diff options
| -rwxr-xr-x | build/make/Makefile | 2 | ||||
| -rw-r--r-- | build/make/obj_int_extract.c | 673 | ||||
| -rw-r--r-- | vp8/encoder/firstpass.c | 188 | ||||
| -rw-r--r-- | vp8/encoder/onyx_if.c | 285 | ||||
| -rw-r--r-- | vp8/encoder/onyx_int.h | 1 | ||||
| -rw-r--r-- | vp8/encoder/rdopt.c | 2 | ||||
| -rw-r--r-- | vp8/encoder/x86/variance_impl_sse2.asm | 388 | ||||
| -rw-r--r-- | vp8/encoder/x86/variance_impl_ssse3.asm | 348 | ||||
| -rw-r--r-- | vp8/encoder/x86/variance_mmx.c | 140 | ||||
| -rw-r--r-- | vp8/encoder/x86/variance_sse2.c | 191 | ||||
| -rw-r--r-- | vp8/encoder/x86/variance_ssse3.c | 116 | ||||
| -rw-r--r-- | vp8/encoder/x86/variance_x86.h | 4 | ||||
| -rw-r--r-- | vp8/encoder/x86/x86_csystemdependent.c | 2 | ||||
| -rw-r--r-- | vp8/vp8cx.mk | 2 | ||||
| -rwxr-xr-x | vpxenc.c | 4 |
15 files changed, 1509 insertions, 837 deletions
diff --git a/build/make/Makefile b/build/make/Makefile index 40fa6d50c..62d139ea4 100755 --- a/build/make/Makefile +++ b/build/make/Makefile @@ -153,7 +153,7 @@ endif # obj_int_extract: build/make/obj_int_extract.c $(if $(quiet),echo " [HOSTCC] $@") - $(qexec)$(HOSTCC) -I. -o $@ $< + $(qexec)$(HOSTCC) -I. -I$(SRC_PATH_BARE) -o $@ $< CLEAN-OBJS += obj_int_extract # diff --git a/build/make/obj_int_extract.c b/build/make/obj_int_extract.c index 3c54b248f..22c5cf2ab 100644 --- a/build/make/obj_int_extract.c +++ b/build/make/obj_int_extract.c @@ -14,7 +14,7 @@ #include "vpx_config.h" -#if defined(_MSC_VER) +#if defined(_MSC_VER) || defined(__MINGW32__) #include <io.h> #include <share.h> #include "vpx/vpx_integer.h" @@ -59,20 +59,47 @@ int parse_macho(uint8_t *base_buf, size_t sz) struct mach_header header; uint8_t *buf = base_buf; int base_data_section = 0; - + int bits = 0; + + /* We can read in mach_header for 32 and 64 bit architectures + * because it's identical to mach_header_64 except for the last + * element (uint32_t reserved), which we don't use. Then, when + * we know which architecture we're looking at, increment buf + * appropriately. + */ memcpy(&header, buf, sizeof(struct mach_header)); - buf += sizeof(struct mach_header); - if (header.magic != MH_MAGIC) + if (header.magic == MH_MAGIC) { - log_msg("Bad magic number for object file. 0x%x expected, 0x%x found.\n", - header.magic, MH_MAGIC); - goto bail; + if (header.cputype == CPU_TYPE_ARM + || header.cputype == CPU_TYPE_X86) + { + bits = 32; + buf += sizeof(struct mach_header); + } + else + { + log_msg("Bad cputype for object file. Currently only tested for CPU_TYPE_[ARM|X86].\n"); + goto bail; + } } - - if (header.cputype != CPU_TYPE_ARM) + else if (header.magic == MH_MAGIC_64) + { + if (header.cputype == CPU_TYPE_X86_64) + { + bits = 64; + buf += sizeof(struct mach_header_64); + } + else + { + log_msg("Bad cputype for object file. Currently only tested for CPU_TYPE_X86_64.\n"); + goto bail; + } + } + else { - log_msg("Bad cputype for object file. Currently only tested for CPU_TYPE_ARM.\n"); + log_msg("Bad magic number for object file. 0x%x or 0x%x expected, 0x%x found.\n", + MH_MAGIC, MH_MAGIC_64, header.magic); goto bail; } @@ -85,8 +112,6 @@ int parse_macho(uint8_t *base_buf, size_t sz) for (i = 0; i < header.ncmds; i++) { struct load_command lc; - struct symtab_command sc; - struct segment_command seg_c; memcpy(&lc, buf, sizeof(struct load_command)); @@ -94,50 +119,99 @@ int parse_macho(uint8_t *base_buf, size_t sz) { uint8_t *seg_buf = buf; struct section s; + struct segment_command seg_c; - memcpy(&seg_c, buf, sizeof(struct segment_command)); - + memcpy(&seg_c, seg_buf, sizeof(struct segment_command)); seg_buf += sizeof(struct segment_command); - for (j = 0; j < seg_c.nsects; j++) + /* Although each section is given it's own offset, nlist.n_value + * references the offset of the first section. This isn't + * apparent without debug information because the offset of the + * data section is the same as the first section. However, with + * debug sections mixed in, the offset of the debug section + * increases but n_value still references the first section. + */ + if (seg_c.nsects < 1) { - memcpy(&s, seg_buf + (j * sizeof(struct section)), sizeof(struct section)); + log_msg("Not enough sections\n"); + goto bail; + } + + memcpy(&s, seg_buf, sizeof(struct section)); + base_data_section = s.offset; + } + else if (lc.cmd == LC_SEGMENT_64) + { + uint8_t *seg_buf = buf; + struct section_64 s; + struct segment_command_64 seg_c; + + memcpy(&seg_c, seg_buf, sizeof(struct segment_command_64)); + seg_buf += sizeof(struct segment_command_64); - // Need to get this offset which is the start of the symbol table - // before matching the strings up with symbols. - base_data_section = s.offset; + /* Explanation in LG_SEGMENT */ + if (seg_c.nsects < 1) + { + log_msg("Not enough sections\n"); + goto bail; } + + memcpy(&s, seg_buf, sizeof(struct section_64)); + base_data_section = s.offset; } else if (lc.cmd == LC_SYMTAB) { - uint8_t *sym_buf = base_buf; - uint8_t *str_buf = base_buf; - if (base_data_section != 0) { + struct symtab_command sc; + uint8_t *sym_buf = base_buf; + uint8_t *str_buf = base_buf; + memcpy(&sc, buf, sizeof(struct symtab_command)); if (sc.cmdsize != sizeof(struct symtab_command)) + { log_msg("Can't find symbol table!\n"); + goto bail; + } sym_buf += sc.symoff; str_buf += sc.stroff; for (j = 0; j < sc.nsyms; j++) { - struct nlist nl; - int val; + /* Location of string is cacluated each time from the + * start of the string buffer. On darwin the symbols + * are prefixed by "_", so we bump the pointer by 1. + * The target value is defined as an int in asm_*_offsets.c, + * which is 4 bytes on all targets we currently use. + */ + if (bits == 32) + { + struct nlist nl; + int val; - memcpy(&nl, sym_buf + (j * sizeof(struct nlist)), sizeof(struct nlist)); + memcpy(&nl, sym_buf, sizeof(struct nlist)); + sym_buf += sizeof(struct nlist); - val = *((int *)(base_buf + base_data_section + nl.n_value)); + memcpy(&val, base_buf + base_data_section + nl.n_value, + sizeof(val)); + printf("%-40s EQU %5d\n", + str_buf + nl.n_un.n_strx + 1, val); + } + else /* if (bits == 64) */ + { + struct nlist_64 nl; + int val; - // Location of string is cacluated each time from the - // start of the string buffer. On darwin the symbols - // are prefixed by "_". On other platforms it is not - // so it needs to be removed. That is the reason for - // the +1. - printf("%-40s EQU %5d\n", str_buf + nl.n_un.n_strx + 1, val); + memcpy(&nl, sym_buf, sizeof(struct nlist_64)); + sym_buf += sizeof(struct nlist_64); + + memcpy(&val, base_buf + base_data_section + nl.n_value, + sizeof(val)); + printf("%-40s EQU %5d\n", + str_buf + nl.n_un.n_strx + 1, val); + } } } } @@ -218,7 +292,7 @@ bail: return EXIT_FAILURE; } -#else +#elif defined(__ELF__) #include "elf.h" #define COPY_STRUCT(dst, buf, ofst, sz) do {\ @@ -237,212 +311,420 @@ bail: typedef struct { - uint8_t *buf; /* Buffer containing ELF data */ - size_t sz; /* Buffer size */ - int le_data; /* Data is little-endian */ - Elf32_Ehdr hdr; + uint8_t *buf; /* Buffer containing ELF data */ + size_t sz; /* Buffer size */ + int le_data; /* Data is little-endian */ + unsigned char e_ident[EI_NIDENT]; /* Magic number and other info */ + int bits; /* 32 or 64 */ + Elf32_Ehdr hdr32; + Elf64_Ehdr hdr64; } elf_obj_t; -int parse_elf32_header(elf_obj_t *elf) +int parse_elf_header(elf_obj_t *elf) { int res; - /* Verify ELF32 header */ - COPY_STRUCT(&elf->hdr, elf->buf, 0, elf->sz); - res = elf->hdr.e_ident[EI_MAG0] == ELFMAG0; - res &= elf->hdr.e_ident[EI_MAG1] == ELFMAG1; - res &= elf->hdr.e_ident[EI_MAG2] == ELFMAG2; - res &= elf->hdr.e_ident[EI_MAG3] == ELFMAG3; - res &= elf->hdr.e_ident[EI_CLASS] == ELFCLASS32; - res &= elf->hdr.e_ident[EI_DATA] == ELFDATA2LSB - || elf->hdr.e_ident[EI_DATA] == ELFDATA2MSB; + /* Verify ELF Magic numbers */ + COPY_STRUCT(&elf->e_ident, elf->buf, 0, elf->sz); + res = elf->e_ident[EI_MAG0] == ELFMAG0; + res &= elf->e_ident[EI_MAG1] == ELFMAG1; + res &= elf->e_ident[EI_MAG2] == ELFMAG2; + res &= elf->e_ident[EI_MAG3] == ELFMAG3; + res &= elf->e_ident[EI_CLASS] == ELFCLASS32 + || elf->e_ident[EI_CLASS] == ELFCLASS64; + res &= elf->e_ident[EI_DATA] == ELFDATA2LSB; if (!res) goto bail; - elf->le_data = elf->hdr.e_ident[EI_DATA] == ELFDATA2LSB; - - ENDIAN_ASSIGN_IN_PLACE(elf->hdr.e_type); - ENDIAN_ASSIGN_IN_PLACE(elf->hdr.e_machine); - ENDIAN_ASSIGN_IN_PLACE(elf->hdr.e_version); - ENDIAN_ASSIGN_IN_PLACE(elf->hdr.e_entry); - ENDIAN_ASSIGN_IN_PLACE(elf->hdr.e_phoff); - ENDIAN_ASSIGN_IN_PLACE(elf->hdr.e_shoff); - ENDIAN_ASSIGN_IN_PLACE(elf->hdr.e_flags); - ENDIAN_ASSIGN_IN_PLACE(elf->hdr.e_ehsize); - ENDIAN_ASSIGN_IN_PLACE(elf->hdr.e_phentsize); - ENDIAN_ASSIGN_IN_PLACE(elf->hdr.e_phnum); - ENDIAN_ASSIGN_IN_PLACE(elf->hdr.e_shentsize); - ENDIAN_ASSIGN_IN_PLACE(elf->hdr.e_shnum); - ENDIAN_ASSIGN_IN_PLACE(elf->hdr.e_shstrndx); + elf->le_data = elf->e_ident[EI_DATA] == ELFDATA2LSB; + + /* Read in relevant values */ + if (elf->e_ident[EI_CLASS] == ELFCLASS32) + { + elf->bits = 32; + COPY_STRUCT(&elf->hdr32, elf->buf, 0, elf->sz); + + ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_type); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_machine); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_version); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_entry); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_phoff); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_shoff); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_flags); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_ehsize); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_phentsize); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_phnum); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_shentsize); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_shnum); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr32.e_shstrndx); + } + else /* if (elf->e_ident[EI_CLASS] == ELFCLASS64) */ + { + elf->bits = 64; + COPY_STRUCT(&elf->hdr64, elf->buf, 0, elf->sz); + + ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_type); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_machine); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_version); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_entry); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_phoff); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_shoff); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_flags); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_ehsize); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_phentsize); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_phnum); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_shentsize); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_shnum); + ENDIAN_ASSIGN_IN_PLACE(elf->hdr64.e_shstrndx); + } + return 