Skip to content

test: update llvm ir expectations for autodiff tests when dlopen #149986

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.

Sign up for GitHub

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

Open
sgasho wants to merge 5 commits into
base: main
Choose a base branch
from
+42 −55
Open

test: update llvm ir expectations for autodiff tests when dlopen #149986

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
2408190
test: update llvm ir expectations for autodiff tests
sgasho Dec 14, 2025
5e6d5bb
update llvm ir expectations
sgasho Dec 17, 2025
348f324
Update tests/codegen-llvm/autodiff/batched.rs
sgasho Dec 18, 2025
10d8b6d
Update tests/codegen-llvm/autodiff/batched.rs
sgasho Dec 18, 2025
d9dc67a
Update tests/codegen-llvm/autodiff/abi_handling.rs
sgasho Dec 18, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions tests/codegen-llvm/autodiff/abi_handling.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -58,7 +58,7 @@ fn f1(x: &[f32; 2]) -> f32 {
// CHECK-NEXT: Function Attrs
// debug-NEXT: define internal { float, float }
// debug-SAME: (ptr %f, float %x, float %dret)
// release-NEXT: define internal fastcc float
// release-NEXT: define internal fastcc{{( noundef)?}} float
Copy link
Member

@workingjubilee workingjubilee Dec 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

good.

Copy link
Member

@ZuseZ4 ZuseZ4 Dec 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

fwiw autodiff only supports main, no older versions. So I'd not mark it as optional, as I hope that we don't regress. Or, at least I'd like to learn about regressions.

Copy link
Member

@workingjubilee workingjubilee Dec 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

sure.

Suggested change
// release-NEXT: define internal fastcc{{( noundef)?}} float
// release-NEXT: define internal fastcc noundef float

// release-SAME: (float noundef %x)

// CHECK-LABEL: ; abi_handling::f2
Expand All @@ -77,7 +77,7 @@ fn f2(f: fn(f32) -> f32, x: f32) -> f32 {
// CHECK-NEXT: Function Attrs
// debug-NEXT: define internal { float, float }
// debug-SAME: (ptr align 4 %x, ptr align 4 %bx_0, ptr align 4 %y, ptr align 4 %by_0)
// release-NEXT: define internal fastcc { float, float }
// release-NEXT: define internal fastcc
Copy link
Member

@workingjubilee workingjubilee Dec 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

what is its return type now?

// release-SAME: (float %x.0.val)

// CHECK-LABEL: ; abi_handling::f3
Expand Down
78 changes: 34 additions & 44 deletions tests/codegen-llvm/autodiff/batched.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,7 +23,7 @@ fn square(x: &f32) -> f32 {
}

// d_square2
// CHECK: define internal [4 x float] @fwddiffe4square(ptr noalias noundef readonly align 4 captures(none) dereferenceable(4) %x, [4 x ptr] %"x'")
// CHECK: define internal fastcc [4 x float] @fwddiffe4square(float %x.0.val, [4 x ptr] %"x'")
Copy link
Member

@workingjubilee workingjubilee Dec 17, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

the linkage probably matters to autodiff tests but the calling convention and the first argument maybe shouldn't, since this is optimization sensitive and we are just trying to verify that batching works: use regexes to ignore these details instead. like this, or just {{( fastcc)?}} like you do elsewhere

Suggested change
// CHECK: define internal fastcc [4 x float] @fwddiffe4square(float %x.0.val, [4 x ptr] %"x'")
// CHECK: define internal {{[ a-z_]*}} [4 x float] @fwddiffe4square({{[ a-z0-9()_]*}} %x{{.*}}, [4 x ptr] %"x'")

