diff options
author | Kartik Agaram <vc@akkartik.com> | 2018-07-25 17:09:54 -0700 |
---|---|---|
committer | Kartik Agaram <vc@akkartik.com> | 2018-07-25 17:09:54 -0700 |
commit | d9ea0ed1bc4dd53b102a8788120bc6b966c9f2f2 (patch) | |
tree | 7c9bb57a983fee7b334d71af8bea9840f794b219 /subx | |
parent | 3d3cb2a166b54e0ce990e420c64f9c37a88f0d7e (diff) | |
download | mu-d9ea0ed1bc4dd53b102a8788120bc6b966c9f2f2.tar.gz |
4410
Temporarily delete support for /imm tags. We'll bring it back momentarily in a totally different way.
Diffstat (limited to 'subx')
-rw-r--r-- | subx/024transform_immediate.cc | 145 |
1 files changed, 0 insertions, 145 deletions
diff --git a/subx/024transform_immediate.cc b/subx/024transform_immediate.cc deleted file mode 100644 index 0451a7c9..00000000 --- a/subx/024transform_immediate.cc +++ /dev/null @@ -1,145 +0,0 @@ -//: Having to manually translate numbers into hex and enter them in -//: little-endian order is tedious and error-prone. Let's automate the -//: translation. -//: -//: We'll convert any immediate operands from decimal to hex and emit the -//: appropriate number of bytes. If they occur in a non-code segment we'll -//: raise an error. - -:(scenario translate_immediate_constants) -== 0x1 -# opcode ModR/M SIB displacement immediate -# instruction mod, reg, Reg/Mem bits scale, index, base -# 1-3 bytes 0/1 byte 0/1 byte 0/1/2/4 bytes 0/1/2/4 bytes - bb 42/imm32 # copy 42 to EBX -+translate: converting '42/imm32' to '2a 00 00 00' -+run: copy imm32 0x0000002a to EBX - -#: we don't have a testable instruction using 8-bit immediates yet, so can't run this instruction -:(scenarios transform) -:(scenario translate_imm8) -== 0x1 - cd 128/imm8 -+translate: converting '128/imm8' to '80' -:(scenarios run) - -:(before "End One-time Setup") -Transform.push_back(transform_immediate); - -:(code) -void transform_immediate(program& p) { - if (p.segments.empty()) return; - transform_immediate(p.segments.at(0)); - for (int i = 1; i < SIZE(p.segments); ++i) - flag_immediate(p.segments.at(i)); -} - -void transform_immediate(segment& seg) { - for (int i = 0; i < SIZE(seg.lines); ++i) - for (int j = 0; j < SIZE(seg.lines.at(i).words); ++j) - transform_immediate(seg.lines.at(i).words, j); -} - -void transform_immediate(vector<word>& line, int index) { - assert(index < SIZE(line)); - if (contains_metadata(line.at(index), "imm32")) - transform_imm32(line, index); - else if (contains_metadata(line.at(index), "imm8")) - transform_imm8(line.at(index)); -} - -bool contains_metadata(const word& curr, const string& m) { - for (int k = 0; k < SIZE(curr.metadata); ++k) - if (curr.metadata.at(k) == m) - return true; - return false; -} - -void transform_imm8(word& w) { - // convert decimal to hex - uint32_t val = parse_int(w.data); - if (trace_contains_errors()) return; - if (val > 0xff) { - raise << "invalid /imm8 word " << w.data << '\n' << end(); - return; - } - w.data = serialize_hex(val); - trace("translate") << "converting '" << w.original << "' to '" << w.data << "'" << end(); -} - -void transform_imm32(vector<word>& line, int index) { - vector<word>::iterator find(vector<word>&, int); - vector<word>::iterator x = find(line, index); - uint32_t val = parse_int(x->data); - if (trace_contains_errors()) return; - string orig = x->original; - x = line.erase(x); - emit_octets(line, x, val, orig); -} - -vector<word>::iterator find(vector<word>& l, int index) { - if (index >= SIZE(l)) { - raise << "find: index too large: " << index << " vs " << SIZE(l) << '\n' << end(); - return l.end(); - } - vector<word>::iterator result = l.begin(); - for (int i = 0; i < index; ++i) - ++result; - return result; -} - -void emit_octets(vector<word>& line, vector<word>::iterator pos, uint32_t val, const string& orig) { - vector<word> new_data; - for (int i = 0; i < /*num bytes*/4; ++i) { - word tmp; - tmp.data = serialize_hex(val & 0xff); // little-endian - new_data.push_back(tmp); - val = val >> 8; - } - trace("translate") << "converting '" << orig << "' to '" << to_string(new_data) << "'" << end(); - line.insert(pos, new_data.begin(), new_data.end()); -} - -string to_string(const vector<word>& in) { - ostringstream out; - for (int i = 0; i < SIZE(in); ++i) { - if (i > 0) out << ' '; - out << HEXBYTE << in.at(i).data; - } - return out.str(); -} - -string serialize_hex(const int val) { - ostringstream out; - out << std::hex << val; - return out.str(); -} - -void flag_immediate(const segment& seg) { - for (int i = 0; i < SIZE(seg.lines); ++i) { - const vector<word>& line = seg.lines.at(i).words; - for (int j = 0; j < SIZE(line); ++j) { - if (contains_metadata(line.at(j), "imm32") - || contains_metadata(line.at(j), "imm8")) - raise << "/imm8 and /imm32 only permitted in code segments, and we currently only allow the very first segment to be code.\n" << end(); - } - } -} - -// helper -void transform(const string& text_bytes) { - program p; - istringstream in(text_bytes); - parse(in, p); - if (trace_contains_errors()) return; - transform(p); -} - -:(scenario translate_immediate_constants_hex) -== 0x1 -# opcode ModR/M SIB displacement immediate -# instruction mod, reg, Reg/Mem bits scale, index, base -# 1-3 bytes 0/1 byte 0/1 byte 0/1/2/4 bytes 0/1/2/4 bytes - bb 0x2a/imm32 # copy 42 to EBX -+translate: converting '0x2a/imm32' to '2a 00 00 00' -+run: copy imm32 0x0000002a to EBX |