Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Multiple disassembler errors on vector instructions #2266

UnlimitedChild opened this issue Dec 15, 2019 · 1 comment

Multiple disassembler errors on vector instructions #2266

UnlimitedChild opened this issue Dec 15, 2019 · 1 comment


Copy link

@UnlimitedChild UnlimitedChild commented Dec 15, 2019


This comment has been minimized.

Copy link

@mappzor mappzor commented Dec 22, 2019

This is actually more related to Zydis rather than x64dbg directly nevertheless it's an interesting case. By looking at syntax used in your example code I assume you were using MASM64 to compile that. The problem is that MASM is quite permissive and allows you to specify things like zeroing-mask in places where you should not be allowed to do so, generating invalid instructions as a result.

If you look up VMOVAPS in Intel's manual you can find that following encoding is allowed:
EVEX.512.0F.W0 29 /r VMOVAPS zmm2/m512 {k1}{z}, zmm1
However above line actually describes two different cases: "register-copy" form when first operand is zmm register and "store-form" which is what you are using (first operand is memory). If you look at pseudocode provided in the manual you can notice that zeroing-mask is actually allowed ONLY for register-copy form but not for store-form.

Therefore according to official documentation those instructions

VMOVAPS zmmword ptr [4 * r14 + r14 + 123456h]{k1}{z} , zmm31
VMOVAPS zmmword ptr [4 * r14 + r14 + 123456h]{k2}{z} , zmm30
VMOVAPS zmmword ptr [4 * r14 + r14 + 123456h]{k3}{z} , zmm29
VMOVAPS zmmword ptr [4 * r14 + r14 + 123456h]{k4}{z} , zmm28
VMOVAPS zmmword ptr [4 * r14 + r14 + 123456h]{k5}{z} , zmm27

are all INVALID, so Zydis handles them correctly by failing to recognize them.

Were you able to test how those instructions behave on real hardware? Unfortunately I don't have any CPU with AVX-512 support to check this. I would expect that they either cause #UD or silently ignore EVEX.z. In latter case it would mean that there are valid yet undocumented alternative encodings not only for VMOVAPS but for several other AVX-512 instructions with similar semantics.

@mrexodia mrexodia added the bug label Jan 16, 2020
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
None yet
3 participants
You can’t perform that action at this time.