Skip to content

Ensure that the SSE fallback for Vector3.Dot masks off the unused element of op1 and op2 #74932

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
tannergooding merged 3 commits into from
Sep 2, 2022
Merged

Ensure that the SSE fallback for Vector3.Dot masks off the unused element of op1 and op2 #74932

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
5b49a4d
Ensure that the SSE fallback for Vector3.Dot masks off the unused ele…
tannergooding Sep 1, 2022
b19aec7
Applying formatting patch
tannergooding Sep 1, 2022
2945d9c
Ensure we use TYP_SIMD16 in for the simdType when generating the fall…
tannergooding Sep 1, 2022
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
66 changes: 49 additions & 17 deletions src/coreclr/jit/lowerxarch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3419,45 +3419,77 @@ GenTree* Lowering::LowerHWIntrinsicDot(GenTreeHWIntrinsic* node)
{
assert(simdBaseType == TYP_FLOAT);

// We need to mask off the most significant element to avoid the shuffle + add
// from including it in the computed result. We need to do this for both op1 and
// op2 in case one of them is `NaN` (because Zero * NaN == NaN)

simd16_t simd16Val = {};

simd16Val.i32[0] = -1;
simd16Val.i32[1] = -1;
simd16Val.i32[2] = -1;
simd16Val.i32[3] = +0;

simdType = TYP_SIMD16;
simdSize = 16;

// We will be constructing the following parts:
// ...
// +--* CNS_INT int -1
// +--* CNS_INT int -1
// +--* CNS_INT int -1
// +--* CNS_INT int 0
// tmp1 = * HWINTRINSIC simd16 T Create
// /--* op2 simd16
// /--* op1 simd16
// +--* tmp1 simd16
// op1 = * HWINTRINSIC simd16 T And
// ...

// This is roughly the following managed code:
// ...
// tmp1 = Vector128.Create(-1, -1, -1, 0);
// op1 = Sse.And(op1, tmp2);
// op1 = Sse.And(op1, tmp1);
// ...

GenTree* cns0 = comp->gtNewIconNode(-1, TYP_INT);
BlockRange().InsertAfter(op1, cns0);
GenTreeVecCon* vecCon1 = comp->gtNewVconNode(simdType, simdBaseJitType);
vecCon1->gtSimd16Val = simd16Val;

GenTree* cns1 = comp->gtNewIconNode(-1, TYP_INT);
BlockRange().InsertAfter(cns0, cns1);
BlockRange().InsertAfter(op1, vecCon1);

GenTree* cns2 = comp->gtNewIconNode(-1, TYP_INT);
BlockRange().InsertAfter(cns1, cns2);
op1 = comp->gtNewSimdHWIntrinsicNode(simdType, op1, vecCon1, NI_SSE_And, simdBaseJitType, simdSize);
BlockRange().InsertAfter(vecCon1, op1);

GenTree* cns3 = comp->gtNewIconNode(0, TYP_INT);
BlockRange().InsertAfter(cns2, cns3);
LowerNode(vecCon1);
LowerNode(op1);

tmp1 = comp->gtNewSimdHWIntrinsicNode(simdType, cns0, cns1, cns2, cns3, NI_Vector128_Create,
CORINFO_TYPE_INT, 16);
BlockRange().InsertAfter(cns3, tmp1);
// We will be constructing the following parts:
// ...
// +--* CNS_INT int -1
// +--* CNS_INT int -1
// +--* CNS_INT int -1
// +--* CNS_INT int 0
// tmp2 = * HWINTRINSIC simd16 T Create
// /--* op2 simd16
// +--* tmp2 simd16
// op2 = * HWINTRINSIC simd16 T And
// ...

op1 = comp->gtNewSimdHWIntrinsicNode(simdType, op1, tmp1, NI_SSE_And, simdBaseJitType, simdSize);
BlockRange().InsertAfter(tmp1, op1);
// This is roughly the following managed code:
// ...
// tmp2 = Vector128.Create(-1, -1, -1, 0);
// op2 = Sse.And(op2, tmp2);
// ...

LowerNode(tmp1);
LowerNode(op1);
GenTreeVecCon* vecCon2 = comp->gtNewVconNode(simdType, simdBaseJitType);
vecCon2->gtSimd16Val = simd16Val;

BlockRange().InsertAfter(op2, vecCon2);

op2 = comp->gtNewSimdHWIntrinsicNode(simdType, op2, vecCon2, NI_SSE_And, simdBaseJitType, simdSize);
BlockRange().InsertAfter(vecCon2, op2);

LowerNode(vecCon2);
LowerNode(op2);
}
}

Expand Down