Skip to content

[SM6.10] LinAlg spec review: identify signature gaps vs 0035-linalg-matrix spec #8180

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 2 commits into
base: main
Choose a base branch
from
+20 −25
Draft

[SM6.10] LinAlg spec review: identify signature gaps vs 0035-linalg-matrix spec #8180

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e9413bb
[SM6.10] Update HLOp gen intrinsics to latest shape
V-FEXrt Feb 4, 2026
4ca0603
Initial plan
Copilot Feb 20, 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
45 changes: 20 additions & 25 deletions utils/hct/gen_intrin_main.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -397,32 +397,27 @@ void [[min_sm=6.10]] __builtin_VectorAccumulate(in LinAlg<c> InputVector, in RWB

// LinAlg intrinsics

// TODO: Replace all int MatrixRef with MatrixRef type
// TODO: Replace all int GroupSharedMem with groupshared memory
void [[min_sm=6.10]] __builtin_LinAlg_FillMatrix(int MatrixRef, numeric value);
void [[min_sm=6.10]] __builtin_LinAlg_CopyConvertMatrix(int MatrixRefDest, int MatrixRefSrc, bool transpose);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixLoadFromDescriptor(int MatrixRef, resource buf, int32_only offset, int32_only stride, int32_only layout);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixLoadFromMemory(int MatrixRef, int GroupSharedMem, int32_only offset, int32_only stride, int32_only layout);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixLength(int MatrixRef);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixGetCoordinate(int MatrixRef, int32_only threadLocalIndex);
numeric [[min_sm=6.10]] __builtin_LinAlg_MatrixGetElement(int MatrixRef, int32_only threadLocalIndex);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixSetElement(int MatrixRef, int32_only threadLocalIndex, numeric value);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixStoreToDescriptor(int MatrixRef, resource buf, int32_only offset, int32_only stride, int32_only layout);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixStoreToMemory(int MatrixRef, int GroupSharedMem, int32_only offset, int32_only stride, int32_only layout);
int32_only [[min_sm=6.10]] __builtin_LinAlg_MatrixQueryAccumulatorLayout();
void [[min_sm=6.10]] __builtin_LinAlg_MatrixMatrixMultiply(int MatrixRefA, int MatrixRefB, int MatrixRefC);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixMatrixMultiplyAccumulate(int MatrixRefA, int MatrixRefB, int MatrixRefC);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixAccumulate(int MatrixRefRHS, int MatrixRefLHS);

// TODO: Fix vector types
void [[min_sm=6.10]] __builtin_LinAlg_MatrixVectorMultiply(int MatrixRef);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixVectorMultiplyAdd(int MatrixRef);

void [[min_sm=6.10]] __builtin_LinAlg_MatrixAccumulateToDescriptor(int MatrixRef, resource buf, int32_only offset, int32_only stride, int32_only layout);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixAccumulateToMemory(int MatrixRef, int GroupSharedMem, int32_only offset, int32_only stride, int32_only layout);

// TODO: Fix vector types
void [[min_sm=6.10]] __builtin_LinAlg_MatrixOuterProduct(int MatrixRef);
void [[min_sm=6.10]] __builtin_LinAlg_FillMatrix(out LinAlgMatrix ret, in numeric value);
void [[min_sm=6.10]] __builtin_LinAlg_CopyConvertMatrix(out LinAlgMatrix ret, in LinAlgMatrix source, in bool transpose);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixLoadFromDescriptor(out LinAlgMatrix ret, in ByteAddressBuffer buf, in uint offset, in uint stride, in uint layout);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixLoadFromDescriptor(out LinAlgMatrix ret, in RWByteAddressBuffer buf, in uint offset, in uint stride, in uint layout);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixLoadFromMemory(out LinAlgMatrix ret, in int GroupSharedMem, in uint offset, in uint stride, in uint layout);
uint [[min_sm=6.10]] __builtin_LinAlg_MatrixLength(in LinAlgMatrix matrix);
uint<2> [[min_sm=6.10]] __builtin_LinAlg_MatrixGetCoordinate(in LinAlgMatrix matrix, in uint threadLocalIndex);
numeric [[min_sm=6.10]] __builtin_LinAlg_MatrixGetElement(in LinAlgMatrix matrix, in uint threadLocalIndex);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixSetElement(out LinAlgMatrix ret, in LinAlgMatrix matrix, in uint threadLocalIndex, in numeric value);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixStoreToDescriptor(in LinAlgMatrix matrix, in RWByteAddressBuffer buf, in uint offset, in uint stride, in uint layout);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixStoreToMemory(in LinAlgMatrix matrix, in int GroupSharedMem, in uint offset, in uint stride, in uint layout);
uint [[min_sm=6.10]] __builtin_LinAlg_MatrixQueryAccumulatorLayout();
void [[min_sm=6.10]] __builtin_LinAlg_MatrixMatrixMultiply(out LinAlgMatrix matrixC, in LinAlgMatrix matrixA, in LinAlgMatrix matrixB);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixMatrixMultiplyAccumulate(out LinAlgMatrix matrixC, in LinAlgMatrix matrixA, in LinAlgMatrix matrixB);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixAccumulate(out LinAlgMatrix matrixC, in LinAlgMatrix matrixLHS, in LinAlgMatrix matrixRHS);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixVectorMultiply(out numeric<> ret, in LinAlgMatrix mat, in numeric<> input, in uint input_interp);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixVectorMultiplyAdd(out numeric<> ret, in LinAlgMatrix mat, in numeric<> input, in uint input_interp, in numeric<> bias, in uint bias_interp);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixAccumulateToDescriptor(in LinAlgMatrix matrix, in RWByteAddressBuffer buf, in uint offset, in uint stride, in uint layout);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixAccumulateToMemory(in LinAlgMatrix matrix, in int GroupSharedMem, in uint offset, in uint stride, in uint layout);
void [[min_sm=6.10]] __builtin_LinAlg_MatrixOuterProduct(out LinAlgMatrix ret, in numeric<> vecA, in numeric<> vecB);

} namespace

Expand Down