diff --git a/.github/workflows/tests.yml b/.github/workflows/tests.yml
index 986c41e0..8eeb069a 100644
--- a/.github/workflows/tests.yml
+++ b/.github/workflows/tests.yml
@@ -45,6 +45,7 @@ jobs:
     runs-on: ubuntu-latest
     needs: [code_style, rule_linter]
     strategy:
+      fail-fast: false
       matrix:
         include:
           - python: 2.7
diff --git a/capa/features/extractors/smda/__init__.py b/capa/features/extractors/smda/__init__.py
new file mode 100644
index 00000000..b7ea3ec1
--- /dev/null
+++ b/capa/features/extractors/smda/__init__.py
@@ -0,0 +1,52 @@
+import sys
+import types
+
+from smda.common.SmdaReport import SmdaReport
+from smda.common.SmdaInstruction import SmdaInstruction
+
+import capa.features.extractors.smda.file
+import capa.features.extractors.smda.insn
+import capa.features.extractors.smda.function
+import capa.features.extractors.smda.basicblock
+from capa.main import UnsupportedRuntimeError
+from capa.features.extractors import FeatureExtractor
+
+
+class SmdaFeatureExtractor(FeatureExtractor):
+    def __init__(self, smda_report: SmdaReport, path):
+        super(SmdaFeatureExtractor, self).__init__()
+        if sys.version_info < (3, 0):
+            raise UnsupportedRuntimeError("SMDA should only be used with Python 3.")
+        self.smda_report = smda_report
+        self.path = path
+
+    def get_base_address(self):
+        return self.smda_report.base_addr
+
+    def extract_file_features(self):
+        for feature, va in capa.features.extractors.smda.file.extract_features(self.smda_report, self.path):
+            yield feature, va
+
+    def get_functions(self):
+        for function in self.smda_report.getFunctions():
+            yield function
+
+    def extract_function_features(self, f):
+        for feature, va in capa.features.extractors.smda.function.extract_features(f):
+            yield feature, va
+
+    def get_basic_blocks(self, f):
+        for bb in f.getBlocks():
+            yield bb
+
+    def extract_basic_block_features(self, f, bb):
+        for feature, va in capa.features.extractors.smda.basicblock.extract_features(f, bb):
+            yield feature, va
+
+    def get_instructions(self, f, bb):
+        for smda_ins in bb.getInstructions():
+            yield smda_ins
+
+    def extract_insn_features(self, f, bb, insn):
+        for feature, va in capa.features.extractors.smda.insn.extract_features(f, bb, insn):
+            yield feature, va
diff --git a/capa/features/extractors/smda/basicblock.py b/capa/features/extractors/smda/basicblock.py
new file mode 100644
index 00000000..07477139
--- /dev/null
+++ b/capa/features/extractors/smda/basicblock.py
@@ -0,0 +1,131 @@
+import sys
+import string
+import struct
+
+from capa.features import Characteristic
+from capa.features.basicblock import BasicBlock
+from capa.features.extractors.helpers import MIN_STACKSTRING_LEN
+
+
+def _bb_has_tight_loop(f, bb):
+    """
+    parse tight loops, true if last instruction in basic block branches to bb start
+    """
+    return bb.offset in f.blockrefs[bb.offset] if bb.offset in f.blockrefs else False
+
+
+def extract_bb_tight_loop(f, bb):
+    """ check basic block for tight loop indicators """
+    if _bb_has_tight_loop(f, bb):
+        yield Characteristic("tight loop"), bb.offset
+
+
+def _bb_has_stackstring(f, bb):
+    """
+    extract potential stackstring creation, using the following heuristics:
+      - basic block contains enough moves of constant bytes to the stack
+    """
+    count = 0
+    for instr in bb.getInstructions():
+        if is_mov_imm_to_stack(instr):
+            count += get_printable_len(instr.getDetailed())
+        if count > MIN_STACKSTRING_LEN:
+            return True
+    return False
+
+
+def get_operands(smda_ins):
+    return [o.strip() for o in smda_ins.operands.split(",")]
+
+
+def extract_stackstring(f, bb):
+    """ check basic block for stackstring indicators """
+    if _bb_has_stackstring(f, bb):
+        yield Characteristic("stack string"), bb.offset
+
+
+def is_mov_imm_to_stack(smda_ins):
+    """
+    Return if instruction moves immediate onto stack
+    """
+    if not smda_ins.mnemonic.startswith("mov"):
+        return False
+
+    try:
+        dst, src = get_operands(smda_ins)
+    except ValueError:
+        # not two operands
+        return False
+
+    try:
+        int(src, 16)
+    except ValueError:
+        return False
+
+    if not any(regname in dst for regname in ["ebp", "rbp", "esp", "rsp"]):
+        return False
+
+    return True
+
+
+def is_printable_ascii(chars):
+    return all(c < 127 and chr(c) in string.printable for c in chars)
+
+
+def is_printable_utf16le(chars):
+    if all(c == 0x00 for c in chars[1::2]):
+        return is_printable_ascii(chars[::2])
+
+
+def get_printable_len(instr):
+    """
+    Return string length if all operand bytes are ascii or utf16-le printable
+
+    Works on a capstone instruction
+    """
+    # should have exactly two operands for mov immediate
+    if len(instr.operands) != 2:
+        return 0
+
+    op_value = instr.operands[1].value.imm
+
+    if instr.imm_size == 1:
+        chars = struct.pack("<B", op_value & 0xFF)
+    elif instr.imm_size == 2:
+        chars = struct.pack("<H", op_value & 0xFFFF)
+    elif instr.imm_size == 4:
+        chars = struct.pack("<I", op_value & 0xFFFFFFFF)
+    elif instr.imm_size == 8:
