3b, byte [r2 + r6] + jnc .continue + mov r3b, 0xff +.continue: + mov byte [r0 + r6], r3b + inc r6 + cmp r6, r5 + jl .stride_loop ; still in scan line + add r0, r1 + add r2, r4 + cmp r2, [rsp] + jl .height_loop + ADD rsp, gprsize + RET + +%macro ADD_BITMAPS 0 + cglobal add_bitmaps, 6,7 + .skip_prologue: + cmp r5, mmsize + %if mmsize == 16 + jl add_bitmaps_x86.skip_prologue + %else + jl add_bitmaps_sse2.skip_prologue + %endif + %if mmsize == 32 + vzeroupper + %endif + imul r4, r3 + add r4, r2 ; last address + .height_loop: + xor r6, r6 ; x offset + .stride_loop: + movu m0, [r0 + r6] + paddusb m0, [r2 + r6] + movu [r0 + r6], m0 + add r6, mmsize + cmp r6, r5 + jl .stride_loop ; still in scan line + add r0, r1 + add r2, r3 + cmp r2, r4 + jl .height_loop + RET +%endmacro + +INIT_XMM sse2 +ADD_BITMAPS +INIT_YMM avx2 +ADD_BITMAPS + +;------------------------------------------------------------------------------ +; void sub_bitmaps( uint8_t *dst, intptr_t dst_stride, +; uint8_t *src, intptr_t src_stride, +; intptr_t height, intptr_t width ); +;------------------------------------------------------------------------------ + +INIT_XMM +cglobal sub_bitmaps_x86, 6,10 +.skip_prologue: + imul r4, r3 + add r4, r2 ; last address + PUSH r4 + mov r4, r3 +.height_loop: + xor r6, r6 ; x offset +.stride_loop: + mov r3b, byte [r0 + r6] + sub r3b, byte [r2 + r6] + jnc .continue + mov r3b, 0x0 +.continue: + mov byte [r0 + r6], r3b + inc r6 + cmp r6, r5 + jl .stride_loop ; still in scan line + add r0, r1 + add r2, r4 + cmp r2, [rsp] + jl .height_loop + ADD rsp, gprsize + RET + +%if ARCH_X86_64 + +%macro SUB_BITMAPS 0 + cglobal sub_bitmaps, 6,10 + .skip_prologue: + cmp r5, mmsize + %if mmsize == 16 + jl sub_bitmaps_x86.skip_prologue + %else + jl sub_bitmaps_sse2.skip_prologue + %endif + %if mmsize == 32 + vzeroupper + %endif + imul r4, r3 + add r4, r2 ; last address + mov r7, r5 + and r7, -mmsize ; &= (16); + xor r9, r9 + .height_loop: + xor r6, r6 ; x offset + .stride_loop: + movu m0, [r0 + r6] + movu m1, [r2 + r6] + psubusb m0, m1 + movu [r0 + r6], m0 + add r6, mmsize + cmp r6, r7 + jl .stride_loop ; still in scan line + .stride_loop2 + cmp r6, r5 + jge .finish + movzx r8, byte [r0 + r6] + sub r8b, byte [r2 + r6] + cmovc r8, r9 + mov byte [r0 + r6], r8b + inc r6 + jmp .stride_loop2 + .finish + add r0, r1 + add r2, r3 + cmp r2, r4 + jl .height_loop + RET +%endmacro + +INIT_XMM sse2 +SUB_BITMAPS +INIT_YMM avx2 +SUB_BITMAPS + +;------------------------------------------------------------------------------ +; void mul_bitmaps( uint8_t *dst, intptr_t dst_stride, +; uint8_t *src1, intptr_t src1_stride, +; uint8_t *src2, intptr_t src2_stride, +; intptr_t width, intptr_t height ); +;------------------------------------------------------------------------------ + +INIT_XMM +cglobal mul_bitmaps_x86, 8,12 +.skip_prologue: + imul r7, r3 + add r7, r2 ; last address +.height_loop: + xor r8, r8 ; x offset +.stride_loop: + movzx r9, byte [r2 + r8] + movzx r10, byte [r4 + r8] + imul r9, r10 + add r9, 255 + shr r9, 8 + mov byte [r0 + r8], r9b + inc r8 + cmp r8, r6 + jl .stride_loop ; still in scan line + add r0, r1 + add r2, r3 + add r4, r5 + cmp r2, r7 + jl .height_loop + RET + +INIT_XMM sse2 +cglobal mul_bitmaps, 8,12 +.skip_prologue: + cmp r6, 8 + jl mul_bitmaps_x86.skip_prologue + imul r7, r3 + add r7, r2 ; last address + pxor xmm2, xmm2 + movdqa xmm3, [words_255 wrt rip] + mov r9, r6 + and r9, -8 ; &= (~8); +.height_loop: + xor r8, r8 ; x offset +.stride_loop: + movq xmm0, [r2 + r8] + movq xmm1, [r4 + r8] + punpcklbw xmm0, xmm2 + punpcklbw xmm1, xmm2 + pmullw xmm0, xmm1 + paddw xmm0, xmm3 + psrlw xmm0, 0x08 + packuswb xmm0, xmm0 + movq [r0 + r8], xmm0 + add r8, 8 + cmp r8, r9 + jl .stride_loop ; still in scan line +.stride_loop2 + cmp r8, r6 + jge .finish + movzx