[x86 disassembler] crc32 not disassembled correctly

[llvmbot]

Bugzilla Link	10702
Resolution	FIXED
Resolved on	Nov 07, 2018 00:17
Version	trunk
OS	All
Blocks	llvm/llvm-bugzilla-archive#10988 llvm/llvm-bugzilla-archive#10832
Reporter	LLVM Bugzilla Contributor
CC	@topperc

Extended Description

From the "Intel® 64 and IA-32 Architectures Software Developer’s Manual
Volume 2 (2A & 2B): Instruction Set Reference, A-Z", p. 3-236:
CRC32 — Accumulate CRC32 Value
F2 0F 38 F0 /r

Clang on OSX assembles:
crc32 %al, %eax

To:
F2 0F 38 F0 C0

With llvm-mc built from trunk revision 135913:
$ echo '0xf2 0x0f 0x38 0xf0 0xc0'| ./llvm-mc -disassemble -triple="x86_64"
:1:1: warning: invalid instruction encoding
0xf2 0x0f 0x38 0xf0 0xc0
^
:1:21: warning: invalid instruction encoding
0xf2 0x0f 0x38 0xf0 0xc0

[llvmbot]

assigned to @topperc

[llvmbot]

Fixed this in r138771.

[topperc]

Reopening because the following case doesn't work

echo '0x66 0xf2 0x0f 0x38 0xf1 0xc0'| ./llvm-mc -disassemble -triple="x86_64"
crc32l %ax, %ax

Should be "crc32w %ax, %eax"

The disassembler doesn't really know what to do with 0x66 and 0xf2 on the same instruction and fails to match the opcode correctly. The destination operand needs to ignore the 0x66 prefix.

Also this should decode to MOVBE which I don't think is supported at all in LLVM.

echo '0x0f 0x38 0xf1 0xc0'| ./llvm-mc -disassemble -triple="x86_64
crc32l %eax, %eax

[llvmbot]

Also this should decode to MOVBE which I don't think is supported at all in
LLVM.

echo '0x0f 0x38 0xf1 0xc0'| ./llvm-mc -disassemble -triple="x86_64
crc32l %eax, %eax

I filed bug 10832 to track the movbe bug.

[llvmbot]

Fix the disassembly of the X86 "crc32w %ax, %eax" instruction in r139014.

[topperc]

But now this doesn't decode correctly

0xf2 0x0f 0x38 0xf1 0xc0

[topperc]

Fixed in r140954.

[llvmbot]

Testing with r140997:

$ echo '0xf2 0x0f 0x38 0xf0 0xc0'| ./llvm-mc -disassemble -triple="x86_64" -x86-asm-syntax=intel
crc32 EAX, AL
$ echo '0x66 0xf2 0x0f 0x38 0xf1 0xc0'| ./llvm-mc -disassemble -triple="x86_64" -x86-asm-syntax=intel
crc32 EAX, AX
$ echo '0xf2 0x0f 0x38 0xf1 0xc0'| ./llvm-mc -disassemble -triple="x86_64" -x86-asm-syntax=intel
crc32 EAX, EAX
$ echo '0xf2 0x48 0x0f 0x38 0xf1 0xc0'| ./llvm-mc -disassemble -triple="x86_64" -x86-asm-syntax=intel
crc32 RAX, RAX

I think those are all of the expected cases; movbe is still broken as of this build.

[llvmbot]

mentioned in issue llvm/llvm-bugzilla-archive#10832

[bcardosolopes]

mentioned in issue llvm/llvm-bugzilla-archive#10988