+        chars = struct.pack("<Q", op_value & 0xFFFFFFFFFFFFFFFF)
+    else:
+        raise ValueError("Unhandled operand data type 0x%x." % instr.imm_size)
+
+    if is_printable_ascii(chars):
+        return instr.imm_size
+    if is_printable_utf16le(chars):
+        return instr.imm_size // 2
+
+    return 0
+
+
+def extract_features(f, bb):
+    """
+    extract features from the given basic block.
+
+    args:
+      f (smda.common.SmdaFunction): the function from which to extract features
+      bb (smda.common.SmdaBasicBlock): the basic block to process.
+
+    yields:
+      Feature, set[VA]: the features and their location found in this basic block.
+    """
+    yield BasicBlock(), bb.offset
+    for bb_handler in BASIC_BLOCK_HANDLERS:
+        for feature, va in bb_handler(f, bb):
+            yield feature, va
+
+
+BASIC_BLOCK_HANDLERS = (
+    extract_bb_tight_loop,
+    extract_stackstring,
+)
diff --git a/capa/features/extractors/smda/file.py b/capa/features/extractors/smda/file.py
new file mode 100644
index 00000000..6f690dbe
--- /dev/null
+++ b/capa/features/extractors/smda/file.py
@@ -0,0 +1,139 @@
+import struct
+
+# if we have SMDA we definitely have lief
+import lief
+
+import capa.features.extractors.helpers
+import capa.features.extractors.strings
+from capa.features import String, Characteristic
+from capa.features.file import Export, Import, Section
+
+
+def carve(pbytes, offset=0):
+    """
+    Return a list of (offset, size, xor) tuples of embedded PEs
+
+    Based on the version from vivisect:
+    https://github.com/vivisect/vivisect/blob/7be4037b1cecc4551b397f840405a1fc606f9b53/PE/carve.py#L19
+    And its IDA adaptation:
+    capa/features/extractors/ida/file.py
+    """
+    mz_xor = [
+        (
+            capa.features.extractors.helpers.xor_static(b"MZ", i),
+            capa.features.extractors.helpers.xor_static(b"PE", i),
+            i,
+        )
+        for i in range(256)
+    ]
+
+    pblen = len(pbytes)
+    todo = [(pbytes.find(mzx, offset), mzx, pex, i) for mzx, pex, i in mz_xor]
+    todo = [(off, mzx, pex, i) for (off, mzx, pex, i) in todo if off != -1]
+
+    while len(todo):
+
+        off, mzx, pex, i = todo.pop()
+
+        # The MZ header has one field we will check
+        # e_lfanew is at 0x3c
+        e_lfanew = off + 0x3C
+        if pblen < (e_lfanew + 4):
+            continue
+
+        newoff = struct.unpack("<I", capa.features.extractors.helpers.xor_static(pbytes[e_lfanew : e_lfanew + 4], i))[0]
+
+        nextres = pbytes.find(mzx, off + 1)
+        if nextres != -1:
+            todo.append((nextres, mzx, pex, i))
+
+        peoff = off + newoff
+        if pblen < (peoff + 2):
+            continue
+
+        if pbytes[peoff : peoff + 2] == pex:
+            yield (off, i)
+
+
+def extract_file_embedded_pe(smda_report, file_path):
+    with open(file_path, "rb") as f:
+        fbytes = f.read()
+
+    for offset, i in carve(fbytes, 1):
+        yield Characteristic("embedded pe"), offset
+
+
+def extract_file_export_names(smda_report, file_path):
+    lief_binary = lief.parse(file_path)
+    if lief_binary is not None:
+        for function in lief_binary.exported_functions:
+            yield Export(function.name), function.address
+
+
+def extract_file_import_names(smda_report, file_path):
+    # extract import table info via LIEF
+    lief_binary = lief.parse(file_path)
+    if not isinstance(lief_binary, lief.PE.Binary):
+        return
+    for imported_library in lief_binary.imports:
+        library_name = imported_library.name.lower()
+        library_name = library_name[:-4] if library_name.endswith(".dll") else library_name
+        for func in imported_library.entries:
+            if func.name:
+                va = func.iat_address + smda_report.base_addr
+                for name in capa.features.extractors.helpers.generate_symbols(library_name, func.name):
+                    yield Import(name), va
+            elif func.is_ordinal:
+                for name in capa.features.extractors.helpers.generate_symbols(library_name, "#%s" % func.ordinal):
+                    yield Import(name), va
+
+
+def extract_file_section_names(smda_report, file_path):
+    lief_binary = lief.parse(file_path)
+    if not isinstance(lief_binary, lief.PE.Binary):
+        return
+    if lief_binary and lief_binary.sections:
+        base_address = lief_binary.optional_header.imagebase
+        for section in lief_binary.sections:
+            yield Section(section.name), base_address + section.virtual_address
+
+
+def extract_file_strings(smda_report, file_path):
+    """
+    extract ASCII and UTF-16 LE strings from file
+    """
+    with open(file_path, "rb") as f:
+        b = f.read()
+
+    for s in capa.features.extractors.strings.extract_ascii_strings(b):
+        yield String(s.s), s.offset
+
+    for s in capa.features.extractors.strings.extract_unicode_strings(b):
+        yield String(s.s), s.offset
+
+
+def extract_features(smda_report, file_path):
+    """
+    extract file features from given workspace
+
+    args:
+      smda_report (smda.common.SmdaReport): a SmdaReport
+      file_path: path to the input file
+
+    yields:
+      Tuple[Feature, VA]: a feature and its location.
+    """
+
+    for file_handler in FILE_HANDLERS:
+        result = file_handler(smda_report, file_path)
