Skip to content

Expand double vector dot products to FMul/FMad instead of emitting invalid Dot2/3/4 intrinsics #8

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

Draft
Copilot wants to merge 3 commits into
base: main
Choose a base branch
from
+40 −0
Draft

Expand double vector dot products to FMul/FMad instead of emitting invalid Dot2/3/4 intrinsics #8

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
b41944b
Initial plan
Copilot Mar 13, 2026
14224ea
Fix mul with double vectors producing invalid DXIL dot intrinsic
Copilot Mar 13, 2026
45f72a9
Expand test to cover both mul and dot with double vectors
Copilot Mar 13, 2026
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
7 changes: 7 additions & 0 deletions lib/HLSL/HLOperationLower.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2590,6 +2590,13 @@ Value *ExpandDot(Value *arg0, Value *arg1, unsigned vecSize, hlsl::OP *hlslOP,

Value *TranslateFDot(Value *arg0, Value *arg1, unsigned vecSize,
hlsl::OP *hlslOP, IRBuilder<> &Builder) {
// Dot2/3/4 DXIL operations only support half and float, not double.
// For double vectors, expand the dot product using FMul and FMad.
if (arg0->getType()->getScalarType()->isDoubleTy()) {
return ExpandDot(arg0, arg1, vecSize, hlslOP, Builder,
DXIL::OpCode::FMad);
}

switch (vecSize) {
case 2:
return TrivialDotOperation(OP::OpCode::Dot2, arg0, arg1, hlslOP, Builder);
Expand Down
33 changes: 33 additions & 0 deletions tools/clang/test/HLSLFileCheck/hlsl/intrinsics/mul/mul_double.hlsl
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,33 @@
// RUN: %dxc -T cs_6_0 -E main -DFUNC=mul -DDIM=4 %s | FileCheck %s
// RUN: %dxc -T cs_6_0 -E main -DFUNC=mul -DDIM=3 %s | FileCheck %s
// RUN: %dxc -T cs_6_0 -E main -DFUNC=mul -DDIM=2 %s | FileCheck %s
// RUN: %dxc -T cs_6_0 -E main -DFUNC=dot -DDIM=4 %s | FileCheck %s
// RUN: %dxc -T cs_6_0 -E main -DFUNC=dot -DDIM=3 %s | FileCheck %s
// RUN: %dxc -T cs_6_0 -E main -DFUNC=dot -DDIM=2 %s | FileCheck %s

// Verify that mul and dot of double vectors do not produce invalid DXIL Dot
// intrinsics (Dot2/3/4 only support half and float). Instead, the dot product
// should be expanded using FMul and FMad.
// %dxc runs validation, so the test implicitly verifies DXIL validity.

// CHECK-NOT: call double @dx.op.dot2.f64
// CHECK-NOT: call double @dx.op.dot3.f64
// CHECK-NOT: call double @dx.op.dot4.f64
// CHECK: fmul fast double
// CHECK: call double @dx.op.tertiary.f64(i32 46,

#if DIM == 4
typedef double4 DVec;
#elif DIM == 3
typedef double3 DVec;
#else
typedef double2 DVec;
#endif

RWStructuredBuffer<DVec> In;
RWStructuredBuffer<double> Out;

[numthreads(1, 1, 1)]
void main() {
Out[0] = FUNC(In[0], In[1]);
}