0; bail: + log_msg("Failed to parse ELF file header"); return 1; } -int parse_elf32_section(elf_obj_t *elf, int idx, Elf32_Shdr *hdr) +int parse_elf_section(elf_obj_t *elf, int idx, Elf32_Shdr *hdr32, Elf64_Shdr *hdr64) { - if (idx >= elf->hdr.e_shnum) - goto bail; + if (hdr32) + { + if (idx >= elf->hdr32.e_shnum) + goto bail; + + COPY_STRUCT(hdr32, elf->buf, elf->hdr32.e_shoff + idx * elf->hdr32.e_shentsize, + elf->sz); + ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_name); + ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_type); + ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_flags); + ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_addr); + ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_offset); + ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_size); + ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_link); + ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_info); + ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_addralign); + ENDIAN_ASSIGN_IN_PLACE(hdr32->sh_entsize); + } + else /* if (hdr64) */ + { + if (idx >= elf->hdr64.e_shnum) + goto bail; + + COPY_STRUCT(hdr64, elf->buf, elf->hdr64.e_shoff + idx * elf->hdr64.e_shentsize, + elf->sz); + ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_name); + ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_type); + ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_flags); + ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_addr); + ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_offset); + ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_size); + ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_link); + ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_info); + ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_addralign); + ENDIAN_ASSIGN_IN_PLACE(hdr64->sh_entsize); + } - COPY_STRUCT(hdr, elf->buf, elf->hdr.e_shoff + idx * elf->hdr.e_shentsize, - elf->sz); - ENDIAN_ASSIGN_IN_PLACE(hdr->sh_name); - ENDIAN_ASSIGN_IN_PLACE(hdr->sh_type); - ENDIAN_ASSIGN_IN_PLACE(hdr->sh_flags); - ENDIAN_ASSIGN_IN_PLACE(hdr->sh_addr); - ENDIAN_ASSIGN_IN_PLACE(hdr->sh_offset); - ENDIAN_ASSIGN_IN_PLACE(hdr->sh_size); - ENDIAN_ASSIGN_IN_PLACE(hdr->sh_link); - ENDIAN_ASSIGN_IN_PLACE(hdr->sh_info); - ENDIAN_ASSIGN_IN_PLACE(hdr->sh_addralign); - ENDIAN_ASSIGN_IN_PLACE(hdr->sh_entsize); return 0; bail: return 1; } -char *parse_elf32_string_table(elf_obj_t *elf, int s_idx, int idx) +char *parse_elf_string_table(elf_obj_t *elf, int s_idx, int idx) { - Elf32_Shdr shdr; - - if (parse_elf32_section(elf, s_idx, &shdr)) + if (elf->bits == 32) { - log_msg("Failed to parse ELF string table: section %d, index %d\n", - s_idx, idx); - return ""; + Elf32_Shdr shdr; + + if (parse_elf_section(elf, s_idx, &shdr, NULL)) + { + log_msg("Failed to parse ELF string table: section %d, index %d\n", + s_idx, idx); + return ""; + } + + return (char *)(elf->buf + shdr.sh_offset + idx); } + else /* if (elf->bits == 64) */ + { + Elf64_Shdr shdr; - return (char *)(elf->buf + shdr.sh_offset + idx); + if (parse_elf_section(elf, s_idx, NULL, &shdr)) + { + log_msg("Failed to parse ELF string table: section %d, index %d\n", + s_idx, idx); + return ""; + } + + return (char *)(elf->buf + shdr.sh_offset + idx); + } } -int parse_elf32_symbol(elf_obj_t *elf, unsigned int ofst, Elf32_Sym *sym) +int parse_elf_symbol(elf_obj_t *elf, unsigned int ofst, Elf32_Sym *sym32, Elf64_Sym *sym64) { - COPY_STRUCT(sym, elf->buf, ofst, elf->sz); - ENDIAN_ASSIGN_IN_PLACE(sym->st_name); - ENDIAN_ASSIGN_IN_PLACE(sym->st_value); - ENDIAN_ASSIGN_IN_PLACE(sym->st_size); - ENDIAN_ASSIGN_IN_PLACE(sym->st_info); - ENDIAN_ASSIGN_IN_PLACE(sym->st_other); - ENDIAN_ASSIGN_IN_PLACE(sym->st_shndx); + if (sym32) + { + COPY_STRUCT(sym32, elf->buf, ofst, elf->sz); + ENDIAN_ASSIGN_IN_PLACE(sym32->st_name); + ENDIAN_ASSIGN_IN_PLACE(sym32->st_value); + ENDIAN_ASSIGN_IN_PLACE(sym32->st_size); + ENDIAN_ASSIGN_IN_PLACE(sym32->st_info); + ENDIAN_ASSIGN_IN_PLACE(sym32->st_other); + ENDIAN_ASSIGN_IN_PLACE(sym32->st_shndx); + } + else /* if (sym64) */ + { + COPY_STRUCT(sym64, elf->buf, ofst, elf->sz); + ENDIAN_ASSIGN_IN_PLACE(sym64->st_name); + ENDIAN_ASSIGN_IN_PLACE(sym64->st_value); + ENDIAN_ASSIGN_IN_PLACE(sym64->st_size); + ENDIAN_ASSIGN_IN_PLACE(sym64->st_info); + ENDIAN_ASSIGN_IN_PLACE(sym64->st_other); + ENDIAN_ASSIGN_IN_PLACE(sym64->st_shndx); + } return 0; bail: return 1; } -int parse_elf32(uint8_t *buf, size_t sz, output_fmt_t mode) +int parse_elf(uint8_t *buf, size_t sz, output_fmt_t mode) { - elf_obj_t elf; - Elf32_Shdr shdr; + elf_obj_t elf; unsigned int ofst; - int i; - Elf32_Off strtab_off; /* save String Table offset for later use */ + int i; + Elf32_Off strtab_off32; + Elf64_Off strtab_off64; /* save String Table offset for later use */ memset(&elf, 0, sizeof(elf)); elf.buf = buf; elf.sz = sz; /* Parse Header */ - if (parse_elf32_header(&elf)) - { - log_msg("Parse error: File does not appear to be valid ELF32\n"); - return 1; - } + if (parse_elf_header(&elf)) + goto bail; - for (i = 0; i < elf.hdr.e_shnum; i++) + if (elf.bits == 32) { - parse_elf32_section(&elf, i, &shdr); - - if (shdr.sh_type == SHT_STRTAB) + Elf32_Shdr shdr; + for (i = 0; i < elf.hdr32.e_shnum; i++) { - char strtsb_name[128]; - - strcpy(strtsb_name, (char *)(elf.buf + shdr.sh_offset + shdr.sh_name)); + parse_elf_section(&elf, i, &shdr, NULL); - if (!(strcmp(strtsb_name, ".shstrtab"))) + if (shdr.sh_type == SHT_STRTAB) { - log_msg("found section: %s\n", strtsb_name); - strtab_off = shdr.sh_offset; - break; + char strtsb_name[128]; + + strcpy(strtsb_name, (char *)(elf.buf + shdr.sh_offset + shdr.sh_name)); + + if (!(strcmp(strtsb_name, ".shstrtab"))) + { + /* log_msg("found section: %s\n", strtsb_name); */ + strtab_off32 = shdr.sh_offset; + break; + } } } } - - /* Parse all Symbol Tables */ - for (i = 0; i < elf.hdr.e_shnum; i++) + else /* if (elf.bits == 64) */ { - - parse_elf32_section(&elf, i, &shdr); - - if (shdr.sh_type == SHT_SYMTAB) + Elf64_Shdr shdr; + for (i = 0; i < elf.hdr64.e_shnum; i++) { - for (ofst = shdr.sh_offset; - ofst < shdr.sh_offset + shdr.sh_size; - ofst += shdr.sh_entsize) - { - Elf32_Sym sym; + parse_elf_section(&elf, i, NULL, &shdr); - parse_elf32_symbol(&elf, ofst, &sym); + if (shdr.sh_type == SHT_STRTAB) + { + char strtsb_name[128]; - /* For all OBJECTS (data objects), extract the value from the - * proper data segment. - */ - if (ELF32_ST_TYPE(sym.st_info) == STT_OBJECT && sym.st_name) - log_msg("found data object %s\n", - parse_elf32_string_table(&elf, - shdr.sh_link, - sym.st_name)); + strcpy(strtsb_name, (char *)(elf.buf + shdr.sh_offset + shdr.sh_name)); - if (ELF32_ST_TYPE(sym.st_info) == STT_OBJECT - && sym.st_size == 4) + if (!(strcmp(strtsb_name, ".shstrtab"))) { - Elf32_Shdr dhdr; - int32_t val; - char section_name[128]; - - parse_elf32_section(&elf, sym.st_shndx, &dhdr); + /* log_msg("found section: %s\n", strtsb_name); */ + strtab_off64 = shdr.sh_offset; + break; + } + } + } + } - /* For explanition - refer to _MSC_VER version of code */ - strcpy(section_name, (char *)(elf.buf + strtab_off + dhdr.sh_name)); - log_msg("Section_name: %s, Section_type: %d\n", section_name, dhdr.sh_type); + /* Parse all Symbol Tables */ + if (elf.bits == 32) + { + Elf32_Shdr shdr; + for (i = 0; i < elf.hdr32.e_shnum; i++) + { + parse_elf_section(&elf, i, &shdr, NULL); - if (!(strcmp(section_name, ".bss"))) - { - val = 0; - } - else + if (shdr.sh_type == SHT_SYMTAB) + { + for (ofst = shdr.sh_offset; + ofst < shdr.sh_offset + shdr.sh_size; + ofst += shdr.sh_entsize) + { + Elf32_Sym sym; + + parse_elf_symbol(&elf, ofst, &sym, NULL); + + /* For all OBJECTS (data objects), extract the value from the + * proper data segment. + */ + /* if (ELF32_ST_TYPE(sym.st_info) == STT_OBJECT && sym.st_name) + log_msg("found data object %s\n", + parse_elf_string_table(&elf, + shdr.sh_link, + sym.st_name)); + */ + + if (ELF32_ST_TYPE(sym.st_info) == STT_OBJECT + && sym.st_size == 4) { - memcpy(&val, - elf.buf + dhdr.sh_offset + sym.st_value, - sizeof(val)); + Elf32_Shdr dhdr; + int val = 0; + char section_name[128]; + + parse_elf_section(&elf, sym.st_shndx, &dhdr, NULL); + + /* For explanition - refer to _MSC_VER version of code */ + strcpy(section_name, (char *)(elf.buf + strtab_off32 + dhdr.sh_name)); + /* log_msg("Section_name: %s, Section_type: %d\n", section_name, dhdr.sh_type); */ + + if (strcmp(section_name, ".bss")) + { + if (sizeof(val) != sym.st_size) + { + /* The target value is declared as an int in + * asm_*_offsets.c, which is 4 bytes on all + * targets we currently use. Complain loudly if + * this is not true. + */ + log_msg("Symbol size is wrong\n"); + goto bail; + } + + memcpy(&val, + elf.buf + dhdr.sh_offset + sym.st_value, + sym.st_size); + } + + if (!elf.le_data) + { + log_msg("Big Endian data not supported yet!\n"); + goto bail; + } + + switch (mode) + { + case OUTPUT_FMT_RVDS: + printf("%-40s EQU %5d\n", + parse_elf_string_table(&elf, + shdr.sh_link, + sym.st_name), + val); + break; + case OUTPUT_FMT_GAS: + printf(".equ %-40s, %5d\n", + parse_elf_string_table(&elf, + shdr.sh_link, + sym.st_name), + val); + break; + default: + printf("%s = %d\n", + parse_elf_string_table(&elf, + shdr.sh_link, + sym.st_name), + val); + } } + } + } + } + } + else /* if (elf.bits == 64) */ + { + Elf64_Shdr shdr; + for (i = 0; i < elf.hdr64.e_shnum; i++) + { + parse_elf_section(&elf, i, NULL, &shdr); - if (!elf.le_data) - { - log_msg("Big Endian data not supported yet!