+        for feature, va in file_handler(smda_report, file_path):
+            yield feature, va
+
+
+FILE_HANDLERS = (
+    extract_file_embedded_pe,
+    extract_file_export_names,
+    extract_file_import_names,
+    extract_file_section_names,
+    extract_file_strings,
+)
diff --git a/capa/features/extractors/smda/function.py b/capa/features/extractors/smda/function.py
new file mode 100644
index 00000000..e1765e67
--- /dev/null
+++ b/capa/features/extractors/smda/function.py
@@ -0,0 +1,38 @@
+from capa.features import Characteristic
+from capa.features.extractors import loops
+
+
+def extract_function_calls_to(f):
+    for inref in f.inrefs:
+        yield Characteristic("calls to"), inref
+
+
+def extract_function_loop(f):
+    """
+    parse if a function has a loop
+    """
+    edges = []
+    for bb_from, bb_tos in f.blockrefs.items():
+        for bb_to in bb_tos:
+            edges.append((bb_from, bb_to))
+
+    if edges and loops.has_loop(edges):
+        yield Characteristic("loop"), f.offset
+
+
+def extract_features(f):
+    """
+    extract features from the given function.
+
+    args:
+      f (smda.common.SmdaFunction): the function from which to extract features
+
+    yields:
+      Feature, set[VA]: the features and their location found in this function.
+    """
+    for func_handler in FUNCTION_HANDLERS:
+        for feature, va in func_handler(f):
+            yield feature, va
+
+
+FUNCTION_HANDLERS = (extract_function_calls_to, extract_function_loop)
diff --git a/capa/features/extractors/smda/insn.py b/capa/features/extractors/smda/insn.py
new file mode 100644
index 00000000..8fa4e221
--- /dev/null
+++ b/capa/features/extractors/smda/insn.py
@@ -0,0 +1,393 @@
+import re
+import string
+import struct
+
+from smda.common.SmdaReport import SmdaReport
+
+import capa.features.extractors.helpers
+from capa.features import (
+    ARCH_X32,
+    ARCH_X64,
+    MAX_BYTES_FEATURE_SIZE,
+    THUNK_CHAIN_DEPTH_DELTA,
+    Bytes,
+    String,
+    Characteristic,
+)
+from capa.features.insn import API, Number, Offset, Mnemonic
+
+# security cookie checks may perform non-zeroing XORs, these are expected within a certain
+# byte range within the first and returning basic blocks, this helps to reduce FP features
+SECURITY_COOKIE_BYTES_DELTA = 0x40
+PATTERN_HEXNUM = re.compile(r"[+\-] (?P<num>0x[a-fA-F0-9]+)")
+PATTERN_SINGLENUM = re.compile(r"[+\-] (?P<num>[0-9])")
+
+
+def get_arch(smda_report):
+    if smda_report.architecture == "intel":
+        if smda_report.bitness == 32:
+            return ARCH_X32
+        elif smda_report.bitness == 64:
+            return ARCH_X64
+    else:
+        raise NotImplementedError
+
+
+def extract_insn_api_features(f, bb, insn):
+    """parse API features from the given instruction."""
+    if insn.offset in f.apirefs:
+        api_entry = f.apirefs[insn.offset]
+        # reformat
+        dll_name, api_name = api_entry.split("!")
+        dll_name = dll_name.split(".")[0]
+        dll_name = dll_name.lower()
+        for name in capa.features.extractors.helpers.generate_symbols(dll_name, api_name):
+            yield API(name), insn.offset
+    elif insn.offset in f.outrefs:
+        current_function = f
+        current_instruction = insn
+        for index in range(THUNK_CHAIN_DEPTH_DELTA):
+            if current_function and len(current_function.outrefs[current_instruction.offset]) == 1:
+                target = current_function.outrefs[current_instruction.offset][0]
+                referenced_function = current_function.smda_report.getFunction(target)
+                if referenced_function:
+                    # TODO SMDA: implement this function for both jmp and call, checking if function has 1 instruction which refs an API
+                    if referenced_function.isApiThunk():
+                        api_entry = (
+                            referenced_function.apirefs[target] if target in referenced_function.apirefs else None
+                        )
+                        if api_entry:
+                            # reformat
+                            dll_name, api_name = api_entry.split("!")
+                            dll_name = dll_name.split(".")[0]
+                            dll_name = dll_name.lower()
+                            for name in capa.features.extractors.helpers.generate_symbols(dll_name, api_name):
+                                yield API(name), insn.offset
+                    elif referenced_function.num_instructions == 1 and referenced_function.num_outrefs == 1:
+                        current_function = referenced_function
+                        current_instruction = [i for i in referenced_function.getInstructions()][0]
+                else:
+                    return
+
+
+def extract_insn_number_features(f, bb, insn):
+    """parse number features from the given instruction."""
+    # example:
+    #
+    #     push    3136B0h         ; dwControlCode
+    operands = [o.strip() for o in insn.operands.split(",")]
+    if insn.mnemonic == "add" and operands[0] in ["esp", "rsp"]:
+        # skip things like:
+        #
+        #    .text:00401140                 call    sub_407E2B
+        #    .text:00401145                 add     esp, 0Ch
+        return
+    for operand in operands:
+        try:
+            yield Number(int(operand, 16)), insn.offset
