Thanks to visit codestin.com
Credit goes to github.com

Skip to content

Expose the FusedMultiplyAdd and MultiplyAddEstimate APIs on relevant vector and scalar types #102181

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

Merged
merged 10 commits into from
May 16, 2024
Merged

Expose the FusedMultiplyAdd and MultiplyAddEstimate APIs on relevant vector and scalar types #102181

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
46caa78
Expose FusedMultiplyAdd and MultiplyAddEstimate on the scalar and vec…
tannergooding May 10, 2024
940862e
Adding tests covering FusedMultiplyAdd and MultiplyAddEstimate for th…
tannergooding May 11, 2024
d5b2ad9
Ensure TensorPrimitives uses the xplat APIs on .NET 9+
tannergooding May 11, 2024
731258f
Apply formatting patch
tannergooding May 14, 2024
4010235
Fix an accidental change to GenericVectorTests
tannergooding May 14, 2024
8a60cff
Ensure Arm64 passes fma operands in the correct order
tannergooding May 14, 2024
5e1e5eb
Apply formatting patch
tannergooding May 14, 2024
85425c2
Ensure all the Arm64 code paths are spilling and swapping operands
tannergooding May 14, 2024
0896be6
Apply formatting patch
tannergooding May 14, 2024
fdd2a83
Don't pop the stack value twice
tannergooding May 14, 2024
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
Prev Previous commit
Next Next commit
Ensure Arm64 passes fma operands in the correct order
  • Loading branch information
@tannergooding
tannergooding committed May 14, 2024
commit 8a60cff4f43b062c057e8055bd65f0835d000f2a
5 changes: 5 additions & 0 deletions src/coreclr/jit/gentree.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23433,6 +23433,11 @@ GenTree* Compiler::gtNewSimdFmaNode(
{
intrinsic = NI_AdvSimd_FusedMultiplyAdd;
}

// AdvSimd.FusedMultiplyAdd expects (addend, left, right), while the APIs take (left, right, addend)
// We expect op1 and op2 to have already been spilled

std::swap(op1, op3);
Copy link
Member

Choose a reason for hiding this comment

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

I am just curious - who's responsible to spill them?

Copy link
Member

Choose a reason for hiding this comment

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

Ah, nvm, I see

Copy link
Member Author

Choose a reason for hiding this comment

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

Typically it's the importer code, prior to calling this method. Latter phases that might call such an API (like morph) are responsible for ensuring the swap is safe in the face of potential side effects.

There's, unfortunately, not really a way for us to do such a swap safely in the API itself (at least that I know of) nor to know if the caller has actually done it. So the typical approach has been to do the swap and comment that callers should be doing the validation.

#else
#error Unsupported platform
#endif // !TARGET_XARCH && !TARGET_ARM64
Expand Down
6 changes: 6 additions & 0 deletions src/coreclr/jit/hwintrinsicarm64.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1353,6 +1353,12 @@ GenTree* Compiler::impSpecialIntrinsic(NamedIntrinsic intrinsic,
assert(sig->numArgs == 3);
assert(varTypeIsFloating(simdBaseType));

impSpillSideEffect(true, verCurrentState.esStackDepth -
3 DEBUGARG("Spilling op1 side effects for FusedMultiplyAdd"));

impSpillSideEffect(true, verCurrentState.esStackDepth -
2 DEBUGARG("Spilling op2 side effects for FusedMultiplyAdd"));

op3 = impSIMDPopStack();
op2 = impSIMDPopStack();
op1 = impSIMDPopStack();
Expand Down
6 changes: 6 additions & 0 deletions src/coreclr/jit/importercalls.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4001,6 +4001,12 @@ GenTree* Compiler::impIntrinsic(GenTree* newobjThis,
// Vector64.Create{ScalarUnsafe}(x)
// ).ToScalar();

impSpillSideEffect(true, verCurrentState.esStackDepth -
3 DEBUGARG("Spilling op1 side effects for FusedMultiplyAdd"));

impSpillSideEffect(true, verCurrentState.esStackDepth -
2 DEBUGARG("Spilling op2 side effects for FusedMultiplyAdd"));

GenTree* op3 = gtNewSimdCreateScalarUnsafeNode(TYP_SIMD8, impPopStack().val, callJitType, 8);
GenTree* op2 = gtNewSimdCreateScalarUnsafeNode(TYP_SIMD8, impPopStack().val, callJitType, 8);
GenTree* op1 = gtNewSimdCreateScalarUnsafeNode(TYP_SIMD8, impPopStack().val, callJitType, 8);
Expand Down
Loading