\n"); - goto bail; - }\ - - switch (mode) + if (shdr.sh_type == SHT_SYMTAB) + { + for (ofst = shdr.sh_offset; + ofst < shdr.sh_offset + shdr.sh_size; + ofst += shdr.sh_entsize) + { + Elf64_Sym sym; + + parse_elf_symbol(&elf, ofst, NULL, &sym); + + /* For all OBJECTS (data objects), extract the value from the + * proper data segment. + */ + /* if (ELF64_ST_TYPE(sym.st_info) == STT_OBJECT && sym.st_name) + log_msg("found data object %s\n", + parse_elf_string_table(&elf, + shdr.sh_link, + sym.st_name)); + */ + + if (ELF64_ST_TYPE(sym.st_info) == STT_OBJECT + && sym.st_size == 4) { - case OUTPUT_FMT_RVDS: - printf("%-40s EQU %5d\n", - parse_elf32_string_table(&elf, - shdr.sh_link, - sym.st_name), - val); - break; - case OUTPUT_FMT_GAS: - printf(".equ %-40s, %5d\n", - parse_elf32_string_table(&elf, - shdr.sh_link, - sym.st_name), - val); - break; - default: - printf("%s = %d\n", - parse_elf32_string_table(&elf, - shdr.sh_link, - sym.st_name), - val); + Elf64_Shdr dhdr; + int val = 0; + char section_name[128]; + + parse_elf_section(&elf, sym.st_shndx, NULL, &dhdr); + + /* For explanition - refer to _MSC_VER version of code */ + strcpy(section_name, (char *)(elf.buf + strtab_off64 + dhdr.sh_name)); + /* log_msg("Section_name: %s, Section_type: %d\n", section_name, dhdr.sh_type); */ + + if ((strcmp(section_name, ".bss"))) + { + if (sizeof(val) != sym.st_size) + { + /* The target value is declared as an int in + * asm_*_offsets.c, which is 4 bytes on all + * targets we currently use. Complain loudly if + * this is not true. + */ + log_msg("Symbol size is wrong\n"); + goto bail; + } + + memcpy(&val, + elf.buf + dhdr.sh_offset + sym.st_value, + sym.st_size); + } + + if (!elf.le_data) + { + log_msg("Big Endian data not supported yet!\n"); + goto bail; + } + + switch (mode) + { + case OUTPUT_FMT_RVDS: + printf("%-40s EQU %5d\n", + parse_elf_string_table(&elf, + shdr.sh_link, + sym.st_name), + val); + break; + case OUTPUT_FMT_GAS: + printf(".equ %-40s, %5d\n", + parse_elf_string_table(&elf, + shdr.sh_link, + sym.st_name), + val); + break; + default: + printf("%s = %d\n", + parse_elf_string_table(&elf, + shdr.sh_link, + sym.st_name), + val); + } } } } @@ -454,7 +736,7 @@ int parse_elf32(uint8_t *buf, size_t sz, output_fmt_t mode) return 0; bail: - log_msg("Parse error: File does not appear to be valid ELF32\n"); + log_msg("Parse error: File does not appear to be valid ELF32 or ELF64\n"); return 1; } @@ -521,8 +803,7 @@ int main(int argc, char **argv) goto bail; } - res = parse_elf32(file_buf, stat_buf.st_size, mode); - //res = parse_coff(file_buf, stat_buf.st_size); + res = parse_elf(file_buf, stat_buf.st_size, mode); free(file_buf); if (!res) @@ -535,7 +816,7 @@ bail: #endif -#if defined(_MSC_VER) +#if defined(_MSC_VER) || defined(__MINGW32__) /* See "Microsoft Portable Executable and Common Object File Format Specification" for reference. */ @@ -549,7 +830,6 @@ int parse_coff(unsigned __int8 *buf, size_t sz) unsigned int i; unsigned __int8 *ptr; unsigned __int32 symoffset; - FILE *fp; char **sectionlist; //this array holds all section names in their correct order. //it is used to check if the symbol is in .bss or .data section. @@ -590,14 +870,6 @@ int parse_coff(unsigned __int8 *buf, size_t sz) //log_msg("COFF: Symbol table at offset %u\n", symtab_ptr); //log_msg("COFF: raw data pointer ofset for section .data is %u\n", sectionrawdata_ptr); - fp = fopen("assembly_offsets.asm", "w"); - - if (fp == NULL) - { - perror("open file"); - goto bail; - } - /* The compiler puts the data with non-zero offset in .data section, but puts the data with zero offset in .bss section. So, if the data in in .bss section, set offset=0. Note from Wiki: In an object module compiled from C, the bss section contains @@ -631,13 +903,13 @@ int parse_coff(unsigned __int8 *buf, size_t sz) char name[9] = {0, 0, 0, 0, 0, 0, 0, 0, 0}; strncpy(name, ptr, 8); //log_msg("COFF: Parsing symbol %s\n",name); - fprintf(fp, "%-40s EQU ", name); + printf("%-40s EQU ", name + 1); } else { //log_msg("COFF: Parsing symbol %s\n", // buf + strtab_ptr + get_le32(ptr+4)); - fprintf(fp, "%-40s EQU ", buf + strtab_ptr + get_le32(ptr + 4)); + printf("%-40s EQU ", buf + strtab_ptr + get_le32(ptr + 4) + 1); } if (!(strcmp(sectionlist[section-1], ".bss"))) @@ -654,14 +926,13 @@ int parse_coff(unsigned __int8 *buf, size_t sz) //log_msg(" Address: %u\n",get_le32(ptr+8)); //log_msg(" Offset: %u\n", symoffset); - fprintf(fp, "%5d\n", symoffset); + printf("%5d\n", symoffset); } ptr += 18; } - fprintf(fp, " END\n"); - fclose(fp); + printf(" END\n"); for (i = 0; i < nsections; i++) { @@ -711,11 +982,7 @@ int main(int argc, char **argv) else f = argv[1]; - if (_sopen_s(&fd, f, _O_BINARY, _SH_DENYNO, _S_IREAD | _S_IWRITE)) - { - perror("Unable to open file"); - goto bail; - } + fd = _sopen(f, _O_BINARY, _SH_DENYNO, _S_IREAD | _S_IWRITE); if (_fstat(fd, &stat_buf)) { diff --git a/vp8/encoder/firstpass.c b/vp8/encoder/firstpass.c index e0b2a8a45..d2cc8482e 100644 --- a/vp8/encoder/firstpass.c +++ b/vp8/encoder/firstpass.c @@ -312,7 +312,9 @@ void vp8_output_stats(const VP8_COMP *cpi, FILE *fpfile; fpfile = fopen("firstpass.stt", "a"); - fprintf(fpfile, "%12.0f %12.0f %12.0f %12.4f %12.4f %12.4f %12.4f %12.4f %12.4f %12.4f %12.4f %12.4f %12.4f %12.4f %12.0f\n", + fprintf(fpfile, "%12.0f %12.0f %12.0f %12.4f %12.4f %12.4f %12.4f" + " %12.4f %12.4f %12.4f %12.4f %12.4f %12.4f %12.4f %12.0f" + " %12.4f\n", stats->frame, stats->intra_error, stats->coded_error, @@ -320,6 +322,7 @@ void vp8_output_stats(const VP8_COMP *cpi, stats->pcnt_inter, stats->pcnt_motion, stats->pcnt_second_ref, + stats->pcnt_neutral, stats->MVr, stats->mvr_abs, stats->MVc, @@ -327,7 +330,8 @@ void vp8_output_stats(const VP8_COMP *cpi, stats->MVrv, stats->MVcv, stats->mv_in_out_count, - stats->count); + stats->count, + stats->duration); fclose(fpfile); @@ -359,6 +363,7 @@ void vp8_zero_stats(FIRSTPASS_STATS *section) section->pcnt_inter = 0.0; section->pcnt_motion = 0.0; section->pcnt_second_ref = 0.0; + section->pcnt_neutral = 0.0; section->MVr = 0.0; section->mvr_abs = 0.0; section->MVc = 0.0; @@ -378,6 +383,7 @@ void vp8_accumulate_stats(FIRSTPASS_STATS *section, FIRSTPASS_STATS *frame) section->pcnt_inter += frame->pcnt_inter; section->pcnt_motion += frame->pcnt_motion; section->pcnt_second_ref += frame->pcnt_second_ref; + section->pcnt_neutral += frame->pcnt_neutral; section->MVr += frame->MVr; section->mvr_abs += frame->mvr_abs; section->MVc += frame->MVc; @@ -398,6 +404,7 @@ void vp8_avg_stats(FIRSTPASS_STATS *section) section->ssim_weighted_pred_err /= section->count; section->pcnt_inter /= section->count; section->pcnt_second_ref /= section->count; + section->pcnt_neutral /= section->count; section->pcnt_motion /= section->count; section->MVr /= section->count; section->mvr_abs /= section->count; @@ -570,6 +577,7 @@ void vp8_first_pass(VP8_COMP *cpi) int intercount = 0; int second_ref_count = 0; int intrapenalty = 256; + int neutral_count = 0; int sum_in_vectors = 0; @@ -726,6 +734,17 @@ void vp8_first_pass(VP8_COMP *cpi) if (motion_error <= this_error) { + // Keep a count of cases where the inter and intra were + // very close and very low. This helps with scene cut + // detection for example in cropped clips with black bars + // at the sides or top and bottom. + if( (((this_error-intrapenalty) * 9) <= + (motion_error*10)) && + (this_error < (2*intrapenalty)) ) + { + neutral_count++; + } + d->bmi.mv.as_mv.row <<= 3; d->bmi.mv.as_mv.col <<= 3; this_error = motion_error; @@ -854,6 +873,7 @@ void vp8_first_pass(VP8_COMP *cpi) fps.pcnt_inter = 1.0 * (double)intercount / cm->MBs; fps.pcnt_second_ref = 1.0 * (double)second_ref_count / cm->MBs; + fps.pcnt_neutral = 1.0 * (double)neutral_count / cm->MBs; if (mvcount > 0) { @@ -1341,7 +1361,7 @@ void vp8_end_second_pass(VP8_COMP *cpi) // This function gives and estimate of how badly we believe // the predicition quality is decaying from frame to frame. -double gf_prediction_decay_rate(VP8_COMP *cpi, FIRSTPASS_STATS *next_frame) +double get_prediction_decay_rate(VP8_COMP *cpi, FIRSTPASS_STATS *next_frame) { double prediction_decay_rate; double motion_decay; @@ -1376,6 +1396,52 @@ double gf_prediction_decay_rate(VP8_COMP *cpi, FIRSTPASS_STATS *next_frame) return prediction_decay_rate; } +// Function to test for a condition where a complex transition is followed +// by a static section. For example in slide shows where there is a fade +// between slides. This is to help with more optimal kf and gf positioning. +BOOL detect_transition_to_still( + VP8_COMP *cpi, + int frame_interval, + int still_interval, + double loop_decay_rate, + double decay_accumulator ) +{ + BOOL trans_to_still = FALSE; + + // Break clause to detect very still sections after motion + // For example a static image after a fade or other transition + // instead of a clean scene cut. + if ( (frame_interval > MIN_GF_INTERVAL) && + (loop_decay_rate >= 0.999) && + (decay_accumulator < 0.9) ) + { + int j; + FIRSTPASS_STATS * position = cpi->stats_in; + FIRSTPASS_STATS tmp_next_frame; + double decay_rate; + + // Look ahead a few frames to see if static condition + // persists... + for ( j = 0; j < still_interval; j++ ) + { + if (EOF == vp8_input_stats(cpi, &tmp_next_frame)) + break; + + decay_rate = get_prediction_decay_rate(cpi, &tmp_next_frame); + if ( decay_rate < 0.999 ) + break; + } + // Reset file position + reset_fpf_position(cpi, position); + + // Only if it does do we signal a transition to still + if ( j == still_interval ) + trans_to_still = TRUE; + } + + return trans_to_still; +} + // Analyse and define a gf/arf group . static void define_gf_group(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) { @@ -1528,7 +1594,7 @@ static void define_gf_group(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) if (r > GF_RMAX) r = GF_RMAX; - loop_decay_rate = gf_prediction_decay_rate(cpi, &next_frame); + loop_decay_rate = get_prediction_decay_rate(cpi, &next_frame); // Cumulative effect of decay decay_accumulator = decay_accumulator * loop_decay_rate; @@ -1537,48 +1603,13 @@ static void define_gf_group(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) boost_score += (decay_accumulator * r); // Break clause to detect very still sections after motion - // For example a staic image after a fade or other transition - // instead of a clean key frame. - if ( (i > MIN_GF_INTERVAL) && - (loop_decay_rate >= 0.999) && - (decay_accumulator < 0.9) ) + // For example a staic image after a fade or other transition. + if ( detect_transition_to_still( cpi, i, 5, + loop_decay_rate, decay_accumulator ) ) { - int j; - FIRSTPASS_STATS * position = cpi->stats_in; - FIRSTPASS_STATS tmp_next_frame; - double decay_rate; - - // Look ahead a few frames to see if static condition - // persists... - for ( j = 0; j < 4; j++ ) - { - if (EOF == vp8_input_stats(cpi, &tmp_next_frame)) - break; - - decay_rate = gf_prediction_decay_rate(cpi, &tmp_next_frame); - if ( decay_rate < 0.999 ) - break; - } - reset_fpf_position(cpi, position); // Reset file position - - // Force GF not alt ref - if ( j == 