+            yield Number(int(operand, 16), arch=get_arch(f.smda_report)), insn.offset
+        except:
+            continue
+
+
+def read_bytes(smda_report, va, num_bytes=None):
+    """
+    read up to MAX_BYTES_FEATURE_SIZE from the given address.
+    """
+
+    rva = va - smda_report.base_addr
+    if smda_report.buffer is None:
+        return
+    buffer_end = len(smda_report.buffer)
+    max_bytes = num_bytes if num_bytes is not None else MAX_BYTES_FEATURE_SIZE
+    if rva + max_bytes > buffer_end:
+        return smda_report.buffer[rva:]
+    else:
+        return smda_report.buffer[rva : rva + max_bytes]
+
+
+def derefs(smda_report, p):
+    """
+    recursively follow the given pointer, yielding the valid memory addresses along the way.
+    useful when you may have a pointer to string, or pointer to pointer to string, etc.
+
+    this is a "do what i mean" type of helper function.
+
+    based on the implementation in viv/insn.py
+    """
+    depth = 0
+    while True:
+        if not smda_report.isAddrWithinMemoryImage(p):
+            return
+        yield p
+
+        bytes_ = read_bytes(smda_report, p, num_bytes=4)
+        val = struct.unpack("I", bytes_)[0]
+
+        # sanity: pointer points to self
+        if val == p:
+            return
+
+        # sanity: avoid chains of pointers that are unreasonably deep
+        depth += 1
+        if depth > 10:
+            return
+
+        p = val
+
+
+def extract_insn_bytes_features(f, bb, insn):
+    """
+    parse byte sequence features from the given instruction.
+    example:
+        #     push    offset iid_004118d4_IShellLinkA ; riid
+    """
+    for data_ref in insn.getDataRefs():
+        for v in derefs(f.smda_report, data_ref):
+            bytes_read = read_bytes(f.smda_report, v)
+            if bytes_read is None:
+                continue
+            if capa.features.extractors.helpers.all_zeros(bytes_read):
+                continue
+
+            yield Bytes(bytes_read), insn.offset
+
+
+def detect_ascii_len(smda_report, offset):
+    if smda_report.buffer is None:
+        return 0
+    ascii_len = 0
+    rva = offset - smda_report.base_addr
+    char = smda_report.buffer[rva]
+    while char < 127 and chr(char) in string.printable:
+        ascii_len += 1
+        rva += 1
+        char = smda_report.buffer[rva]
+    if char == 0:
+        return ascii_len
+    return 0
+
+
+def detect_unicode_len(smda_report, offset):
+    if smda_report.buffer is None:
+        return 0
+    unicode_len = 0
+    rva = offset - smda_report.base_addr
+    char = smda_report.buffer[rva]
+    second_char = smda_report.buffer[rva + 1]
+    while char < 127 and chr(char) in string.printable and second_char == 0:
+        unicode_len += 2
+        rva += 2
+        char = smda_report.buffer[rva]
+        second_char = smda_report.buffer[rva + 1]
+    if char == 0 and second_char == 0:
+        return unicode_len
+    return 0
+
+
+def read_string(smda_report, offset):
+    alen = detect_ascii_len(smda_report, offset)
+    if alen > 1:
+        return read_bytes(smda_report, offset, alen).decode("utf-8")
+    ulen = detect_unicode_len(smda_report, offset)
+    if ulen > 2:
+        return read_bytes(smda_report, offset, ulen).decode("utf-16")
+
+
+def extract_insn_string_features(f, bb, insn):
+    """parse string features from the given instruction."""
+    # example:
+    #
+    #     push    offset aAcr     ; "ACR  > "
+    for data_ref in insn.getDataRefs():
+        for v in derefs(f.smda_report, data_ref):
+            string_read = read_string(f.smda_report, v)
+            if string_read:
+                yield String(string_read.rstrip("\x00")), insn.offset
+
+
+def extract_insn_offset_features(f, bb, insn):
+    """parse structure offset features from the given instruction."""
+    # examples:
+    #
+    #     mov eax, [esi + 4]
+    #     mov eax, [esi + ecx + 16384]
+    operands = [o.strip() for o in insn.operands.split(",")]
+    for operand in operands:
+        if not "ptr" in operand:
+            continue
+        if "esp" in operand or "ebp" in operand or "rbp" in operand:
+            continue
+        number = 0
+        number_hex = re.search(PATTERN_HEXNUM, operand)
+        number_int = re.search(PATTERN_SINGLENUM, operand)
+        if number_hex:
+            number = int(number_hex.group("num"), 16)
+            number = -1 * number if number_hex.group().startswith("-") else number
+        elif number_int:
+            number = int(number_int.group("num"))
+            number = -1 * number if number_int.group().startswith("-") else number
+        yield Offset(number), insn.offset
+        yield Offset(number, arch=get_arch(f.smda_report)), insn.offset
+
+
+def is_security_cookie(f, bb, insn):
+    """
+    check if an instruction is related to security cookie checks
+    """
+    # security cookie check should use SP or BP
+    operands = [o.strip() for o in insn.operands.split(",")]
+    if operands[1] not in ["esp", "ebp", "rsp", "rbp"]:
+        return False
+    for index, block in enumerate(f.getBlocks()):
+        # expect security cookie init in first basic block within first bytes (instructions)
+        block_instructions = [i for i in block.getInstructions()]
+        if index == 0 and insn.offset < (block_instructions[0].offset + SECURITY_COOKIE_BYTES_DELTA):
+            return True
+        # ... or within last bytes (instructions) before a return
