-
Notifications
You must be signed in to change notification settings - Fork 172
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
Added constraint for SXLEN and UXLEN #530
base: master
Are you sure you want to change the base?
Conversation
I'm not sure what point this is when they're all hard-coded to the same value? If this is part of a larger series to do dynamic XLEN then sure, but that is a quite large undertaking. |
model/riscv_types.sail
Outdated
0b01 => 32, | ||
0b10 => 64, | ||
0b11 => 128, | ||
0b00 => 0 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Tab not spaces
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
And this should really be an abort, not silently give a junk result...
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Tab not spaces
sorry about that, fixed
And this should really be an abort, not silently give a junk result...
replace with Option(int)
Agree, It's really of no use if you can make sure of it. The biggest use maybe to reassure beginners who don’t know much about model.
Not yet. But I discussed the issue of dynamic xlen with some people, about usage scenarios and requirements. Here are some conclusions:
Personally, I would do a little research on how to implement this in the model |
I have thought quite a bit about dynamic XLEN, and the biggest problem is the CSRs that change width. Essentially what happens is as soon as you change the appropriate bits to change UXLEN or SXLEN, then any UXLEN or SXLEN register gets width-modulated down (or up) to the new size, according to the CSR width modulation algorithm in the spec, and what this means is a field specified in a register at UXLEN - N really does 'move' in some sense when UXLEN changes. The width-modulation algorithm (at least as I read it) is destructive and appears to allow producing illegal values in registers, so presumably some magic needs to happen to ensure that the width-modulated register remains legal. Other architectures do the sensible thing of saying registers are exactly 64-bits and then defining the 32-bit view of the register when executing 32-bit code, or providing two registers for the lower and upper 32-bits (RISC-V does this for some registers). There are also some CSRs that are defined as current XLEN-width, so presumably they need to be width-modulated every time you change modes that have a different XLEN? |
According to the spec:
SXLEN≥UXLEN
.MXLEN=64
, if S-mode is not supported, then SXL is read-only zero.MXLEN=64
, if U-mode is not supported, then UXL is read-only zero.This PR adds assertion and some conditional statements to satisfy the above constraints.