4 ) - { - if (0) - { - FILE *f = fopen("fadegf.stt", "a"); - fprintf(f, " %8d %8d %10.4f %10.4f %10.4f\n", - cpi->common.current_video_frame+i, i, - loop_decay_rate, decay_accumulator, - boost_score ); - fclose(f); - } - - allow_alt_ref = FALSE; - - boost_score = old_boost_score; - break; - } + allow_alt_ref = FALSE; + boost_score = old_boost_score; + break; } // Break out conditions. @@ -2285,7 +2316,7 @@ static BOOL test_candidate_kf(VP8_COMP *cpi, FIRSTPASS_STATS *last_frame, FIRST (next_frame->pcnt_second_ref < 0.10) && ((this_frame->pcnt_inter < 0.05) || ( - (this_frame->pcnt_inter < .25) && + ((this_frame->pcnt_inter - this_frame->pcnt_neutral) < .25) && ((this_frame->intra_error / DOUBLE_DIVIDE_CHECK(this_frame->coded_error)) < 2.5) && ((fabs(last_frame->coded_error - this_frame->coded_error) / DOUBLE_DIVIDE_CHECK(this_frame->coded_error) > .40) || (fabs(last_frame->intra_error - this_frame->intra_error) / DOUBLE_DIVIDE_CHECK(this_frame->intra_error) > .40) || @@ -2332,7 +2363,9 @@ static BOOL test_candidate_kf(VP8_COMP *cpi, FIRSTPASS_STATS *last_frame, FIRST // Test various breakout clauses if ((local_next_frame.pcnt_inter < 0.05) || (next_iiratio < 1.5) || - ((local_next_frame.pcnt_inter < 0.20) && (next_iiratio < 3.0)) || + (((local_next_frame.pcnt_inter - + local_next_frame.pcnt_neutral) < 0.20) && + (next_iiratio < 3.0)) || ((boost_score - old_boost_score) < 0.5) || (local_next_frame.intra_error < 200) ) @@ -2363,13 +2396,13 @@ static BOOL test_candidate_kf(VP8_COMP *cpi, FIRSTPASS_STATS *last_frame, FIRST } void vp8_find_next_key_frame(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) { - int i; + int i,j; FIRSTPASS_STATS last_frame; FIRSTPASS_STATS first_frame; FIRSTPASS_STATS next_frame; FIRSTPASS_STATS *start_position; - double decay_accumulator = 0; + double decay_accumulator = 1.0; double boost_score = 0; double old_boost_score = 0.0; double loop_decay_rate; @@ -2379,6 +2412,7 @@ void vp8_find_next_key_frame(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) double kf_group_intra_err = 0.0; double kf_group_coded_err = 0.0; double two_pass_min_rate = (double)(cpi->oxcf.target_bandwidth * cpi->oxcf.two_pass_vbrmin_section / 100); + double recent_loop_decay[8] = {1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0}; vpx_memset(&next_frame, 0, sizeof(next_frame)); // assure clean @@ -2407,6 +2441,7 @@ void vp8_find_next_key_frame(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) kf_mod_err = calculate_modified_err(cpi, this_frame); // find the next keyframe + i = 0; while (cpi->stats_in < cpi->stats_in_end) { // Accumulate kf group error @@ -2425,9 +2460,34 @@ void vp8_find_next_key_frame(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) if (cpi->oxcf.auto_key && lookup_next_frame_stats(cpi, &next_frame) != EOF) { + // Normal scene cut check if (test_candidate_kf(cpi, &last_frame, this_frame, &next_frame)) break; + // How fast is prediction quality decaying + loop_decay_rate = get_prediction_decay_rate(cpi, &next_frame); + + // We want to know something about the recent past... rather than + // as used elsewhere where we are concened with decay in prediction + // quality since the last GF or KF. + recent_loop_decay[i%8] = loop_decay_rate; + decay_accumulator = 1.0; + for (j = 0; j < 8; j++) + { + decay_accumulator = decay_accumulator * recent_loop_decay[j]; + } + + // Special check for transition or high motion followed by a + // to a static scene. + if ( detect_transition_to_still( cpi, i, + (cpi->key_frame_frequency-i), + loop_decay_rate, + decay_accumulator ) ) + { + break; + } + + // Step on to the next frame cpi->frames_to_key ++; @@ -2437,6 +2497,8 @@ void vp8_find_next_key_frame(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) break; } else cpi->frames_to_key ++; + + i++; } // If there is a max kf interval set by the user we must obey it. @@ -2588,32 +2650,8 @@ void vp8_find_next_key_frame(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) if (r > RMAX) r = RMAX; - // Adjust loop decay rate - //if ( next_frame.pcnt_inter < loop_decay_rate ) - loop_decay_rate = next_frame.pcnt_inter; - - // High % motion -> somewhat higher decay rate - motion_pct = next_frame.pcnt_motion; - motion_decay = (1.0 - (motion_pct / 20.0)); - if (motion_decay < loop_decay_rate) - loop_decay_rate = motion_decay; - - // Adjustment to decay rate based on speed of motion - { - double this_mv_rabs; - double this_mv_cabs; - double distance_factor; - - this_mv_rabs = fabs(next_frame.mvr_abs * motion_pct); - this_mv_cabs = fabs(next_frame.mvc_abs * motion_pct); - - distance_factor = sqrt((this_mv_rabs * this_mv_rabs) + - (this_mv_cabs * this_mv_cabs)) / 250.0; - distance_factor = ((distance_factor > 1.0) - ? 0.0 : (1.0 - distance_factor)); - if (distance_factor < loop_decay_rate) - loop_decay_rate = distance_factor; - } + // How fast is prediction quality decaying + loop_decay_rate = get_prediction_decay_rate(cpi, &next_frame); decay_accumulator = decay_accumulator * loop_decay_rate; decay_accumulator = decay_accumulator < 0.1 ? 0.1 : decay_accumulator; diff --git a/vp8/encoder/onyx_if.c b/vp8/encoder/onyx_if.c index 39610a73f..797e18b30 100644 --- a/vp8/encoder/onyx_if.c +++ b/vp8/encoder/onyx_if.c @@ -1517,252 +1517,29 @@ void vp8_init_config(VP8_PTR ptr, VP8_CONFIG *oxcf) cpi->oxcf = *oxcf; - switch (cpi->oxcf.Mode) - { - - case MODE_REALTIME: - cpi->pass = 0; - cpi->compressor_speed = 2; - - if (cpi->oxcf.cpu_used < -16) - { - cpi->oxcf.cpu_used = -16; - } - - if (cpi->oxcf.cpu_used > 16) - cpi->oxcf.cpu_used = 16; - - break; - -#if !(CONFIG_REALTIME_ONLY) - case MODE_GOODQUALITY: - cpi->pass = 0; - cpi->compressor_speed = 1; - - if (cpi->oxcf.cpu_used < -5) - { - cpi->oxcf.cpu_used = -5; - } - - if (cpi->oxcf.cpu_used > 5) - cpi->oxcf.cpu_used = 5; - - break; - - case MODE_BESTQUALITY: - cpi->pass = 0; - cpi->compressor_speed = 0; - break; - - case MODE_FIRSTPASS: - cpi->pass = 1; - cpi->compressor_speed = 1; - break; - case MODE_SECONDPASS: - cpi->pass = 2; - cpi->compressor_speed = 1; - - if (cpi->oxcf.cpu_used < -5) - { - cpi->oxcf.cpu_used = -5; - } - - if (cpi->oxcf.cpu_used > 5) - cpi->oxcf.cpu_used = 5; - - break; - case MODE_SECONDPASS_BEST: - cpi->pass = 2; - cpi->compressor_speed = 0; - break; -#endif - } - - if (cpi->pass == 0) - cpi->auto_worst_q = 1; - - cpi->oxcf.worst_allowed_q = q_trans[oxcf->worst_allowed_q]; - cpi->oxcf.best_allowed_q = q_trans[oxcf->best_allowed_q]; - cpi->oxcf.cq_level = q_trans[cpi->oxcf.cq_level]; - - if (oxcf->fixed_q >= 0) - { - if (oxcf->worst_allowed_q < 0) - cpi->oxcf.fixed_q = q_trans[0]; - else - cpi->oxcf.fixed_q = q_trans[oxcf->worst_allowed_q]; - - if (oxcf->alt_q < 0) - cpi->oxcf.alt_q = q_trans[0]; - else - cpi->oxcf.alt_q = q_trans[oxcf->alt_q]; - - if (oxcf->key_q < 0) - cpi->oxcf.key_q = q_trans[0]; - else - cpi->oxcf.key_q = q_trans[oxcf->key_q]; - - if (oxcf->gold_q < 0) - cpi->oxcf.gold_q = q_trans[0]; - else - cpi->oxcf.gold_q = q_trans[oxcf->gold_q]; - - } - - cpi->baseline_gf_interval = cpi->oxcf.alt_freq ? cpi->oxcf.alt_freq : DEFAULT_GF_INTERVAL; - cpi->ref_frame_flags = VP8_ALT_FLAG | VP8_GOLD_FLAG | VP8_LAST_FLAG; - - //cpi->use_golden_frame_only = 0; - //cpi->use_last_frame_only = 0; - cm->refresh_golden_frame = 0; - cm->refresh_last_frame = 1; - cm->refresh_entropy_probs = 1; - - if (cpi->oxcf.token_partitions >= 0 && cpi->oxcf.token_partitions <= 3) - cm->multi_token_partition = (TOKEN_PARTITION) cpi->oxcf.token_partitions; - - setup_features(cpi); - - { - int i; - - for (i = 0; i < MAX_MB_SEGMENTS; i++) - cpi->segment_encode_breakout[i] = cpi->oxcf.encode_breakout; - } - - // At the moment the first order values may not be > MAXQ - if (cpi->oxcf.fixed_q > MAXQ) - cpi->oxcf.fixed_q = MAXQ; - - // local file playback mode == really big buffer - if (cpi->oxcf.end_usage == USAGE_LOCAL_FILE_PLAYBACK) - { - cpi->oxcf.starting_buffer_level = 60000; - cpi->oxcf.optimal_buffer_level = 60000; - cpi->oxcf.maximum_buffer_size = 240000; - - } - - // Convert target bandwidth from Kbit/s to Bit/s cpi->oxcf.target_bandwidth *= 1000; cpi->oxcf.starting_buffer_level = rescale(cpi->oxcf.starting_buffer_level, cpi->oxcf.target_bandwidth, 1000); - if (cpi->oxcf.optimal_buffer_level == 0) - cpi->oxcf.optimal_buffer_level = cpi->oxcf.target_bandwidth / 8; - else - cpi->oxcf.optimal_buffer_level = - rescale(cpi->oxcf.optimal_buffer_level, - cpi->oxcf.target_bandwidth, 1000); - - if (cpi->oxcf.maximum_buffer_size == 0) - cpi->oxcf.maximum_buffer_size = cpi->oxcf.target_bandwidth / 8; - else - cpi->oxcf.maximum_buffer_size = - rescale(cpi->oxcf.maximum_buffer_size, - cpi->oxcf.target_bandwidth, 1000); - - cpi->buffer_level = cpi->oxcf.starting_buffer_level; + cpi->buffer_level = cpi->oxcf.starting_buffer_level; cpi->bits_off_target = cpi->oxcf.starting_buffer_level; - vp8_new_frame_rate(cpi, cpi->oxcf.frame_rate); - cpi->worst_quality = cpi->oxcf.worst_allowed_q; cpi->active_worst_quality = cpi->oxcf.worst_allowed_q; - cpi->avg_frame_qindex = cpi->oxcf.worst_allowed_q; - cpi->best_quality = cpi->oxcf.best_allowed_q; cpi->active_best_quality = cpi->oxcf.best_allowed_q; - cpi->cq_target_quality = cpi->oxcf.cq_level; - - cpi->buffered_mode = (cpi->oxcf.optimal_buffer_level > 0) ? TRUE : FALSE; + cpi->avg_frame_qindex = cpi->oxcf.worst_allowed_q; cpi->rolling_target_bits = cpi->av_per_frame_bandwidth; cpi->rolling_actual_bits = cpi->av_per_frame_bandwidth; - cpi->long_rolling_target_bits = cpi->av_per_frame_bandwidth; - cpi->long_rolling_actual_bits = cpi->av_per_frame_bandwidth; + cpi->long_rolling_target_bits = cpi->av_per_frame_bandwidth; + cpi->long_rolling_actual_bits = cpi->av_per_frame_bandwidth; cpi->total_actual_bits = 0; - cpi->total_target_vs_actual = 0; - - // Only allow dropped frames in buffered mode - cpi->drop_frames_allowed = cpi->oxcf.allow_df && cpi->buffered_mode; + cpi->total_target_vs_actual = 0; - cm->filter_type = (LOOPFILTERTYPE) cpi->filter_type; - - if (!cm->use_bilinear_mc_filter) - cm->mcomp_filter_type = SIXTAP; - else - cm->mcomp_filter_type = BILINEAR; - - cpi->target_bandwidth = cpi->oxcf.target_bandwidth; - - cm->Width = cpi->oxcf.Width ; - cm->Height = cpi->oxcf.Height ; - - cpi->intra_frame_target = (4 * (cm->Width + cm->Height) / 15) * 1000; // As per VP8 - - cm->horiz_scale = cpi->horiz_scale; - cm->vert_scale = cpi->vert_scale ; - - // VP8 sharpness level mapping 0-7 (vs 0-10 in general VPx dialogs) - if (cpi->oxcf.Sharpness > 7) - cpi->oxcf.Sharpness = 7; - - cm->sharpness_level = cpi->oxcf.Sharpness; - - if (cm->horiz_scale != NORMAL || cm->vert_scale != NORMAL) - { - int UNINITIALIZED_IS_SAFE(hr), UNINITIALIZED_IS_SAFE(hs); - int UNINITIALIZED_IS_SAFE(vr), UNINITIALIZED_IS_SAFE(vs); - - Scale2Ratio(cm->horiz_scale, &hr, &hs); - Scale2Ratio(cm->vert_scale, &vr, &vs); - - // always go to the next whole number - cm->Width = (hs - 1 + cpi->oxcf.Width * hr) / hs; - cm->Height = (vs - 1 + cpi->oxcf.Height * vr) / vs; - } - - if (((cm->Width + 15) & 0xfffffff0) != cm->yv12_fb[cm->lst_fb_idx].y_width || - ((cm->Height + 15) & 0xfffffff0) != cm->yv12_fb[cm->lst_fb_idx].y_height || - cm->yv12_fb[cm->lst_fb_idx].y_width == 0) - { - alloc_raw_frame_buffers(cpi); - vp8_alloc_compressor_data(cpi); - } - - // Clamp KF frame size to quarter of data rate - if (cpi->intra_frame_target > cpi->target_bandwidth >> 2) - cpi->intra_frame_target = cpi->target_bandwidth >> 2; - - if (cpi->oxcf.fixed_q >= 0) - { - cpi->last_q[0] = cpi->oxcf.fixed_q; - cpi->last_q[1] = cpi->oxcf.fixed_q; - } - - cpi->Speed = cpi->oxcf.cpu_used; - - // force to allowlag to 0 if lag_in_frames is 0; - if (cpi->oxcf.lag_in_frames == 0) - { - cpi->oxcf.allow_lag = 0; - } - // Limit on lag buffers as these are not currently dynamically allocated - else if (cpi->oxcf.lag_in_frames > MAX_LAG_BUFFERS) - cpi->oxcf.lag_in_frames = MAX_LAG_BUFFERS; - - // YX Temp - cpi->last_alt_ref_sei = -1; - cpi->is_src_frame_alt_ref = 0; - cpi->is_next_src_alt_ref = 0; - -#if 0 - // Experimental RD Code - cpi->frame_distortion = 0; - cpi->last_frame_distortion = 0; -#endif + // change includes all joint functionality + vp8_change_config(ptr, oxcf); #if VP8_TEMPORAL_ALT_REF @@ -1779,12 +1556,6 @@ void vp8_init_config(VP8_PTR ptr, VP8_CONFIG *oxcf) #endif } -/* - * This function needs more clean up, i.e. be more tuned torwards - * change_config rather than init_config !!!!!!!!!!!!!!!! - * YX - 5/28/2009 - * - */ void vp8_change_config(VP8_PTR ptr, VP8_CONFIG *oxcf) { @@ -1935,10 +1706,6 @@ void vp8_change_config(VP8_PTR ptr, VP8_CONFIG *oxcf) // Convert target bandwidth from Kbit/s to Bit/s cpi->oxcf.target_bandwidth *= 1000; - cpi->oxcf.starting_buffer_level = - rescale(cpi->oxcf.starting_buffer_level, - cpi->oxcf.target_bandwidth, 1000); - if (cpi->oxcf.optimal_buffer_level == 0) cpi->oxcf.optimal_buffer_level = cpi->oxcf.target_bandwidth / 8; else @@ -1953,29 +1720,36 @@ void vp8_change_config(VP8_PTR ptr, VP8_CONFIG *oxcf) rescale(cpi->oxcf.maximum_buffer_size, cpi->oxcf.target_bandwidth, 1000); - cpi->buffer_level = cpi->oxcf.starting_buffer_level; - cpi->bits_off_target = cpi->oxcf.starting_buffer_level; - vp8_new_frame_rate(cpi, cpi->oxcf.frame_rate); cpi->worst_quality = cpi->oxcf.worst_allowed_q; - cpi->active_worst_quality = cpi->oxcf.worst_allowed_q; - cpi->avg_frame_qindex = cpi->oxcf.worst_allowed_q; cpi->best_quality = cpi->oxcf.best_allowed_q; - cpi->active_best_quality = cpi->oxcf.best_allowed_q; - cpi->buffered_mode = (cpi->oxcf.optimal_buffer_level > 0) ? TRUE : FALSE; - cpi->cq_target_quality = cpi->oxcf.cq_level; + // active values should only be modified if out of new range + if (cpi->active_worst_quality > cpi->oxcf.worst_allowed_q) + { + cpi->active_worst_quality = cpi->oxcf.worst_allowed_q; + } + // less likely + else if (cpi->active_worst_quality < cpi->oxcf.best_allowed_q) + { + cpi->active_worst_quality = cpi->oxcf.best_allowed_q; + } + if (cpi->active_best_quality < cpi->oxcf.best_allowed_q) + { + cpi->active_best_quality = cpi->oxcf.best_allowed_q; + } + // less likely + else if (cpi->active_best_quality > cpi->oxcf.worst_allowed_q) + { + cpi->active_best_quality = cpi->oxcf.worst_allowed_q; + } - cpi->rolling_target_bits = cpi->av_per_frame_bandwidth; - cpi->rolling_actual_bits = cpi->av_per_frame_bandwidth; - cpi->long_rolling_target_bits = cpi->av_per_frame_bandwidth; - cpi->long_rolling_actual_bits = cpi->av_per_frame_bandwidth; + cpi->buffered_mode = (cpi->oxcf.optimal_buffer_level > 0) ? TRUE : FALSE; - cpi->total_actual_bits = 0; - cpi->total_target_vs_actual = 0; + cpi->cq_target_quality = cpi->oxcf.cq_level; // Only allow dropped frames in buffered mode - cpi->drop_frames_allowed = cpi->oxcf.allow_df && cpi->buffered_mode; + cpi->drop_frames_allowed = cpi->oxcf.allow_df && cpi->buffered_mode; cm->filter_type = (LOOPFILTERTYPE) cpi->filter_type; @@ -3542,6 +3316,7 @@ static void encode_frame_to_data_rate int drop_mark50 = drop_mark / 4; int drop_mark25 = drop_mark / 8; + // Clear down mmx registers to allow floating point in what follows vp8_clear_system_state(); diff --git a/vp8/encoder/onyx_int.h b/vp8/encoder/onyx_int.h index b66131d15..6d0cbd9fc 100644 --- a/vp8/encoder/onyx_int.h +++ b/vp8/encoder/onyx_int.h @@ -99,6 +99,7 @@ typedef struct double pcnt_inter; double pcnt_motion; double pcnt_second_ref; + double pcnt_neutral; double MVr; double mvr_abs; double MVc; diff --git a/vp8/encoder/rdopt.c b/vp8/encoder/rdopt.c index 8aaca0917..b0dcfe0a4 100644 --- a/vp8/encoder/rdopt.c +++ b/vp8/encoder/rdopt.c @@ -1968,7 +1968,7 @@ int vp8_rd_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, int else cpi->zbin_mode_boost = LF_ZEROMV_ZBIN_BOOST; } - else if (vp8_ref_frame_order[mode_index] == SPLITMV) + else if (vp8_mode_order[mode_index] == SPLITMV) cpi->zbin_mode_boost = 0; else cpi->zbin_mode_boost = MV_ZBIN_BOOST; diff --git a/vp8/encoder/x86/variance_impl_sse2.asm b/vp8/encoder/x86/variance_impl_sse2.asm index 6cdc47bc9..c2c30deb2 100644 --- a/vp8/encoder/x86/variance_impl_sse2.asm +++ b/vp8/encoder/x86/variance_impl_sse2.asm @@ -627,6 +627,10 @@ filter_block2d_bil_var_sse2_loop: filter_block2d_bil_var_sse2_sp_only: movsxd rdx, dword ptr arg(6) ; yoffset + + cmp rdx, 0 ; skip all if both xoffset=0 and yoffset=0 + je filter_block2d_bil_var_sse2_full_pixel + shl rdx, 5 lea rdx, [rdx + rcx] ; VFilter @@ -671,6 +675,35 @@ filter_block2d_bil_sp_only_loop: jmp filter_block2d_bil_variance +filter_block2d_bil_var_sse2_full_pixel: + mov rsi, arg(0) ;ref_ptr + mov rdi, arg(2) ;src_ptr + movsxd rcx, dword ptr arg(4) ;Height + movsxd rax, dword ptr arg(1) ;ref_pixels_per_line + movsxd rbx, dword ptr arg(3) ;src_pixels_per_line + pxor xmm0, xmm0 ; + +filter_block2d_bil_full_pixel_loop: + movq xmm1, QWORD PTR [rsi] ; + punpcklbw xmm1, xmm0 ; + + movq xmm2, QWORD PTR [rdi] ; + punpcklbw xmm2, xmm0 ; + + psubw xmm1, xmm2 ; + paddw xmm6, xmm1 ; + + pmaddwd xmm1, xmm1 ; + paddd xmm7, xmm1 ; + + lea rsi, [rsi + rax] ;ref_pixels_per_line + lea rdi, [rdi + rbx] ;src_pixels_per_line + + sub rcx, 1 ; + jnz filter_block2d_bil_full_pixel_loop ; + + jmp filter_block2d_bil_variance + filter_block2d_bil_var_sse2_fp_only: mov rsi, arg(0) ;ref_ptr mov rdi, arg(2) ;src_ptr @@ -757,7 +790,7 @@ filter_block2d_bil_variance: ret -;void vp8_half_horiz_vert_variance16x_h_sse2 +;void vp8_half_horiz_vert_variance8x_h_sse2 ;( ; unsigned char *ref_ptr, ; int ref_pixels_per_line, @@ -767,8 +800,8 @@ filter_block2d_bil_variance: ; int *sum, ; unsigned int *sumsquared ;) -global sym(vp8_half_horiz_vert_variance16x_h_sse2) -sym(vp8_half_horiz_vert_variance16x_h_sse2): +global sym(vp8_half_horiz_vert_variance8x_h_sse2) +sym(vp8_half_horiz_vert_variance8x_h_sse2): push rbp mov rbp, rsp SHADOW_ARGS_TO_STACK 7 @@ -802,7 +835,7 @@ sym(vp8_half_horiz_vert_variance16x_h_sse2): add rsi, r8 %endif -vp8_half_horiz_vert_variance16x_h_1: +vp8_half_horiz_vert_variance8x_h_1: movq xmm1, QWORD PTR [rsi] ; movq xmm2, QWORD PTR [rsi+1] ; @@ -830,7 +863,7 @@ vp8_half_horiz_vert_variance16x_h_1: %endif sub rcx, 1 ; - jnz vp8_half_horiz_vert_variance16x_h_1 ; + jnz vp8_half_horiz_vert_variance8x_h_1 ; movdq2q mm6, xmm6 ; movdq2q mm7, xmm7 ; @@ -877,8 +910,123 @@ vp8_half_horiz_vert_variance16x_h_1: pop rbp ret +;void vp8_half_horiz_vert_variance16x_h_sse2 +;( +; unsigned char *ref_ptr, +; int ref_pixels_per_line, +; unsigned char *src_ptr, +; int src_pixels_per_line, +; unsigned int Height, +; int *sum, +; unsigned int *sumsquared +;) +global sym(vp8_half_horiz_vert_variance16x_h_sse2) +sym(vp8_half_horiz_vert_variance16x_h_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 7 + SAVE_XMM + GET_GOT rbx + push rsi + push rdi + ; end prolog -;void vp8_half_vert_variance16x_h_sse2 + pxor xmm6, xmm6 ; error accumulator + pxor xmm7, xmm7 ; sse eaccumulator + mov rsi, arg(0) ;ref_ptr ; + + mov rdi, arg(2) ;src_ptr ; + movsxd rcx, dword ptr arg(4) ;Height ; + movsxd rax, dword ptr arg(1) ;ref_pixels_per_line + movsxd rdx, dword ptr arg(3) ;src_pixels_per_line + + pxor xmm0, xmm0 ; + + movdqu xmm5, XMMWORD PTR [rsi] + movdqu xmm3, XMMWORD PTR [rsi+1] + pavgb xmm5, xmm3 ; xmm5 = avg(xmm1,xmm3) horizontal line 1 + + lea rsi, [rsi + rax] + +vp8_half_horiz_vert_variance16x_h_1: + movdqu xmm1, XMMWORD PTR [rsi] ; + movdqu xmm2, XMMWORD PTR [rsi+1] ; + pavgb xmm1, xmm2 ; xmm1 = avg(xmm1,xmm3) horizontal line i+1 + + pavgb xmm5, xmm1 ; xmm = vertical average of the above + + movdqa xmm4, xmm5 + punpcklbw xmm5, xmm0 ; xmm5 = words of above + punpckhbw xmm4, xmm0 + + movq xmm3, QWORD PTR [rdi] ; xmm3 = d0,d1,d2..d7 + punpcklbw xmm3, xmm0 ; xmm3 = words of above + psubw xmm5, xmm3 ; xmm5 -= xmm3 + + movq xmm3, QWORD PTR [rdi+8] + punpcklbw xmm3, xmm0 + psubw xmm4, xmm3 + + paddw xmm6, xmm5 ; xmm6 += accumulated column differences + paddw xmm6, xmm4 + pmaddwd xmm5, xmm5 ; xmm5 *= xmm5 + pmaddwd xmm4, xmm4 + paddd xmm7, xmm5 ; xmm7 += accumulated square column differences + paddd xmm7, xmm4 + + movdqa xmm5, xmm1 ; save xmm1 for use on the next row + + lea rsi, [rsi + rax] + lea rdi, [rdi + rdx] + + sub rcx, 1 ; + jnz vp8_half_horiz_vert_variance16x_h_1 ; + + pxor xmm1, xmm1 + pxor xmm5, xmm5 + + punpcklwd xmm0, xmm6 + punpckhwd xmm1, xmm6 + psrad xmm0, 16 + psrad xmm1, 16 + paddd xmm0, xmm1 + movdqa xmm1, xmm0 + + movdqa xmm6, xmm7 + punpckldq xmm6, xmm5 + punpckhdq xmm7, xmm5 + paddd xmm6, xmm7 + + punpckldq xmm0, xmm5 + punpckhdq xmm1, xmm5 + paddd xmm0, xmm1 + + movdqa xmm7, xmm6 + movdqa xmm1, xmm0 + + psrldq xmm7, 8 + psrldq xmm1, 8 + + paddd xmm6, xmm7 + paddd xmm0, xmm1 + + mov rsi, arg(5) ;[Sum] + mov rdi, arg(6) ;[SSE] + + movd [rsi], xmm0 + movd [rdi], xmm6 + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_half_vert_variance8x_h_sse2 ;( ; unsigned char *ref_ptr, ; int ref_pixels_per_line, @@ -888,8 +1036,8 @@ vp8_half_horiz_vert_variance16x_h_1: ; int *sum, ; unsigned int *sumsquared ;) -global sym(vp8_half_vert_variance16x_h_sse2) -sym(vp8_half_vert_variance16x_h_sse2): +global