+        if block_instructions[-1].mnemonic.startswith("ret") and insn.offset > (
+            block_instructions[-1].offset - SECURITY_COOKIE_BYTES_DELTA
+        ):
+            return True
+    return False
+
+
+def extract_insn_nzxor_characteristic_features(f, bb, insn):
+    """
+    parse non-zeroing XOR instruction from the given instruction.
+    ignore expected non-zeroing XORs, e.g. security cookies.
+    """
+
+    if insn.mnemonic != "xor":
+        return
+
+    operands = [o.strip() for o in insn.operands.split(",")]
+    if operands[0] == operands[1]:
+        return
+
+    if is_security_cookie(f, bb, insn):
+        return
+
+    yield Characteristic("nzxor"), insn.offset
+
+
+def extract_insn_mnemonic_features(f, bb, insn):
+    """parse mnemonic features from the given instruction."""
+    yield Mnemonic(insn.mnemonic), insn.offset
+
+
+def extract_insn_peb_access_characteristic_features(f, bb, insn):
+    """
+    parse peb access from the given function. fs:[0x30] on x86, gs:[0x60] on x64
+    """
+
+    if insn.mnemonic not in ["push", "mov"]:
+        return
+
+    operands = [o.strip() for o in insn.operands.split(",")]
+    for operand in operands:
+        if "fs:" in operand and "0x30" in operand:
+            yield Characteristic("peb access"), insn.offset
+        elif "gs:" in operand and "0x60" in operand:
+            yield Characteristic("peb access"), insn.offset
+
+
+def extract_insn_segment_access_features(f, bb, insn):
+    """ parse the instruction for access to fs or gs """
+    operands = [o.strip() for o in insn.operands.split(",")]
+    for operand in operands:
+        if "fs:" in operand:
+            yield Characteristic("fs access"), insn.offset
+        elif "gs:" in operand:
+            yield Characteristic("gs access"), insn.offset
+
+
+def extract_insn_cross_section_cflow(f, bb, insn):
+    """
+    inspect the instruction for a CALL or JMP that crosses section boundaries.
+    """
+    if insn.mnemonic in ["call", "jmp"]:
+        if insn.offset in f.apirefs:
+            return
+
+        smda_report = insn.smda_function.smda_report
+        if insn.offset in f.outrefs:
+            for target in f.outrefs[insn.offset]:
+                if smda_report.getSection(insn.offset) != smda_report.getSection(target):
+                    yield Characteristic("cross section flow"), insn.offset
+        elif insn.operands.startswith("0x"):
+            target = int(insn.operands, 16)
+            if smda_report.getSection(insn.offset) != smda_report.getSection(target):
+                yield Characteristic("cross section flow"), insn.offset
+
+
+# this is a feature that's most relevant at the function scope,
+# however, its most efficient to extract at the instruction scope.
+def extract_function_calls_from(f, bb, insn):
+    if insn.mnemonic != "call":
+        return
+
+    if insn.offset in f.outrefs:
+        for outref in f.outrefs[insn.offset]:
+            yield Characteristic("calls from"), outref
+
+            if outref == f.offset:
+                # if we found a jump target and it's the function address
+                # mark as recursive
+                yield Characteristic("recursive call"), outref
+    if insn.offset in f.apirefs:
+        yield Characteristic("calls from"), f.apirefs[insn.offset]
+
+
+# this is a feature that's most relevant at the function or basic block scope,
+# however, its most efficient to extract at the instruction scope.
+def extract_function_indirect_call_characteristic_features(f, bb, insn):
+    """
+    extract indirect function call characteristic (e.g., call eax or call dword ptr [edx+4])
+    does not include calls like => call ds:dword_ABD4974
+    """
+    if insn.mnemonic != "call":
+        return
+    if insn.operands.startswith("0x"):
+        return False
+    if "qword ptr" in insn.operands and "rip" in insn.operands:
+        return False
+    if insn.operands.startswith("dword ptr [0x"):
+        return False
+    # call edx
+    # call dword ptr [eax+50h]
+    # call qword ptr [rsp+78h]
+    yield Characteristic("indirect call"), insn.offset
+
+
+def extract_features(f, bb, insn):
+    """
+    extract features from the given insn.
+
+    args:
+      f (smda.common.SmdaFunction): the function to process.
+      bb (smda.common.SmdaBasicBlock): the basic block to process.
+      insn (smda.common.SmdaInstruction): the instruction to process.
+
+    yields:
+      Feature, set[VA]: the features and their location found in this insn.
+    """
+    for insn_handler in INSTRUCTION_HANDLERS:
+        for feature, va in insn_handler(f, bb, insn):
+            yield feature, va
+
+
+INSTRUCTION_HANDLERS = (
+    extract_insn_api_features,
+    extract_insn_number_features,
+    extract_insn_string_features,
+    extract_insn_bytes_features,
+    extract_insn_offset_features,
+    extract_insn_nzxor_characteristic_features,
+    extract_insn_mnemonic_features,
+    extract_insn_peb_access_characteristic_features,
+    extract_insn_cross_section_cflow,
+    extract_insn_segment_access_features,
+    extract_function_calls_from,
+    extract_function_indirect_call_characteristic_features,
+)
diff --git a/capa/main.py b/capa/main.py
index f1ca7760..5b6e1a54 100644
--- a/capa/main.py
+++ b/capa/main.py
@@ -295,7 +295,19 @@ class UnsupportedRuntimeError(RuntimeError):
 
