rol, ror and rori have a similar bug in their Sail pseudo-code
slobodova opened this issue · comments
Chapter 2, Sail code for ror, rol and rori uses X(rs2) instead of X(rs1) in the shift. X(rs2) should be used only for the computation of shamt.
One more comment, I assume that when X(rd) is used on the left-hand-side and right-hand-side has more than XLEN bits, the low XLEN bits are taken. But I haven't seen any comment about it.
rol current:
let shamt = if xlen == 32
then X(rs2)[4..0]
else X(rs2)[5..0];
let result = (X(rs1) << shamt) | (X(rs2) >> (xlen - shamt));
X(rd) = result;
rol suggested:
let shamt = if xlen == 32
then X(rs2)[4..0]
else X(rs2)[5..0];
let result = (X(rs1) << shamt) | (X(rs1) >> (xlen - shamt));
X(rd) = result;
ror current:
let shamt = if xlen == 32
then X(rs2)[4..0]
else X(rs2)[5..0];
let result = (X(rs1) >> shamt) | (X(rs2) << (xlen - shamt));
X(rd) = result;
ror suggested:
let shamt = if xlen == 32
then X(rs2)[4..0]
else X(rs2)[5..0];
let result = (X(rs1) >> shamt) | (X(rs1) << (xlen - shamt));
X(rd) = result;
rori current:
let shamt = if xlen == 32
then shamt[4..0]
else shamt[5..0];
let result = (X(rs1) >> shamt) | (X(rs2) << (xlen - shamt));
X(rd) = result;
rori suggested:
let shamt = if xlen == 32
then shamt[4..0]
else shamt[5..0];
let result = (X(rs1) >> shamt) | (X(rs1) << (xlen - shamt));
X(rd) = result;
Thanks for the reviews of the SAIL code, very much appreciated; every helping hand on SAIL is currently needed.
Could you submit pull-requests for these changes? This would allow us to queue them up already and avoid any mistakes in applying your suggestions ourselves.
Do you have any special procedure for that?
The issue with the syntax of rol, ror, and rori is still not fixed in the bitmanip-1.0.0 public review document. Is this intentional or you are waiting on the next release. cc@cetola
This is fixed in one of the follow-on PDFs (1.0.0-.pdf)
However, I have been bringing up the need to create a final document repeatedly at Tech Chairs without any result.
I'll try to escalate this topic once more.
@ptomsich thanks