sym(vp8_half_vert_variance8x_h_sse2) +sym(vp8_half_vert_variance8x_h_sse2): push rbp mov rbp, rsp SHADOW_ARGS_TO_STACK 7 @@ -912,7 +1060,7 @@ sym(vp8_half_vert_variance16x_h_sse2): movsxd rax, dword ptr arg(1) ;ref_pixels_per_line pxor xmm0, xmm0 ; -vp8_half_vert_variance16x_h_1: +vp8_half_vert_variance8x_h_1: movq xmm5, QWORD PTR [rsi] ; xmm5 = s0,s1,s2..s8 movq xmm3, QWORD PTR [rsi+rax] ; xmm3 = s1,s2,s3..s9 @@ -936,7 +1084,7 @@ vp8_half_vert_variance16x_h_1: %endif sub rcx, 1 ; - jnz vp8_half_vert_variance16x_h_1 ; + jnz vp8_half_vert_variance8x_h_1 ; movdq2q mm6, xmm6 ; movdq2q mm7, xmm7 ; @@ -983,8 +1131,115 @@ vp8_half_vert_variance16x_h_1: pop rbp ret +;void vp8_half_vert_variance16x_h_sse2 +;( +; unsigned char *ref_ptr, +; int ref_pixels_per_line, +; unsigned char *src_ptr, +; int src_pixels_per_line, +; unsigned int Height, +; int *sum, +; unsigned int *sumsquared +;) +global sym(vp8_half_vert_variance16x_h_sse2) +sym(vp8_half_vert_variance16x_h_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 7 + SAVE_XMM + GET_GOT rbx + push rsi + push rdi + ; end prolog + + pxor xmm6, xmm6 ; error accumulator + pxor xmm7, xmm7 ; sse eaccumulator + mov rsi, arg(0) ;ref_ptr + + mov rdi, arg(2) ;src_ptr + movsxd rcx, dword ptr arg(4) ;Height + movsxd rax, dword ptr arg(1) ;ref_pixels_per_line + movsxd rdx, dword ptr arg(3) ;src_pixels_per_line + + movdqu xmm5, XMMWORD PTR [rsi] + lea rsi, [rsi + rax ] + pxor xmm0, xmm0 + +vp8_half_vert_variance16x_h_1: + movdqu xmm3, XMMWORD PTR [rsi] + + pavgb xmm5, xmm3 ; xmm5 = avg(xmm1,xmm3) + movdqa xmm4, xmm5 + punpcklbw xmm5, xmm0 + punpckhbw xmm4, xmm0 + + movq xmm2, QWORD PTR [rdi] + punpcklbw xmm2, xmm0 + psubw xmm5, xmm2 + movq xmm2, QWORD PTR [rdi+8] + punpcklbw xmm2, xmm0 + psubw xmm4, xmm2 + + paddw xmm6, xmm5 ; xmm6 += accumulated column differences + paddw xmm6, xmm4 + pmaddwd xmm5, xmm5 ; xmm5 *= xmm5 + pmaddwd xmm4, xmm4 + paddd xmm7, xmm5 ; xmm7 += accumulated square column differences + paddd xmm7, xmm4 + + movdqa xmm5, xmm3 + + lea rsi, [rsi + rax] + lea rdi, [rdi + rdx] + + sub rcx, 1 + jnz vp8_half_vert_variance16x_h_1 + + pxor xmm1, xmm1 + pxor xmm5, xmm5 + + punpcklwd xmm0, xmm6 + punpckhwd xmm1, xmm6 + psrad xmm0, 16 + psrad xmm1, 16 + paddd xmm0, xmm1 + movdqa xmm1, xmm0 + + movdqa xmm6, xmm7 + punpckldq xmm6, xmm5 + punpckhdq xmm7, xmm5 + paddd xmm6, xmm7 + + punpckldq xmm0, xmm5 + punpckhdq xmm1, xmm5 + paddd xmm0, xmm1 + + movdqa xmm7, xmm6 + movdqa xmm1, xmm0 + + psrldq xmm7, 8 + psrldq xmm1, 8 + + paddd xmm6, xmm7 + paddd xmm0, xmm1 + + mov rsi, arg(5) ;[Sum] + mov rdi, arg(6) ;[SSE] + + movd [rsi], xmm0 + movd [rdi], xmm6 + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + -;void vp8_half_horiz_variance16x_h_sse2 +;void vp8_half_horiz_variance8x_h_sse2 ;( ; unsigned char *ref_ptr, ; int ref_pixels_per_line, @@ -994,8 +1249,8 @@ vp8_half_vert_variance16x_h_1: ; int *sum, ; unsigned int *sumsquared ;) -global sym(vp8_half_horiz_variance16x_h_sse2) -sym(vp8_half_horiz_variance16x_h_sse2): +global sym(vp8_half_horiz_variance8x_h_sse2) +sym(vp8_half_horiz_variance8x_h_sse2): push rbp mov rbp, rsp SHADOW_ARGS_TO_STACK 7 @@ -1017,7 +1272,7 @@ sym(vp8_half_horiz_variance16x_h_sse2): movsxd rcx, dword ptr arg(4) ;Height ; pxor xmm0, xmm0 ; -vp8_half_horiz_variance16x16_1: +vp8_half_horiz_variance8x_h_1: movq xmm5, QWORD PTR [rsi] ; xmm5 = s0,s1,s2..s8 movq xmm3, QWORD PTR [rsi+1] ; xmm3 = s1,s2,s3..s9 @@ -1040,7 +1295,7 @@ vp8_half_horiz_variance16x16_1: add rdi, r9 %endif sub rcx, 1 ; - jnz vp8_half_horiz_variance16x16_1 ; + jnz vp8_half_horiz_variance8x_h_1 ; movdq2q mm6, xmm6 ; movdq2q mm7, xmm7 ; @@ -1087,6 +1342,109 @@ vp8_half_horiz_variance16x16_1: pop rbp ret +;void vp8_half_horiz_variance16x_h_sse2 +;( +; unsigned char *ref_ptr, +; int ref_pixels_per_line, +; unsigned char *src_ptr, +; int src_pixels_per_line, +; unsigned int Height, +; int *sum, +; unsigned int *sumsquared +;) +global sym(vp8_half_horiz_variance16x_h_sse2) +sym(vp8_half_horiz_variance16x_h_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 7 + SAVE_XMM + GET_GOT rbx + push rsi + push rdi + ; end prolog + + pxor xmm6, xmm6 ; error accumulator + pxor xmm7, xmm7 ; sse eaccumulator + mov rsi, arg(0) ;ref_ptr ; + + mov rdi, arg(2) ;src_ptr ; + movsxd rcx, dword ptr arg(4) ;Height ; + movsxd rax, dword ptr arg(1) ;ref_pixels_per_line + movsxd rdx, dword ptr arg(3) ;src_pixels_per_line + + pxor xmm0, xmm0 ; + +vp8_half_horiz_variance16x_h_1: + movdqu xmm5, XMMWORD PTR [rsi] ; xmm5 = s0,s1,s2..s15 + movdqu xmm3, XMMWORD PTR [rsi+1] ; xmm3 = s1,s2,s3..s16 + + pavgb xmm5, xmm3 ; xmm5 = avg(xmm1,xmm3) + movdqa xmm1, xmm5 + punpcklbw xmm5, xmm0 ; xmm5 = words of above + punpckhbw xmm1, xmm0 + + movq xmm3, QWORD PTR [rdi] ; xmm3 = d0,d1,d2..d7 + punpcklbw xmm3, xmm0 ; xmm3 = words of above + movq xmm2, QWORD PTR [rdi+8] + punpcklbw xmm2, xmm0 + + psubw xmm5, xmm3 ; xmm5 -= xmm3 + psubw xmm1, xmm2 + paddw xmm6, xmm5 ; xmm6 += accumulated column differences + paddw xmm6, xmm1 + pmaddwd xmm5, xmm5 ; xmm5 *= xmm5 + pmaddwd xmm1, xmm1 + paddd xmm7, xmm5 ; xmm7 += accumulated square column differences + paddd xmm7, xmm1 + + lea rsi, [rsi + rax] + lea rdi, [rdi + rdx] + + sub rcx, 1 ; + jnz vp8_half_horiz_variance16x_h_1 ; + + pxor xmm1, xmm1 + pxor xmm5, xmm5 + + punpcklwd xmm0, xmm6 + punpckhwd xmm1, xmm6 + psrad xmm0, 16 + psrad xmm1, 16 + paddd xmm0, xmm1 + movdqa xmm1, xmm0 + + movdqa xmm6, xmm7 + punpckldq xmm6, xmm5 + punpckhdq xmm7, xmm5 + paddd xmm6, xmm7 + + punpckldq xmm0, xmm5 + punpckhdq xmm1, xmm5 + paddd xmm0, xmm1 + + movdqa xmm7, xmm6 + movdqa xmm1, xmm0 + + psrldq xmm7, 8 + psrldq xmm1, 8 + + paddd xmm6, xmm7 + paddd xmm0, xmm1 + + mov rsi, arg(5) ;[Sum] + mov rdi, arg(6) ;[SSE] + + movd [rsi], xmm0 + movd [rdi], xmm6 + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret SECTION_RODATA ; short xmm_bi_rd[8] = { 64, 64, 64, 64,64, 64, 64, 64}; diff --git a/vp8/encoder/x86/variance_impl_ssse3.asm b/vp8/encoder/x86/variance_impl_ssse3.asm new file mode 100644 index 000000000..b1976328d --- /dev/null +++ b/vp8/encoder/x86/variance_impl_ssse3.asm @@ -0,0 +1,348 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + +%include "vpx_ports/x86_abi_support.asm" + +%define xmm_filter_shift 7 + + +;void vp8_filter_block2d_bil_var_ssse3 +;( +; unsigned char *ref_ptr, +; int ref_pixels_per_line, +; unsigned char *src_ptr, +; int src_pixels_per_line, +; unsigned int Height, +; int xoffset, +; int yoffset, +; int *sum, +; unsigned int *sumsquared;; +; +;) +;Note: The filter coefficient at offset=0 is 128. Since the second register +;for Pmaddubsw is signed bytes, we must calculate zero offset seperately. +global sym(vp8_filter_block2d_bil_var_ssse3) +sym(vp8_filter_block2d_bil_var_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 9 + SAVE_XMM + GET_GOT rbx + push rsi + push rdi + push rbx + ; end prolog + + pxor xmm6, xmm6 + pxor xmm7, xmm7 + + lea rcx, [GLOBAL(vp8_bilinear_filters_ssse3)] + movsxd rax, dword ptr arg(5) ; xoffset + + cmp rax, 0 ; skip first_pass filter if xoffset=0 + je filter_block2d_bil_var_ssse3_sp_only + + shl rax, 4 ; point to filter coeff with xoffset + lea rax, [rax + rcx] ; HFilter + + movsxd rdx, dword ptr arg(6) ; yoffset + + cmp rdx, 0 ; skip second_pass filter if yoffset=0 + je filter_block2d_bil_var_ssse3_fp_only + + shl rdx, 4 + lea rdx, [rdx + rcx] ; VFilter + + mov rsi, arg(0) ;ref_ptr + mov rdi, arg(2) ;src_ptr + movsxd rcx, dword ptr arg(4) ;Height + + movdqu xmm0, XMMWORD PTR [rsi] + movdqu xmm1, XMMWORD PTR [rsi+1] + movdqa xmm2, xmm0 + + punpcklbw xmm0, xmm1 + punpckhbw xmm2, xmm1 + pmaddubsw xmm0, [rax] + pmaddubsw xmm2, [rax] + + paddw xmm0, [GLOBAL(xmm_bi_rd)] + paddw xmm2, [GLOBAL(xmm_bi_rd)] + psraw xmm0, xmm_filter_shift + psraw xmm2, xmm_filter_shift + + packuswb xmm0, xmm2 + + movsxd rbx, dword ptr arg(1) ;ref_pixels_per_line + lea rsi, [rsi + rbx] +%if ABI_IS_32BIT=0 + movsxd r9, dword ptr arg(3) ;src_pixels_per_line +%endif + +filter_block2d_bil_var_ssse3_loop: + movdqu xmm1, XMMWORD PTR [rsi] + movdqu xmm2, XMMWORD PTR [rsi+1] + movdqa xmm3, xmm1 + + punpcklbw xmm1, xmm2 + punpckhbw xmm3, xmm2 + pmaddubsw xmm1, [rax] + pmaddubsw xmm3, [rax] + + paddw xmm1, [GLOBAL(xmm_bi_rd)] + paddw xmm3, [GLOBAL(xmm_bi_rd)] + psraw xmm1, xmm_filter_shift + psraw xmm3, xmm_filter_shift + packuswb xmm1, xmm3 + + movdqa xmm2, xmm0 + movdqa xmm0, xmm1 + movdqa xmm3, xmm2 + + punpcklbw xmm2, xmm1 + punpckhbw xmm3, xmm1 + pmaddubsw xmm2, [rdx] + pmaddubsw xmm3, [rdx] + + paddw xmm2, [GLOBAL(xmm_bi_rd)] + paddw xmm3, [GLOBAL(xmm_bi_rd)] + psraw xmm2, xmm_filter_shift + psraw xmm3, xmm_filter_shift + + movq xmm1, QWORD PTR [rdi] + pxor xmm4, xmm4 + punpcklbw xmm1, xmm4 + movq xmm5, QWORD PTR [rdi+8] + punpcklbw xmm5, xmm4 + + psubw xmm2, xmm1 + psubw xmm3, xmm5 + paddw xmm6, xmm2 + paddw xmm6, xmm3 + pmaddwd xmm2, xmm2 + pmaddwd xmm3, xmm3 + paddd xmm7, xmm2 + paddd xmm7, xmm3 + + lea rsi, [rsi + rbx] ;ref_pixels_per_line +%if ABI_IS_32BIT + add rdi, dword ptr arg(3) ;src_pixels_per_line +%else + lea rdi, [rdi + r9] +%endif + + sub rcx, 1 + jnz filter_block2d_bil_var_ssse3_loop + + jmp filter_block2d_bil_variance + +filter_block2d_bil_var_ssse3_sp_only: + movsxd rdx, dword ptr arg(6) ; yoffset + + cmp rdx, 0 ; Both xoffset =0 and yoffset=0 + je filter_block2d_bil_var_ssse3_full_pixel + + shl rdx, 4 + lea rdx, [rdx + rcx] ; VFilter + + mov rsi, arg(0) ;ref_ptr + mov rdi, arg(2) ;src_ptr + movsxd rcx, dword ptr arg(4) ;Height + movsxd rax, dword ptr arg(1) ;ref_pixels_per_line + + movdqu xmm1, XMMWORD PTR [rsi] + movdqa xmm0, xmm1 + + movsxd rbx, dword ptr arg(3) ;src_pixels_per_line + lea rsi, [rsi + rax] + +filter_block2d_bil_sp_only_loop: + movdqu xmm3, XMMWORD PTR [rsi] + movdqa xmm2, xmm1 + movdqa xmm0, xmm3 + + punpcklbw xmm1, xmm3 + punpckhbw xmm2, xmm3 + pmaddubsw xmm1, [rdx] + pmaddubsw xmm2, [rdx] + + paddw xmm1, [GLOBAL(xmm_bi_rd)] + paddw xmm2, [GLOBAL(xmm_bi_rd)] + psraw xmm1, xmm_filter_shift + psraw xmm2, xmm_filter_shift + + movq xmm3, QWORD PTR [rdi] + pxor xmm4, xmm4 + punpcklbw xmm3, xmm4 + movq xmm5, QWORD PTR [rdi+8] + punpcklbw xmm5, xmm4 + + psubw xmm1, xmm3 + psubw xmm2, xmm5 + paddw xmm6, xmm1 + paddw xmm6, xmm2 + pmaddwd xmm1, xmm1 + pmaddwd xmm2, xmm2 + paddd xmm7, xmm1 + paddd xmm7, xmm2 + + movdqa xmm1, xmm0 + lea rsi, [rsi + rax] ;ref_pixels_per_line + lea rdi, [rdi + rbx] ;src_pixels_per_line + + sub rcx, 1 + jnz filter_block2d_bil_sp_only_loop + + jmp