 
 def get_extractor_py3(path, format, disable_progress=False):
-    raise UnsupportedRuntimeError()
+    from smda.SmdaConfig import SmdaConfig
+    from smda.Disassembler import Disassembler
+
+    import capa.features.extractors.smda
+
+    smda_report = None
+    with halo.Halo(text="analyzing program", spinner="simpleDots", stream=sys.stderr, enabled=not disable_progress):
+        config = SmdaConfig()
+        config.STORE_BUFFER = True
+        smda_disasm = Disassembler(config)
+        smda_report = smda_disasm.disassembleFile(path)
+
+    return capa.features.extractors.smda.SmdaFeatureExtractor(smda_report, path)
 
 
 def get_extractor(path, format, disable_progress=False):
@@ -446,14 +458,23 @@ def main(argv=None):
     parser = argparse.ArgumentParser(
         description=desc, epilog=epilog, formatter_class=argparse.RawDescriptionHelpFormatter
     )
-    parser.add_argument(
-        # in #328 we noticed that the sample path is not handled correctly if it contains non-ASCII characters
-        # https://stackoverflow.com/a/22947334/ offers a solution and decoding using getfilesystemencoding works
-        # in our testing, however other sources suggest `sys.stdin.encoding` (https://stackoverflow.com/q/4012571/)
-        "sample",
-        type=lambda s: s.decode(sys.getfilesystemencoding()),
-        help="path to sample to analyze",
-    )
+
+    if sys.version_info >= (3, 0):
+        parser.add_argument(
+            # Python 3 str handles non-ASCII arguments correctly
+            "sample",
+            type=str,
+            help="path to sample to analyze",
+        )
+    else:
+        parser.add_argument(
+            # in #328 we noticed that the sample path is not handled correctly if it contains non-ASCII characters
+            # https://stackoverflow.com/a/22947334/ offers a solution and decoding using getfilesystemencoding works
+            # in our testing, however other sources suggest `sys.stdin.encoding` (https://stackoverflow.com/q/4012571/)
+            "sample",
+            type=lambda s: s.decode(sys.getfilesystemencoding()),
+            help="path to sample to analyze",
+        )
     parser.add_argument("--version", action="version", version="%(prog)s {:s}".format(capa.version.__version__))
     parser.add_argument(
         "-r",
@@ -550,7 +571,7 @@ def main(argv=None):
             # during the load of the RuleSet, we extract subscope statements into their own rules
             # that are subsequently `match`ed upon. this inflates the total rule count.
             # so, filter out the subscope rules when reporting total number of loaded rules.
-            len(filter(lambda r: "capa/subscope-rule" not in r.meta, rules.rules.values())),
+            len([i for i in filter(lambda r: "capa/subscope-rule" not in r.meta, rules.rules.values())]),
         )
         if args.tag:
             rules = rules.filter_rules_by_meta(args.tag)
diff --git a/setup.py b/setup.py
index 4c09dd04..4cf4aa8b 100644
--- a/setup.py
+++ b/setup.py
@@ -28,6 +28,7 @@ requirements = [
 if sys.version_info >= (3, 0):
     # py3
     requirements.append("networkx")
+    requirements.append("smda")
 else:
     # py2
     requirements.append("enum34==1.1.6")  # v1.1.6 is needed by halo 0.0.30 / spinners 0.0.24
diff --git a/tests/fixtures.py b/tests/fixtures.py
index 3ff40f6c..640bc222 100644
--- a/tests/fixtures.py
+++ b/tests/fixtures.py
@@ -81,6 +81,21 @@ def get_viv_extractor(path):
     return capa.features.extractors.viv.VivisectFeatureExtractor(vw, path)
 
