-
Notifications
You must be signed in to change notification settings - Fork 4.7k
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
[RISC-V] Fix GitHub_* tests #88640
[RISC-V] Fix GitHub_* tests #88640
Conversation
src/coreclr/jit/codegenriscv64.cpp
Outdated
@@ -6510,12 +6510,19 @@ void CodeGen::genJmpMethod(GenTree* jmp) | |||
{ | |||
var_types loadType = TYP_UNDEF; | |||
|
|||
// NOTE for RISCV64: not supports the HFA. | |||
assert(!varDsc->lvIsHfaRegArg()); |
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.
HFA is ARM only concept. I think this assert isn't needed.
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.
Removed. Thank you.
@@ -6532,14 +6539,21 @@ void CodeGen::genJmpMethod(GenTree* jmp) | |||
regSet.AddMaskVars(genRegMask(argReg)); | |||
gcInfo.gcMarkRegPtrVal(argReg, loadType); | |||
|
|||
if (compiler->lvaIsMultiregStruct(varDsc, compiler->info.compIsVarArgs)) | |||
if (varDsc->GetOtherArgReg() < REG_STK) |
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.
It's code from LA64, when I made genJmpMethod()
implementation, I've took ARM64 version since it's more clear. E.g. what's the meaning of if (varDsc->GetOtherArgReg() < REG_STK)
?
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.
Because ARM64 and RISCV have different calling convention. Right, ARM64 seems more clearer, but RISCV follows LA64's. So we need to update like LA64.
GetOtherArgReg
is the second register for multi argument register for RISCV and LA64.
In case of ARM64 in the test, it passes two float register arguments using one int register (REG_A0). So compiler->lvaIsMultiregStruct
returns false
. In method prolog in a callee, it makes two float values from REG_A0 in ARM64.
argRegNext = genRegArgNext(argReg); | ||
argRegNext = varDsc->GetOtherArgReg(); | ||
|
||
if (emitter::isFloatReg(argRegNext)) |
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 why special case for float is need?
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.
Because it cannot get the exact type by using varDsc->GetLayout()->GetGCPtrType(1)
(only returns TYP_I_IMPL for all primitive and non-gc value type) and GetEmitter()->emitIns_R_S
needs the exact type for the arguments.
If you want to update calling convention like ARM64, we can do. However, I think it is better to investigate after RISC-V is stable.
@jkotas @jakobbotsch Could you please review? Thank you so much. |
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.
The virtualcallstubcpu.hpp change looks good.
@jakobbotsch Could you please review the JIT change?
genJmpMethod
incodegenriscv64.cpp
Part of #84834
cc @wscho77 @HJLeee @JongHeonChoi @t-mustafin @alpencolt @gbalykov