filter_block2d_bil_variance + +filter_block2d_bil_var_ssse3_full_pixel: + mov rsi, arg(0) ;ref_ptr + mov rdi, arg(2) ;src_ptr + movsxd rcx, dword ptr arg(4) ;Height + movsxd rax, dword ptr arg(1) ;ref_pixels_per_line + movsxd rbx, dword ptr arg(3) ;src_pixels_per_line + pxor xmm0, xmm0 + +filter_block2d_bil_full_pixel_loop: + movq xmm1, QWORD PTR [rsi] + punpcklbw xmm1, xmm0 + movq xmm2, QWORD PTR [rsi+8] + punpcklbw xmm2, xmm0 + + movq xmm3, QWORD PTR [rdi] + punpcklbw xmm3, xmm0 + movq xmm4, QWORD PTR [rdi+8] + punpcklbw xmm4, xmm0 + + psubw xmm1, xmm3 + psubw xmm2, xmm4 + paddw xmm6, xmm1 + paddw xmm6, xmm2 + pmaddwd xmm1, xmm1 + pmaddwd xmm2, xmm2 + paddd xmm7, xmm1 + paddd xmm7, xmm2 + + lea rsi, [rsi + rax] ;ref_pixels_per_line + lea rdi, [rdi + rbx] ;src_pixels_per_line + sub rcx, 1 + jnz filter_block2d_bil_full_pixel_loop + + jmp filter_block2d_bil_variance + +filter_block2d_bil_var_ssse3_fp_only: + mov rsi, arg(0) ;ref_ptr + mov rdi, arg(2) ;src_ptr + movsxd rcx, dword ptr arg(4) ;Height + movsxd rdx, dword ptr arg(1) ;ref_pixels_per_line + + pxor xmm0, xmm0 + movsxd rbx, dword ptr arg(3) ;src_pixels_per_line + +filter_block2d_bil_fp_only_loop: + movdqu xmm1, XMMWORD PTR [rsi] + movdqu xmm2, XMMWORD PTR [rsi+1] + movdqa xmm3, xmm1 + + punpcklbw xmm1, xmm2 + punpckhbw xmm3, xmm2 + pmaddubsw xmm1, [rax] + pmaddubsw xmm3, [rax] + + paddw xmm1, [GLOBAL(xmm_bi_rd)] + paddw xmm3, [GLOBAL(xmm_bi_rd)] + psraw xmm1, xmm_filter_shift + psraw xmm3, xmm_filter_shift + + movq xmm2, XMMWORD PTR [rdi] + pxor xmm4, xmm4 + punpcklbw xmm2, xmm4 + movq xmm5, QWORD PTR [rdi+8] + punpcklbw xmm5, xmm4 + + psubw xmm1, xmm2 + psubw xmm3, xmm5 + paddw xmm6, xmm1 + paddw xmm6, xmm3 + pmaddwd xmm1, xmm1 + pmaddwd xmm3, xmm3 + paddd xmm7, xmm1 + paddd xmm7, xmm3 + + lea rsi, [rsi + rdx] + lea rdi, [rdi + rbx] ;src_pixels_per_line + + sub rcx, 1 + jnz filter_block2d_bil_fp_only_loop + + jmp filter_block2d_bil_variance + +filter_block2d_bil_variance: + pxor xmm0, xmm0 + pxor xmm1, xmm1 + pxor xmm5, xmm5 + + punpcklwd xmm0, xmm6 + punpckhwd xmm1, xmm6 + psrad xmm0, 16 + psrad xmm1, 16 + paddd xmm0, xmm1 + movdqa xmm1, xmm0 + + movdqa xmm6, xmm7 + punpckldq xmm6, xmm5 + punpckhdq xmm7, xmm5 + paddd xmm6, xmm7 + + punpckldq xmm0, xmm5 + punpckhdq xmm1, xmm5 + paddd xmm0, xmm1 + + movdqa xmm7, xmm6 + movdqa xmm1, xmm0 + + psrldq xmm7, 8 + psrldq xmm1, 8 + + paddd xmm6, xmm7 + paddd xmm0, xmm1 + + mov rsi, arg(7) ;[Sum] + mov rdi, arg(8) ;[SSE] + + movd [rsi], xmm0 + movd [rdi], xmm6 + + ; begin epilog + pop rbx + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + + +SECTION_RODATA +align 16 +xmm_bi_rd: + times 8 dw 64 +align 16 +vp8_bilinear_filters_ssse3: + times 8 db 128, 0 + times 8 db 112, 16 + times 8 db 96, 32 + times 8 db 80, 48 + times 8 db 64, 64 + times 8 db 48, 80 + times 8 db 32, 96 + times 8 db 16, 112 diff --git a/vp8/encoder/x86/variance_mmx.c b/vp8/encoder/x86/variance_mmx.c index 6eed98e07..07358c0c7 100644 --- a/vp8/encoder/x86/variance_mmx.c +++ b/vp8/encoder/x86/variance_mmx.c @@ -456,146 +456,6 @@ unsigned int vp8_sub_pixel_variance8x16_mmx return (xxsum - ((xsum * xsum) >> 7)); } -unsigned int vp8_i_variance16x16_mmx( - const unsigned char *src_ptr, - int source_stride, - const unsigned char *ref_ptr, - int recon_stride, - unsigned int *sse) -{ - unsigned int sse0, sse1, sse2, sse3, var; - int sum0, sum1, sum2, sum3, avg; - - - vp8_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ; - vp8_get8x8var_mmx(src_ptr + 8, source_stride, ref_ptr + 8, recon_stride, &sse1, &sum1); - vp8_get8x8var_mmx(src_ptr + (source_stride >> 1), source_stride, ref_ptr + (recon_stride >> 1), recon_stride, &sse2, &sum2) ; - vp8_get8x8var_mmx(src_ptr + (source_stride >> 1) + 8, source_stride, ref_ptr + (recon_stride >> 1) + 8, recon_stride, &sse3, &sum3); - - var = sse0 + sse1 + sse2 + sse3; - avg = sum0 + sum1 + sum2 + sum3; - *sse = var; - return (var - ((avg * avg) >> 8)); - -} - -unsigned int vp8_i_variance8x16_mmx( - const unsigned char *src_ptr, - int source_stride, - const unsigned char *ref_ptr, - int recon_stride, - unsigned int *sse) -{ - unsigned int sse0, sse1, var; - int sum0, sum1, avg; - vp8_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ; - vp8_get8x8var_mmx(src_ptr + (source_stride >> 1), source_stride, ref_ptr + (recon_stride >> 1), recon_stride, &sse1, &sum1) ; - - var = sse0 + sse1; - avg = sum0 + sum1; - - *sse = var; - return (var - ((avg * avg) >> 7)); - -} - -unsigned int vp8_i_sub_pixel_variance16x16_mmx -( - const unsigned char *src_ptr, - int src_pixels_per_line, - int xoffset, - int yoffset, - const unsigned char *dst_ptr, - int dst_pixels_per_line, - unsigned int *sse -) -{ - int xsum0, xsum1; - unsigned int xxsum0, xxsum1; - int f2soffset = (src_pixels_per_line >> 1); - int f2doffset = (dst_pixels_per_line >> 1); - - - vp8_filter_block2d_bil_var_mmx( - src_ptr, src_pixels_per_line, - dst_ptr, dst_pixels_per_line, 8, - vp8_vp7_bilinear_filters_mmx[xoffset], vp8_vp7_bilinear_filters_mmx[yoffset], - &xsum0, &xxsum0 - ); - - - vp8_filter_block2d_bil_var_mmx( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 8, - vp8_vp7_bilinear_filters_mmx[xoffset], vp8_vp7_bilinear_filters_mmx[yoffset], - &xsum1, &xxsum1 - ); - - xsum0 += xsum1; - xxsum0 += xxsum1; - - vp8_filter_block2d_bil_var_mmx( - src_ptr + f2soffset, src_pixels_per_line, - dst_ptr + f2doffset, dst_pixels_per_line, 8, - vp8_vp7_bilinear_filters_mmx[xoffset], vp8_vp7_bilinear_filters_mmx[yoffset], - &xsum1, &xxsum1 - ); - - xsum0 += xsum1; - xxsum0 += xxsum1; - - vp8_filter_block2d_bil_var_mmx( - src_ptr + f2soffset + 8, src_pixels_per_line, - dst_ptr + f2doffset + 8, dst_pixels_per_line, 8, - vp8_vp7_bilinear_filters_mmx[xoffset], vp8_vp7_bilinear_filters_mmx[yoffset], - &xsum1, &xxsum1 - ); - - xsum0 += xsum1; - xxsum0 += xxsum1; - *sse = xxsum0; - return (xxsum0 - ((xsum0 * xsum0) >> 8)); -} - - -unsigned int vp8_i_sub_pixel_variance8x16_mmx -( - const unsigned char *src_ptr, - int src_pixels_per_line, - int xoffset, - int yoffset, - const unsigned char *dst_ptr, - int dst_pixels_per_line, - unsigned int *sse -) -{ - int xsum0, xsum1; - unsigned int xxsum0, xxsum1; - int f2soffset = (src_pixels_per_line >> 1); - int f2doffset = (dst_pixels_per_line >> 1); - - - vp8_filter_block2d_bil_var_mmx( - src_ptr, src_pixels_per_line, - dst_ptr, dst_pixels_per_line, 8, - vp8_vp7_bilinear_filters_mmx[xoffset], vp8_vp7_bilinear_filters_mmx[yoffset], - &xsum0, &xxsum0 - ); - - - vp8_filter_block2d_bil_var_mmx( - src_ptr + f2soffset, src_pixels_per_line, - dst_ptr + f2doffset, dst_pixels_per_line, 8, - vp8_vp7_bilinear_filters_mmx[xoffset], vp8_vp7_bilinear_filters_mmx[yoffset], - &xsum1, &xxsum1 - ); - - xsum0 += xsum1; - xxsum0 += xxsum1; - *sse = xxsum0; - return (xxsum0 - ((xsum0 * xsum0) >> 7)); -} - unsigned int vp8_variance_halfpixvar16x16_h_mmx( const unsigned char *src_ptr, diff --git a/vp8/encoder/x86/variance_sse2.c b/vp8/encoder/x86/variance_sse2.c index 7cf6a6308..0edda3062 100644 --- a/vp8/encoder/x86/variance_sse2.c +++ b/vp8/encoder/x86/variance_sse2.c @@ -81,6 +81,16 @@ void vp8_filter_block2d_bil_var_sse2 int *sum, unsigned int *sumsquared ); +void vp8_half_horiz_vert_variance8x_h_sse2 +( + const unsigned char *ref_ptr, + int ref_pixels_per_line, + const unsigned char *src_ptr, + int src_pixels_per_line, + unsigned int Height, + int *sum, + unsigned int *sumsquared +); void vp8_half_horiz_vert_variance16x_h_sse2 ( const unsigned char *ref_ptr, @@ -91,6 +101,16 @@ void vp8_half_horiz_vert_variance16x_h_sse2 int *sum, unsigned int *sumsquared ); +void vp8_half_horiz_variance8x_h_sse2 +( + const unsigned char *ref_ptr, + int ref_pixels_per_line, + const unsigned char *src_ptr, + int src_pixels_per_line, + unsigned int Height, + int *sum, + unsigned int *sumsquared +); void vp8_half_horiz_variance16x_h_sse2 ( const unsigned char *ref_ptr, @@ -101,6 +121,16 @@ void vp8_half_horiz_variance16x_h_sse2 int *sum, unsigned int *sumsquared ); +void vp8_half_vert_variance8x_h_sse2 +( + const unsigned char *ref_ptr, + int ref_pixels_per_line, + const unsigned char *src_ptr, + int src_pixels_per_line, + unsigned int Height, + int *sum, + unsigned int *sumsquared +); void vp8_half_vert_variance16x_h_sse2 ( const unsigned char *ref_ptr, @@ -262,21 +292,21 @@ unsigned int vp8_sub_pixel_variance8x8_wmt if (xoffset == 4 && yoffset == 0) { - vp8_half_horiz_variance16x_h_sse2( + vp8_half_horiz_variance8x_h_sse2( src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 8, &xsum, &xxsum); } else if (xoffset == 0 && yoffset == 4) { - vp8_half_vert_variance16x_h_sse2( + vp8_half_vert_variance8x_h_sse2( src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 8, &xsum, &xxsum); } else if (xoffset == 4 && yoffset == 4) { - vp8_half_horiz_vert_variance16x_h_sse2( + vp8_half_horiz_vert_variance8x_h_sse2( src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 8, &xsum, &xxsum); @@ -317,11 +347,6 @@ unsigned int vp8_sub_pixel_variance16x16_wmt src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 16, &xsum0, &xxsum0); - - vp8_half_horiz_variance16x_h_sse2( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 16, - &xsum1, &xxsum1); } else if (xoffset == 0 && yoffset == 4) { @@ -329,11 +354,6 @@ unsigned int vp8_sub_pixel_variance16x16_wmt src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 16, &xsum0, &xxsum0); - - vp8_half_vert_variance16x_h_sse2( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 16, - &xsum1, &xxsum1); } else if (xoffset == 4 && yoffset == 4) { @@ -341,11 +361,6 @@ unsigned int vp8_sub_pixel_variance16x16_wmt src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 16, &xsum0, &xxsum0); - - vp8_half_horiz_vert_variance16x_h_sse2( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 16, - &xsum1, &xxsum1); } else { @@ -356,17 +371,16 @@ unsigned int vp8_sub_pixel_variance16x16_wmt &xsum0, &xxsum0 ); - vp8_filter_block2d_bil_var_sse2( src_ptr + 8, src_pixels_per_line, dst_ptr + 8, dst_pixels_per_line, 16, xoffset, yoffset, &xsum1, &xxsum1 ); + xsum0 += xsum1; + xxsum0 += xxsum1; } - xsum0 += xsum1; - xxsum0 += xxsum1; *sse = xxsum0; return (xxsum0 - ((xsum0 * xsum0) >> 8)); } @@ -406,11 +420,6 @@ unsigned int vp8_sub_pixel_variance16x8_wmt src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 8, &xsum0, &xxsum0); - - vp8_half_horiz_variance16x_h_sse2( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 8, - &xsum1, &xxsum1); } else if (xoffset == 0 && yoffset == 4) { @@ -418,11 +427,6 @@ unsigned int vp8_sub_pixel_variance16x8_wmt src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 8, &xsum0, &xxsum0); - - vp8_half_vert_variance16x_h_sse2( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 8, - &xsum1, &xxsum1); } else if (xoffset == 4 && yoffset == 4) { @@ -430,11 +434,6 @@ unsigned int vp8_sub_pixel_variance16x8_wmt src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 8, &xsum0, &xxsum0); - - vp8_half_horiz_vert_variance16x_h_sse2( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 8, - &xsum1, &xxsum1); } else { @@ -449,11 +448,10 @@ unsigned int vp8_sub_pixel_variance16x8_wmt dst_ptr + 8, dst_pixels_per_line, 8, xoffset, yoffset, &xsum1, &xxsum1); + xsum0 += xsum1; + xxsum0 += xxsum1; } - xsum0 += xsum1; - xxsum0 += xxsum1; - *sse = xxsum0; return (xxsum0 - ((xsum0 * xsum0) >> 7)); } @@ -474,21 +472,21 @@ unsigned int vp8_sub_pixel_variance8x16_wmt if (xoffset == 4 && yoffset == 0) { - vp8_half_horiz_variance16x_h_sse2( + vp8_half_horiz_variance8x_h_sse2( src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 16, &xsum, &xxsum); } else if (xoffset == 0 && yoffset == 4) { - vp8_half_vert_variance16x_h_sse2( + vp8_half_vert_variance8x_h_sse2( src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 16, &xsum, &xxsum); } else if (xoffset == 4 && yoffset == 4) { - vp8_half_horiz_vert_variance16x_h_sse2( + vp8_half_horiz_vert_variance8x_h_sse2( src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 16, &xsum, &xxsum); @@ -506,81 +504,6 @@ unsigned int vp8_sub_pixel_variance8x16_wmt return (xxsum - ((xsum * xsum) >> 7)); } -unsigned int vp8_i_variance16x16_wmt( - const unsigned char *src_ptr, - int source_stride, - const unsigned char *ref_ptr, - int recon_stride, - unsigned int *sse) -{ - unsigned int sse0, sse1, sse2, sse3, var; - int sum0, sum1, sum2, sum3, avg; - - - vp8_get8x8var_sse2(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ; - vp8_get8x8var_sse2(src_ptr + 8, source_stride, ref_ptr + 8, recon_stride, &sse1, &sum1); - vp8_get8x8var_sse2(src_ptr + (source_stride >> 1), source_stride, ref_ptr + (recon_stride >> 1), recon_stride, &sse2, &sum2) ; - vp8_get8x8var_sse2(src_ptr + (source_stride >> 1) + 8, source_stride, ref_ptr + (recon_stride >> 1) + 8, recon_stride, &sse3, &sum3); - - var = sse0 + sse1 + sse2 + sse3; - avg = sum0 + sum1 + sum2 + sum3; - - *sse = var; - return (var - ((avg * avg) >> 8)); - -} - -unsigned int vp8_i_variance8x16_wmt( - const unsigned char *src_ptr, - int source_stride, - const unsigned char *ref_ptr, - int recon_stride, - unsigned int *sse) -{ - unsigned int sse0, sse1, var; - int sum0, sum1, avg; - vp8_get8x8var_sse2(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ; - vp8_get8x8var_sse2(src_ptr + (source_stride >> 1), source_stride, ref_ptr + (recon_stride >> 1), recon_stride, &sse1, &sum1) ; - - var = sse0 + sse1; - avg = sum0 + sum1; - - *sse = var; - return (var - ((avg * avg) >> 7)); - -} - - -unsigned int vp8_i_sub_pixel_variance16x16_wmt -( - const unsigned char *src_ptr, - int src_pixels_per_line, - int xoffset, - int yoffset, - const unsigned char *dst_ptr, - int dst_pixels_per_line, - unsigned int *sse -) -{ - return vp8_sub_pixel_variance16x16_wmt(src_ptr, (src_pixels_per_line >> 1), xoffset, yoffset, dst_ptr, (dst_pixels_per_line >> 1), sse); -} - - -unsigned int vp8_i_sub_pixel_variance8x16_wmt -( - const unsigned char *src_ptr, - int src_pixels_per_line, - int xoffset, - int yoffset, - const unsigned char *dst_ptr, - int dst_pixels_per_line, - unsigned int *sse -) -{ - - return vp8_sub_pixel_variance8x16_wmt(src_ptr, (src_pixels_per_line >> 1), xoffset, yoffset, dst_ptr, (dst_pixels_per_line >> 1), sse); -} - unsigned int vp8_variance_halfpixvar16x16_h_wmt( const unsigned char *src_ptr, @@ -589,21 +512,14 @@ unsigned int vp8_variance_halfpixvar16x16_h_wmt( int dst_pixels_per_line, unsigned int *sse) { - int xsum0, xsum1; - unsigned int xxsum0, xxsum1; + int xsum0; + unsigned int xxsum0; vp8_half_horiz_variance16x_h_sse2( src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 16, &xsum0, &xxsum0); - vp8_half_horiz_variance16x_h_sse2( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 16, - &xsum1, &xxsum1); - - xsum0 += xsum1; - xxsum0 += xxsum1; *sse = xxsum0; return (xxsum0 - ((xsum0 * xsum0) >> 8)); } @@ -616,21 +532,13 @@ unsigned int vp8_variance_halfpixvar16x16_v_wmt( int dst_pixels_per_line, unsigned int *sse) { - int xsum0, xsum1; - unsigned int xxsum0, xxsum1; - + int xsum0; + unsigned int xxsum0; vp8_half_vert_variance16x_h_sse2( src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 16, &xsum0, &xxsum0); - vp8_half_vert_variance16x_h_sse2( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 16, - &xsum1, &xxsum1); - - xsum0 += xsum1; - xxsum0 += xxsum1; *sse = xxsum0; return (xxsum0 - ((xsum0 * xsum0) >> 8)); } @@ -643,21 +551,14 @@ unsigned int vp8_variance_halfpixvar16x16_hv_wmt( int dst_pixels_per_line, unsigned int *sse) { - int xsum0, xsum1; - unsigned int xxsum0, xxsum1; + int xsum0; + unsigned int xxsum0; vp8_half_horiz_vert_variance16x_h_sse2( src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 16, &xsum0, &xxsum0); - vp8_half_horiz_vert_variance16x_h_sse2( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 16, - &xsum1, &xxsum1); - - xsum0 += xsum1; - xxsum0 += xxsum1; *sse = xxsum0; return (xxsum0 - ((xsum0 * xsum0) >> 8)); } diff --git a/vp8/encoder/x86/variance_ssse3.c b/vp8/encoder/x86/variance_ssse3.c new file mode 100644 index 000000000..d50ae3ade --- /dev/null +++ b/vp8/encoder/x86/variance_ssse3.c @@ -0,0 +1,116 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "vp8/encoder/variance.h" +#include "vp8/common/pragmas.h" +#include "vpx_ports/mem.h" + +extern unsigned int vp8_get16x16var_sse2 +( + const unsigned char *src_ptr, + int source_stride, + const unsigned char *ref_ptr, + int recon_stride, + unsigned int *SSE, + int *Sum +); +extern void vp8_half_horiz_vert_variance16x_h_sse2 +( + const unsigned char *ref_ptr, + int ref_pixels_per_line, + const unsigned char *src_ptr, + int src_pixels_per_line, + unsigned int Height, + int *sum, + unsigned int *sumsquared +); +extern void vp8_half_horiz_variance16x_h_sse2 +( + const unsigned char *ref_ptr, + int ref_pixels_per_line, + const unsigned char *src_ptr, + int src_pixels_per_line, + unsigned int Height, + int *sum, + unsigned int *sumsquared +); +extern void vp8_half_vert_variance16x_h_sse2 +( + const unsigned char *ref_ptr, + int ref_pixels_per_line, + const unsigned char *src_ptr, + int src_pixels_per_line, + unsigned int Height, + int *sum, + unsigned int *sumsquared +); +extern void vp8_filter_block2d_bil_var_ssse3 +( + const unsigned char *ref_ptr, + int ref_pixels_per_line, + const unsigned char *src_ptr, + int src_pixels_per_line, + unsigned int Height, + int xoffset, + int yoffset, + int *sum, + unsigned int *sumsquared +); + +unsigned int vp8_sub_pixel_variance16x16_ssse3 +( + const unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + const unsigned char *dst_ptr, + int dst_pixels_per_line, + unsigned int *sse +) +{ + int xsum0, xsum1; + unsigned int xxsum0, xxsum1; + + // note we could avoid these if statements if the calling function + // just called the appropriate functions inside. + if (xoffset == 4 && yoffset == 0) + { + vp8_half_horiz_variance16x_h_sse2( + src_ptr, src_pixels_per_line, + dst_ptr, dst_pixels_per_line, 16, + &xsum0, &xxsum0); + } + else if (xoffset == 0 && yoffset == 4) + { + vp8_half_vert_variance16x_h_sse2( + src_ptr, src_pixels_per_line, + dst_ptr, dst_pixels_per_line, 16, + &xsum0, &xxsum0); + } + else if (xoffset == 4 && yoffset == 4) + { + vp8_half_horiz_vert_variance16x_h_sse2( + src_ptr, src_pixels_per_line, + dst_ptr, dst_pixels_per_line, 16, + &xsum0, &xxsum0); + } + else + { + vp8_filter_block2d_bil_var_ssse3( + src_ptr, src_pixels_per_line, + dst_ptr, dst_pixels_per_line, 16, + xoffset, yoffset, + &xsum0, &xxsum0); + } + + *sse = xxsum0; + return (xxsum0 - ((xsum0 * xsum0) >> 8)); +} diff --git a/vp8/encoder/x86/variance_x86.h b/vp8/encoder/x86/variance_x86.h index 6bea15ebc..1e2fb3490 100644 --- a/vp8/encoder/x86/variance_x86.h +++ b/vp8/encoder/x86/variance_x86.h @@ -286,6 +286,7 @@ extern prototype_sad_multi_dif_address(vp8_sad4x4x4d_sse3); #if HAVE_SSSE3 extern prototype_sad_multi_same_address(vp8_sad16x16x3_ssse3); extern prototype_sad_multi_same_address(vp8_sad16x8x3_ssse3); +extern prototype_subpixvariance(vp8_sub_pixel_variance16x16_ssse3); #if !CONFIG_RUNTIME_CPU_DETECT #undef vp8_variance_sad16x16x3 @@ -294,6 +295,9 @@ extern prototype_sad_multi_same_address(vp8_sad16x8x3_ssse3); #undef vp8_variance_sad16x8x3 #define vp8_variance_sad16x8x3 vp8_sad16x8x3_ssse3 +#undef vp8_variance_subpixvar16x16 +#define vp8_variance_subpixvar16x16 vp8_sub_pixel_variance16x16_ssse3 + #endif #endif diff --git a/vp8/encoder/x86/x86_csystemdependent.c b/vp8/encoder/x86/x86_csystemdependent.c index 61c603229..c7639a7e4 100644 --- a/vp8/encoder/x86/x86_csystemdependent.c +++ b/vp8/encoder/x86/x86_csystemdependent.c @@ -334,6 +334,8 @@ void vp8_arch_x86_encoder_init(VP8_COMP *cpi) cpi->rtcd.variance.sad16x16x3 = vp8_sad16x16x3_ssse3; cpi->rtcd.variance.sad16x8x3 = vp8_sad16x8x3_ssse3; + cpi->rtcd.variance.subpixvar16x16 = vp8_sub_pixel_variance16x16_ssse3; + cpi->rtcd.quantize.fastquantb = vp8_fast_quantize_b_ssse3; } diff --git a/vp8/vp8cx.mk b/vp8/vp8cx.mk index ed2feddae..c0ae250f5 100644 --- a/vp8/vp8cx.mk +++ b/vp8/vp8cx.mk @@ -110,6 +110,8 @@ VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/subtract_sse2.asm VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/temporal_filter_apply_sse2.asm VP8_CX_SRCS-$(HAVE_SSE3) += encoder/x86/sad_sse3.asm VP8_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/sad_ssse3.asm +VP8_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/variance_ssse3.c +VP8_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/variance_impl_ssse3.asm VP8_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/quantize_ssse3.asm VP8_CX_SRCS-$(HAVE_SSE4_1) += encoder/x86/sad_sse4.asm VP8_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/quantize_mmx.asm @@ -163,8 +163,8 @@ int stats_open_file(stats_io_t *stats, const char *fpf, int pass) if (!stats->buf.buf) { - fprintf(stderr, "Failed to allocate first-pass stats buffer (%d bytes)\n", - stats->buf_alloc_sz); + fprintf(stderr, "Failed to allocate first-pass stats buffer (%lu bytes)\n", + (unsigned long)stats->buf_alloc_sz); exit(EXIT_FAILURE); } |