 
+@lru_cache()
+def get_smda_extractor(path):
+    from smda.SmdaConfig import SmdaConfig
+    from smda.Disassembler import Disassembler
+
+    import capa.features.extractors.smda
+
+    config = SmdaConfig()
+    config.STORE_BUFFER = True
+    disasm = Disassembler(config)
+    report = disasm.disassembleFile(path)
+
+    return capa.features.extractors.smda.SmdaFeatureExtractor(report, path)
+
+
 @lru_cache()
 def extract_file_features(extractor):
     features = collections.defaultdict(set)
@@ -129,6 +144,8 @@ def get_data_path_by_name(name):
         return os.path.join(CD, "data", "Practical Malware Analysis Lab 21-01.exe_")
     elif name == "al-khaser x86":
         return os.path.join(CD, "data", "al-khaser_x86.exe_")
+    elif name == "al-khaser x64":
+        return os.path.join(CD, "data", "al-khaser_x64.exe_")
     elif name.startswith("39c05"):
         return os.path.join(CD, "data", "39c05b15e9834ac93f206bc114d0a00c357c888db567ba8f5345da0529cbed41.dll_")
     elif name.startswith("499c2"):
@@ -377,7 +394,7 @@ FEATURE_PRESENCE_TESTS = [
     ),
     ("kernel32-64", "function=0x1800202B0", capa.features.insn.API("RtlCaptureContext"), True),
     # insn/api: x64 nested thunk
-    ("82bf6", "function=0x140059342", capa.features.insn.API("ElfClearEventLogFile"), True),
+    ("al-khaser x64", "function=0x14004B4F0", capa.features.insn.API("__vcrt_GetModuleHandle"), True),
     # insn/api: call via jmp
     ("mimikatz", "function=0x40B3C6", capa.features.insn.API("LocalFree"), True),
     ("c91887...", "function=0x40156F", capa.features.insn.API("CloseClipboard"), True),
@@ -473,7 +490,7 @@ def do_test_feature_count(get_extractor, sample, scope, feature, expected):
 