// CHECK-NEXT: start:
// CHECK-NEXT: %0 = extractvalue [4 x ptr] %"x'", 0
// CHECK-NEXT: %"_2'ipl" = load float, ptr %0, align 4
Expand All @@ -33,28 +33,23 @@ fn square(x: &f32) -> f32 {
// CHECK-NEXT: %"_2'ipl2" = load float, ptr %2, align 4
// CHECK-NEXT: %3 = extractvalue [4 x ptr] %"x'", 3
// CHECK-NEXT: %"_2'ipl3" = load float, ptr %3, align 4
// CHECK-NEXT: %_2 = load float, ptr %x, align 4
// CHECK-NEXT: %4 = fmul fast float %"_2'ipl", %_2
// CHECK-NEXT: %5 = fmul fast float %"_2'ipl1", %_2
// CHECK-NEXT: %6 = fmul fast float %"_2'ipl2", %_2
// CHECK-NEXT: %7 = fmul fast float %"_2'ipl3", %_2
// CHECK-NEXT: %8 = fmul fast float %"_2'ipl", %_2
// CHECK-NEXT: %9 = fmul fast float %"_2'ipl1", %_2
// CHECK-NEXT: %10 = fmul fast float %"_2'ipl2", %_2
// CHECK-NEXT: %11 = fmul fast float %"_2'ipl3", %_2
// CHECK-NEXT: %12 = fadd fast float %4, %8
// CHECK-NEXT: %13 = insertvalue [4 x float] undef, float %12, 0
// CHECK-NEXT: %14 = fadd fast float %5, %9
// CHECK-NEXT: %15 = insertvalue [4 x float] %13, float %14, 1
// CHECK-NEXT: %16 = fadd fast float %6, %10
// CHECK-NEXT: %17 = insertvalue [4 x float] %15, float %16, 2
// CHECK-NEXT: %18 = fadd fast float %7, %11
// CHECK-NEXT: %19 = insertvalue [4 x float] %17, float %18, 3
// CHECK-NEXT: ret [4 x float] %19
// CHECK-NEXT: }
// CHECK-NEXT: %4 = fadd fast float %"_2'ipl", %"_2'ipl"
// CHECK-NEXT: %5 = fmul fast float %4, %x.0.val
// CHECK-NEXT: %6 = insertvalue [4 x float] undef, float %5, 0
// CHECK-NEXT: %7 = fadd fast float %"_2'ipl1", %"_2'ipl1"
// CHECK-NEXT: %8 = fmul fast float %7, %x.0.val
// CHECK-NEXT: %9 = insertvalue [4 x float] %6, float %8, 1
// CHECK-NEXT: %10 = fadd fast float %"_2'ipl2", %"_2'ipl2"
// CHECK-NEXT: %11 = fmul fast float %10, %x.0.val
// CHECK-NEXT: %12 = insertvalue [4 x float] %9, float %11, 2
// CHECK-NEXT: %13 = fadd fast float %"_2'ipl3", %"_2'ipl3"
// CHECK-NEXT: %14 = fmul fast float %13, %x.0.val
Comment on lines +36 to +46
Copy link
Member

@workingjubilee workingjubilee Dec 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Unless ZuseZ4 prefers this sort of exact matching, this could use CHECK-DAG so we can capture the operations we're actually concerned about: four insertvalues, and an fadd and fmul apiece, in any order.

// CHECK-NEXT: %15 = insertvalue [4 x float] %12, float %14, 3
// CHECK-NEXT: ret [4 x float] %15
Copy link
Member

@workingjubilee workingjubilee Dec 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This one could use a capture here to avoid being so sensitive to specific values

Suggested change
// CHECK-NEXT: %15 = insertvalue [4 x float] %12, float %14, 3
// CHECK-NEXT: ret [4 x float] %15
// CHECK-NEXT: %[[RETVAR:[0-9]+]] = insertvalue [4 x float] %{{[0-9]+}, float %{{[0-9]+}, 3
// CHECK-NEXT: ret [4 x float] %[[RETVAR]]

// CHECK-NEXT: }

// d_square3, the extra float is the original return value (x * x)
// CHECK: define internal { float, [4 x float] } @fwddiffe4square.1(ptr noalias noundef readonly align 4 captures(none) dereferenceable(4) %x, [4 x ptr] %"x'")
// d_square1, the extra float is the original return value (x * x)
// CHECK: define internal fastcc { float, [4 x float] } @fwddiffe4square.{{[0-9]+}}(float %x.0.val, [4 x ptr] %"x'")
Copy link
Member

@workingjubilee workingjubilee Dec 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