r10, byte [r2 + r8] + movzx r11, byte [r4 + r8] + imul r10, r11 + add r10, 255 + shr r10, 8 + mov byte [r0 + r8], r10b + inc r8 + jmp .stride_loop2 +.finish: + add r0, r1 + add r2, r3 + add r4, r5 + cmp r2, r7 + jl .height_loop + RET + +INIT_YMM avx2 +cglobal mul_bitmaps, 8,12 + cmp r6, 16 + jl mul_bitmaps_sse2.skip_prologue + %if mmsize == 32 + vzeroupper + %endif + imul r7, r3 + add r7, r2 ; last address + vpxor ymm2, ymm2 + vmovdqa ymm3, [words_255 wrt rip] + mov r9, r6 + and r9, -16 ; &= (~16); +.height_loop: + xor r8, r8 ; x offset +.stride_loop: + vmovdqu xmm0, [r2 + r8] + vpermq ymm0, ymm0, 0x10 + vmovdqu xmm1, [r4 + r8] + vpermq ymm1, ymm1, 0x10 + vpunpcklbw ymm0, ymm0, ymm2 + vpunpcklbw ymm1, ymm1, ymm2 + vpmullw ymm0, ymm0, ymm1 + vpaddw ymm0, ymm0, ymm3 + vpsrlw ymm0, ymm0, 0x08 + vextracti128 xmm4, ymm0, 0x1 + vpackuswb ymm0, ymm0, ymm4 + vmovdqa [r0 + r8], xmm0 + add r8, 16 + cmp r8, r9 + jl .stride_loop ; still in scan line +.stride_loop2 + cmp r8, r6 + jge .finish + movzx r10, byte [r2 + r8] + movzx r11, byte [r4 + r8] + imul r10, r11 + add r10, 255 + shr r10, 8 + mov byte [r0 + r8], r10b + inc r8 + jmp .stride_loop2 +.finish: + add r0, r1 + add r2, r3 + add r4, r5 + cmp r2, r7 + jl .height_loop + RET + +%endif diff --git a/libass/x86/blend_bitmaps.h b/libass/x86/blend_bitmaps.h new file mode 100644 index 0000000..30058ed --- /dev/null +++ b/libass/x86/blend_bitmaps.h @@ -0,0 +1,56 @@ +/* + * Copyright (C) 2013 Rodger Combs <rcombs@rcombs.me> + * + * This file is part of libass. + * + * Permission to use, copy, modify, and distribute this software for any + * purpose with or without fee is hereby granted, provided that the above + * copyright notice and this permission notice appear in all copies. + * + * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES + * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF + * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR + * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES + * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN + * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF + * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. + */ + +#ifndef INTEL_BLEND_BITMAPS_H +#define INTEL_BLEND_BITMAPS_H + +void ass_add_bitmaps_avx2( uint8_t *dst, intptr_t dst_stride, + uint8_t *src, intptr_t src_stride, + intptr_t height, intptr_t width ); + +void ass_add_bitmaps_sse2( uint8_t *dst, intptr_t dst_stride, + uint8_t *src, intptr_t src_stride, + intptr_t height, intptr_t width ); + +void ass_add_bitmaps_x86( uint8_t *dst, intptr_t dst_stride, + uint8_t *src, intptr_t src_stride, + intptr_t height, intptr_t width ); + +void ass_sub_bitmaps_avx2( uint8_t *dst, intptr_t dst_stride, + uint8_t *src, intptr_t src_stride, + intptr_t height, intptr_t width ); + +void ass_sub_bitmaps_sse2( uint8_t *dst, intptr_t dst_stride, + uint8_t *src, intptr_t src_stride, + intptr_t height, intptr_t width ); + +void ass_sub_bitmaps_x86( uint8_t *dst, intptr_t dst_stride, + uint8_t *src, intptr_t src_stride, + intptr_t height, intptr_t width ); + +void ass_mul_bitmaps_avx2( uint8_t *dst, intptr_t dst_stride, + uint8_t *src1, intptr_t src1_stride, + uint8_t *src2, intptr_t src2_stride, + intptr_t width, intptr_t height ); + +void ass_mul_bitmaps_sse2( uint8_t *dst, intptr_t dst_stride, + uint8_t *src1, intptr_t src1_stride, + uint8_t *src2, intptr_t src2_stride, + intptr_t width, intptr_t height ); + +#endif diff --git a/libass/x86/cpuid.asm b/libass/x86/cpuid.asm new file mode 100644 index 0000000..ca0792c --- /dev/null +++ b/libass/x86/cpuid.asm @@ -0,0 +1,32 @@ +;****************************************************************************** +;* add_bitmaps.asm: SSE2 and x86 add_bitmaps +;****************************************************************************** + +%include "x86inc.asm" + +SECTION .text + +;------------------------------------------------------------------------------ +; void get_cpuid( uint32_t *eax, uint32_t *ebx, uint32_t *ecx, uint32_t *edx); +;------------------------------------------------------------------------------ + +INIT_XMM +cglobal get_cpuid, 4, 5, 0 + push rbx + push r3 + push r2 + push r1 + push r0 + mov eax, [r0] + xor ecx, ecx + cpuid + pop r4 + mov [r4], eax + pop r4 + mov [r4], ebx + pop r4 + mov [r4], ecx + pop r4 + mov [r4], edx + pop rbx + RET diff --git a/libass/x86/cpuid.h b/libass/x86/cpuid.h new file mode 100644 index 0000000..34e4b19 --- /dev/null +++ b/libass/x86/cpuid.h @@ -0,0 +1,24 @@ +/* + * Copyright (C) 2013 Rodger Combs <rcombs@rcombs.me> + * + * This file is part of libass. + * + * Permission to use, copy, modify, and distribute this software for any + * purpose with or without fee is hereby granted, provided that the above + * copyright notice and this permission notice appear in all copies. + * + * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES + * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF + * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR + * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES + * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN + * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF + * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. + */ + +#ifndef INTEL_CPUID_H +#define INTEL_CPUID_H + +void ass_get_cpuid( uint32_t *eax, uint32_t *ebx, uint32_t *ecx, uint32_t *edx); + +#endif diff --git a/libass/x86/x86inc.asm b/libass/x86/x86inc.asm new file mode 100644 index 0000000..53e104d --- /dev/null +++ b/libass/x86/x86inc.asm @@ -0,0 +1,1450 @@ +;***************************************************************************** +;* x86inc.asm: x264asm abstraction layer +;***************************************************************************** +;* Copyright (C) 2005-2013 x264 project +;* +;* Authors: Loren Merritt <lorenm@u.washington.edu> +;* Anton Mitrofanov <BugMaster@narod.ru> +;* Jason Garrett-Glaser <darkshikari@gmail.com> +;* Henrik Gramner <henrik@gramner.com> +;* +;* Permission to use, copy, modify, and/or distribute this software for any +;* purpose with or without fee is hereby granted, provided that the above +;* copyright notice and this permission notice appear in all copies. +;* +;* THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES +;* WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF +;* MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR +;* ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES +;* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN +;* ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF +;* OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. +;***************************************************************************** + +; This is a header file for the x264ASM assembly language, which uses +; NASM/YASM syntax combined with a large number of macros to provide easy +; abstraction between different calling conventions (x86_32, win64, linux64). +; It also has various other useful features to simplify writing the kind of +; DSP functions that are most often used in x264. + +; Unlike the rest of x264, this file is available under an ISC license, as it +; has significant usefulness outside of x264 and we want it to be available +; to the largest audience possible. Of course, if you modify it for your own +; purposes to add a new feature, we strongly encourage contributing a patch +; as this feature might be useful for others as well. Send patches or ideas +; to x264-devel@videolan.org . + +%ifndef private_prefix + %define private_prefix ass +%endif + +%ifndef public_prefix + %define public_prefix private_prefix +%endif |