[Wasm RyuJIT] Initial scaffolding for calls and helper calls #123044
Conversation
kg
added
arch-wasm
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch |
kg
force-pushed
the
wasm-genemithelpercall
branch
from
741ddb8 to
e805a10
Compare
AndyAyersMS
left a comment
AndyAyersMS
left a comment
There was a problem hiding this comment.
We still need to figure out in detail how to specify what method we want to call.
We know roughly what kind of Wasm we want to produce (per the calling convention doc we'll be making lots of indirect calls) but we need help from the JIT host to make this all work out.
I think it might be ok for now to just defer this part of the work and try and get everything else lined up?
| // Generate a direct call to a non-virtual user defined or helper method | ||
| assert(call->IsHelperCall() || (call->gtCallType == CT_USER_FUNC)); | ||
|
|
||
| if (call->gtEntryPoint.addr != NULL) |
There was a problem hiding this comment.
Per the calling convention, for user and helper calls we'll also be invoking them indirectly.
There was a problem hiding this comment.
Right. I'm not clear on whether that means there won't be an addr filled in
kg
force-pushed
the
wasm-genemithelpercall
branch
from
b95ce18 to
499127d
Compare
|
This branch now hits this in crossgen2: |
|
For this: [MethodImpl(MethodImplOptions.NoInlining)]
static void voidFunc () {
}
[MethodImpl(MethodImplOptions.NoInlining)]
static void callVoidFunc () {
voidFunc();
}R2R now compiles without crashing or asserting, and disasmo generates this: ; Method Program:callVoidFunc() (FullOpts)
G_M45341_IG01: ;; offset=0x0000
local.cnt 4
local[0] type=i32
local[1] type=i64
local[2] type=f32
local[3] type=f64
;; size=9 bbWeight=1 PerfScore 5.00
G_M45341_IG02: ;; offset=0x0009
i32.const 140722950757336
i32.load 0 0
call_indirect 0 0
;; size=14 bbWeight=1 PerfScore 3.00
G_M45341_IG03: ;; offset=0x0017
end
;; size=1 bbWeight=1 PerfScore 1.00
; Total bytes of code: 24It's not correct but it's a start. |
|
cc @dotnet/jit-contrib need to figure out how this should actually be factored and shaped and implemented, now that it "works". I don't understand 75% of the code I touched here, for example whether we use EA_8BYTE or a different one still makes no sense to me, and the way we historically separate out |
|
|
||
| genConsumeRegs(op1); | ||
|
|
||
| genEmitHelperCall(CORINFO_HELP_THROWNULLREF, 0, EA_UNKNOWN, REG_NA); |
There was a problem hiding this comment.
why does GT_NULLCHECK just unconditionally calls CORINFO_HELP_THROWNULLREF?
There was a problem hiding this comment.
Was trying to just get the JIT to emit a helper call, since Andy is working on actual null checks. But I never found a way to get a GT_NULLCHECK node
There was a problem hiding this comment.
Something like that should?
unsafe void Foo(int* x)
{
_ = *x;
}TransformUnusedIndirection is expected to convert this unused GT_IND into GT_NULLCHECK
There was a problem hiding this comment.
Thanks for pointing this out though, I had forgotten to put a comment in
There was a problem hiding this comment.
Something like that should?
unsafe void Foo(int* x) { _ = *x; }
TransformUnusedIndirectionis expected to convert this unused GT_IND into GT_NULLCHECK
This indeed appears to work! Thank you very much
3 participants
Depends on #123021
Starting to put together all the code we need for helper calls and calls in general.