same idea as my comment on R26

// CHECK-NEXT: start:
// CHECK-NEXT: %0 = extractvalue [4 x ptr] %"x'", 0
// CHECK-NEXT: %"_2'ipl" = load float, ptr %0, align 4
Expand All @@ -64,28 +59,23 @@ fn square(x: &f32) -> f32 {
// CHECK-NEXT: %"_2'ipl2" = load float, ptr %2, align 4
// CHECK-NEXT: %3 = extractvalue [4 x ptr] %"x'", 3
// CHECK-NEXT: %"_2'ipl3" = load float, ptr %3, align 4
// CHECK-NEXT: %_2 = load float, ptr %x, align 4
// CHECK-NEXT: %_0 = fmul float %_2, %_2
// CHECK-NEXT: %4 = fmul fast float %"_2'ipl", %_2
// CHECK-NEXT: %5 = fmul fast float %"_2'ipl1", %_2
// CHECK-NEXT: %6 = fmul fast float %"_2'ipl2", %_2
// CHECK-NEXT: %7 = fmul fast float %"_2'ipl3", %_2
// CHECK-NEXT: %8 = fmul fast float %"_2'ipl", %_2
// CHECK-NEXT: %9 = fmul fast float %"_2'ipl1", %_2
// CHECK-NEXT: %10 = fmul fast float %"_2'ipl2", %_2
// CHECK-NEXT: %11 = fmul fast float %"_2'ipl3", %_2
// CHECK-NEXT: %12 = fadd fast float %4, %8
// CHECK-NEXT: %13 = insertvalue [4 x float] undef, float %12, 0
// CHECK-NEXT: %14 = fadd fast float %5, %9
// CHECK-NEXT: %15 = insertvalue [4 x float] %13, float %14, 1
// CHECK-NEXT: %16 = fadd fast float %6, %10
// CHECK-NEXT: %17 = insertvalue [4 x float] %15, float %16, 2
// CHECK-NEXT: %18 = fadd fast float %7, %11
// CHECK-NEXT: %19 = insertvalue [4 x float] %17, float %18, 3
// CHECK-NEXT: %20 = insertvalue { float, [4 x float] } undef, float %_0, 0
// CHECK-NEXT: %21 = insertvalue { float, [4 x float] } %20, [4 x float] %19, 1
// CHECK-NEXT: ret { float, [4 x float] } %21
// CHECK-NEXT: }
// CHECK-NEXT: %_0 = fmul float %x.0.val, %x.0.val
// CHECK-NEXT: %4 = fadd fast float %"_2'ipl", %"_2'ipl"
// CHECK-NEXT: %5 = fmul fast float %4, %x.0.val
// CHECK-NEXT: %6 = insertvalue [4 x float] undef, float %5, 0
// CHECK-NEXT: %7 = fadd fast float %"_2'ipl1", %"_2'ipl1"
// CHECK-NEXT: %8 = fmul fast float %7, %x.0.val
// CHECK-NEXT: %9 = insertvalue [4 x float] %6, float %8, 1
// CHECK-NEXT: %10 = fadd fast float %"_2'ipl2", %"_2'ipl2"
// CHECK-NEXT: %11 = fmul fast float %10, %x.0.val
// CHECK-NEXT: %12 = insertvalue [4 x float] %9, float %11, 2
// CHECK-NEXT: %13 = fadd fast float %"_2'ipl3", %"_2'ipl3"
// CHECK-NEXT: %14 = fmul fast float %13, %x.0.val
// CHECK-NEXT: %15 = insertvalue [4 x float] %12, float %14, 3
// CHECK-NEXT: %16 = insertvalue { float, [4 x float] } undef, float %_0, 0
// CHECK-NEXT: %17 = insertvalue { float, [4 x float] } %16, [4 x float] %15, 1
// CHECK-NEXT: ret { float, [4 x float] } %17
// CHECK-NEXT: }
Comment on lines +62 to +78
Copy link
Member

@workingjubilee workingjubilee Dec 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

all I would say here is reiterating previous commentary


fn main() {
let x = std::hint::black_box(3.0);
Expand Down
9 changes: 3 additions & 6 deletions tests/codegen-llvm/autodiff/generic.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,12 +22,9 @@ fn square<T: std::ops::Mul<Output = T> + Copy>(x: &T) -> T {

// Ensure that `d_square::<f64>` code is generated even if `square::<f64>` was never called
//
// CHECK: ; generic::square
// CHECK-NEXT: ; Function Attrs:
// CHECK-NEXT: define internal {{.*}} double
// CHECK-NEXT: start:
// CHECK-NOT: ret
// CHECK: fmul double
// CHECK: define internal{{.*}} @diffe_{{.*}}generic6square
// CHECK-SAME: double
// CHECK: f{{add|mul}} {{.*}} double
Comment on lines +25 to +27
Copy link
Member

@workingjubilee workingjubilee Dec 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

hm. this looks very different, now?


fn main() {
let xf32: f32 = std::hint::black_box(3.0);
Expand Down
6 changes: 3 additions & 3 deletions tests/codegen-llvm/autodiff/identical_fnc.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,14 +27,14 @@ fn square2(x: &f64) -> f64 {

// CHECK:; identical_fnc::main
// CHECK-NEXT:; Function Attrs:
// CHECK-NEXT:define internal void @_ZN13identical_fnc4main17h6009e4f751bf9407E()
// CHECK-NEXT:define internal void @{{[^ (]+identical_fnc4main[^ (]*}}()
// CHECK-NEXT:start:
// CHECK-NOT:br
// CHECK-NOT:ret
// CHECK:; call identical_fnc::d_square
// CHECK-NEXT:call fastcc void @_ZN13identical_fnc8d_square[[HASH:.+]](double %x.val, ptr noalias noundef align 8 dereferenceable(8) %dx1)
// CHECK-NEXT:{{(tail )?}}call fastcc void @[[DSQUARE:[^ (]+identical_fnc8d_square[^ (]*]](double %x.val, ptr noalias noundef align 8 dereferenceable(8) %dx1)
// CHECK:; call identical_fnc::d_square
// CHECK-NEXT:call fastcc void @_ZN13identical_fnc8d_square[[HASH]](double %x.val, ptr noalias noundef align 8 dereferenceable(8) %dx2)
// CHECK-NEXT:{{(tail )?}}call fastcc void @[[DSQUARE]](double %x.val, ptr noalias noundef align 8 dereferenceable(8) %dx2)

fn main() {
let x = std::hint::black_box(3.0);
Expand Down
Loading