 def get_extractor(path):
     if sys.version_info >= (3, 0):
-        raise RuntimeError("no supported py3 backends yet")
+        extractor = get_smda_extractor(path)
     else:
         extractor = get_viv_extractor(path)
 
diff --git a/tests/test_main.py b/tests/test_main.py
index 6ceae34a..783fc95c 100644
--- a/tests/test_main.py
+++ b/tests/test_main.py
@@ -19,7 +19,6 @@ import capa.features
 from capa.engine import *
 
 
-@pytest.mark.xfail(sys.version_info >= (3, 0), reason="vivsect only works on py2")
 def test_main(z9324d_extractor):
     # tests rules can be loaded successfully and all output modes
     path = z9324d_extractor.path
@@ -29,7 +28,6 @@ def test_main(z9324d_extractor):
     assert capa.main.main([path]) == 0
 
 
-@pytest.mark.xfail(sys.version_info >= (3, 0), reason="vivsect only works on py2")
 def test_main_single_rule(z9324d_extractor, tmpdir):
     # tests a single rule can be loaded successfully
     RULE_CONTENT = textwrap.dedent(
@@ -58,7 +56,6 @@ def test_main_single_rule(z9324d_extractor, tmpdir):
     )
 
 
-@pytest.mark.xfail(sys.version_info >= (3, 0), reason="vivsect only works on py2")
 def test_main_non_ascii_filename(pingtaest_extractor, tmpdir, capsys):
     # on py2.7, need to be careful about str (which can hold bytes)
     #  vs unicode (which is only unicode characters).
@@ -71,18 +68,22 @@ def test_main_non_ascii_filename(pingtaest_extractor, tmpdir, capsys):
     std = capsys.readouterr()
     # but here, we have to use a unicode instance,
     # because capsys has decoded the output for us.
-    assert pingtaest_extractor.path.decode("utf-8") in std.out
+    if sys.version_info >= (3, 0):
+        assert pingtaest_extractor.path in std.out
+    else:
+        assert pingtaest_extractor.path.decode("utf-8") in std.out
 
 
-@pytest.mark.xfail(sys.version_info >= (3, 0), reason="vivsect only works on py2")
 def test_main_non_ascii_filename_nonexistent(tmpdir, caplog):
     NON_ASCII_FILENAME = "täst_not_there.exe"
     assert capa.main.main(["-q", NON_ASCII_FILENAME]) == -1
 
-    assert NON_ASCII_FILENAME.decode("utf-8") in caplog.text
+    if sys.version_info >= (3, 0):
+        assert NON_ASCII_FILENAME in caplog.text
+    else:
+        assert NON_ASCII_FILENAME.decode("utf-8") in caplog.text
 
 
-@pytest.mark.xfail(sys.version_info >= (3, 0), reason="vivsect only works on py2")
 def test_main_shellcode(z499c2_extractor):
     path = z499c2_extractor.path
     assert capa.main.main([path, "-vv", "-f", "sc32"]) == 0
@@ -137,7 +138,6 @@ def test_ruleset():
     assert len(rules.basic_block_rules) == 1
 
 
-@pytest.mark.xfail(sys.version_info >= (3, 0), reason="vivsect only works on py2")
 def test_match_across_scopes_file_function(z9324d_extractor):
     rules = capa.rules.RuleSet(
         [
@@ -201,7 +201,6 @@ def test_match_across_scopes_file_function(z9324d_extractor):
     assert ".text section and install service" in capabilities
 
 
-@pytest.mark.xfail(sys.version_info >= (3, 0), reason="vivsect only works on py2")
 def test_match_across_scopes(z9324d_extractor):
     rules = capa.rules.RuleSet(
         [
@@ -264,7 +263,6 @@ def test_match_across_scopes(z9324d_extractor):
     assert "kill thread program" in capabilities
 
 
-@pytest.mark.xfail(sys.version_info >= (3, 0), reason="vivsect only works on py2")
 def test_subscope_bb_rules(z9324d_extractor):
     rules = capa.rules.RuleSet(
         [
@@ -289,7 +287,6 @@ def test_subscope_bb_rules(z9324d_extractor):
     assert "test rule" in capabilities
 
 
-@pytest.mark.xfail(sys.version_info >= (3, 0), reason="vivsect only works on py2")
 def test_byte_matching(z9324d_extractor):
     rules = capa.rules.RuleSet(
         [
@@ -312,7 +309,6 @@ def test_byte_matching(z9324d_extractor):
     assert "byte match test" in capabilities
 
 
-@pytest.mark.xfail(sys.version_info >= (3, 0), reason="vivsect only works on py2")
 def test_count_bb(z9324d_extractor):
     rules = capa.rules.RuleSet(
         [
@@ -336,7 +332,6 @@ def test_count_bb(z9324d_extractor):
     assert "count bb" in capabilities
 
 
-@pytest.mark.xfail(sys.version_info >= (3, 0), reason="vivsect only works on py2")
 def test_fix262(pma16_01_extractor, capsys):
     # tests rules can be loaded successfully and all output modes
     path = pma16_01_extractor.path
@@ -347,7 +342,6 @@ def test_fix262(pma16_01_extractor, capsys):
     assert "www.practicalmalwareanalysis.com" not in std.out
 
 
-@pytest.mark.xfail(sys.version_info >= (3, 0), reason="vivsect only works on py2")
 def test_not_render_rules_also_matched(z9324d_extractor, capsys):
     # rules that are also matched by other rules should not get rendered by default.
     # this cuts down on the amount of output while giving approx the same detail.
diff --git a/tests/test_smda_features.py b/tests/test_smda_features.py
new file mode 100644
index 00000000..68790f8d
--- /dev/null
+++ b/tests/test_smda_features.py
@@ -0,0 +1,30 @@
+# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import sys
+
+from fixtures import *
+
+
+@parametrize(
+    "sample,scope,feature,expected",
+    FEATURE_PRESENCE_TESTS,
+    indirect=["sample", "scope"],
+)
+def test_smda_features(sample, scope, feature, expected):
+    with xfail(sys.version_info < (3, 0), reason="SMDA only works on py3"):
+        do_test_feature_presence(get_smda_extractor, sample, scope, feature, expected)
+
+
+@parametrize(
+    "sample,scope,feature,expected",
+    FEATURE_COUNT_TESTS,
+    indirect=["sample", "scope"],
+)
+def test_smda_feature_counts(sample, scope, feature, expected):
+    with xfail(sys.version_info < (3, 0), reason="SMDA only works on py3"):
+        do_test_feature_count(get_smda_extractor, sample, scope, feature, expected)