Skip to content

[SM6.10] Reserve LinAlg HLSL intrinsics #8026

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
V-FEXrt merged 2 commits into from
Dec 31, 2025
+115 −3
Merged

[SM6.10] Reserve LinAlg HLSL intrinsics #8026

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
25fd2c8
[SM6.10] Reserve LinAlg HLSL intrinsics
V-FEXrt Dec 22, 2025
66dab2f
chore: autopublish 2025-12-31T19:23:00Z
github-actions[bot] Dec 31, 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
22 changes: 21 additions & 1 deletion include/dxc/HlslIntrinsicOp.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -111,6 +111,26 @@ enum class IntrinsicOp {
IOP_WorldToObject = 99,
IOP_WorldToObject3x4 = 100,
IOP_WorldToObject4x3 = 101,
IOP___builtin_LinAlg_CopyConvertMatrix = 405,
Copy link

@inbelic inbelic Dec 31, 2025

Choose a reason for hiding this comment

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

nit: Is there a required order for these? In the other files they seemed to be sorted by opcode

Copy link
Collaborator Author

@V-FEXrt V-FEXrt Dec 31, 2025

Choose a reason for hiding this comment

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

Ah yeah, this is an interesting topic. There is some info below but there are a lot of moving parts. Happy to chat more if below is unclear

  • This is generated code, and this is where it was placed by the generator
  • They are actually ordered, but they are ordered alphabetically not by opcode value. The extra __ for __builtin somewhat hides that fact since it's not visually obvious
  • The values here are internal so they can be renumbered without issue but it was decided they shouldn't be

This may raise the question "how is it that 99% of the opcodes have the number that matches their alphabetic order" to which the answer is: until very recently the generator would renumber every opcode any time a new intrinsic was added. This created a lot of ordering churn, large diffs, and merge conflicts and was generally not a good thing to be doing so it was turned off

Copy link

@inbelic inbelic Dec 31, 2025

Choose a reason for hiding this comment

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

I see, sounds good

IOP___builtin_LinAlg_CreateMatrix = 406,
IOP___builtin_LinAlg_FillMatrix = 407,
IOP___builtin_LinAlg_MatrixAccumulate = 416,
IOP___builtin_LinAlg_MatrixAccumulateToDescriptor = 420,
IOP___builtin_LinAlg_MatrixAccumulateToMemory = 421,
IOP___builtin_LinAlg_MatrixGetCoordinate = 408,
IOP___builtin_LinAlg_MatrixGetElement = 409,
IOP___builtin_LinAlg_MatrixLength = 410,
IOP___builtin_LinAlg_MatrixLoadFromDescriptor = 411,
IOP___builtin_LinAlg_MatrixLoadFromMemory = 412,
IOP___builtin_LinAlg_MatrixMatrixMultiply = 417,
IOP___builtin_LinAlg_MatrixMatrixMultiplyAccumulate = 418,
IOP___builtin_LinAlg_MatrixOuterProduct = 422,
IOP___builtin_LinAlg_MatrixQueryAccumulatorLayout = 419,
IOP___builtin_LinAlg_MatrixSetElement = 413,
IOP___builtin_LinAlg_MatrixStoreToDescriptor = 414,
IOP___builtin_LinAlg_MatrixStoreToMemory = 415,
IOP___builtin_LinAlg_MatrixVectorMultiply = 423,
IOP___builtin_LinAlg_MatrixVectorMultiplyAdd = 424,
IOP___builtin_MatVecMul = 390,
IOP___builtin_MatVecMulAdd = 391,
IOP___builtin_OuterProductAccumulate = 392,
Expand Down Expand Up @@ -411,7 +431,7 @@ enum class IntrinsicOp {
IOP_usign = 355,
MOP_InterlockedUMax = 356,
MOP_InterlockedUMin = 357,
Num_Intrinsics = 405,
Num_Intrinsics = 425,
};
inline bool HasUnsignedIntrinsicOpcode(IntrinsicOp opcode) {
switch (opcode) {
Expand Down
41 changes: 41 additions & 0 deletions lib/HLSL/HLOperationLower.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7536,6 +7536,47 @@ constexpr IntrinsicLower gLowerTable[] = {
DXIL::OpCode::RayQuery_CommittedTriangleObjectPosition},
{IntrinsicOp::MOP_DxHitObject_TriangleObjectPosition, EmptyLower,
DXIL::OpCode::HitObject_TriangleObjectPosition},

{IntrinsicOp::IOP___builtin_LinAlg_CopyConvertMatrix, EmptyLower,
DXIL::OpCode::CopyConvertMatrix},
{IntrinsicOp::IOP___builtin_LinAlg_CreateMatrix, EmptyLower,
DXIL::OpCode::CreateMatrix},
{IntrinsicOp::IOP___builtin_LinAlg_FillMatrix, EmptyLower,
DXIL::OpCode::FillMatrix},
{IntrinsicOp::IOP___builtin_LinAlg_MatrixGetCoordinate, EmptyLower,
DXIL::OpCode::MatrixGetCoordinate},
{IntrinsicOp::IOP___builtin_LinAlg_MatrixGetElement, EmptyLower,
DXIL::OpCode::MatrixGetElement},
{IntrinsicOp::IOP___builtin_LinAlg_MatrixLength, EmptyLower,
DXIL::OpCode::MatrixLength},
{IntrinsicOp::IOP___builtin_LinAlg_MatrixLoadFromDescriptor, EmptyLower,
DXIL::OpCode::MatrixLoadFromDescriptor},
{IntrinsicOp::IOP___builtin_LinAlg_MatrixLoadFromMemory, EmptyLower,
DXIL::OpCode::MatrixLoadFromMemory},
{IntrinsicOp::IOP___builtin_LinAlg_MatrixSetElement, EmptyLower,
DXIL::OpCode::MatrixSetElement},
{IntrinsicOp::IOP___builtin_LinAlg_MatrixStoreToDescriptor, EmptyLower,
DXIL::OpCode::MatrixStoreToDescriptor},
{IntrinsicOp::IOP___builtin_LinAlg_MatrixStoreToMemory, EmptyLower,
DXIL::OpCode::MatrixStoreToMemory},
{IntrinsicOp::IOP___builtin_LinAlg_MatrixAccumulate, EmptyLower,
DXIL::OpCode::MatrixAccumulate},
{IntrinsicOp::IOP___builtin_LinAlg_MatrixMatrixMultiply, EmptyLower,
DXIL::OpCode::MatrixMulOp},
{IntrinsicOp::IOP___builtin_LinAlg_MatrixMatrixMultiplyAccumulate,
EmptyLower, DXIL::OpCode::MatrixMulOp},
{IntrinsicOp::IOP___builtin_LinAlg_MatrixQueryAccumulatorLayout, EmptyLower,
DXIL::OpCode::MatrixQueryAccumulatorLayout},
{IntrinsicOp::IOP___builtin_LinAlg_MatrixAccumulateToDescriptor, EmptyLower,
DXIL::OpCode::MatrixAccumulateToDescriptor},
{IntrinsicOp::IOP___builtin_LinAlg_MatrixAccumulateToMemory, EmptyLower,
DXIL::OpCode::MatrixAccumulateToMemory},
{IntrinsicOp::IOP___builtin_LinAlg_MatrixOuterProduct, EmptyLower,
DXIL::OpCode::MatrixOuterProduct},
{IntrinsicOp::IOP___builtin_LinAlg_MatrixVectorMultiply, EmptyLower,
DXIL::OpCode::MatrixVecMul},
{IntrinsicOp::IOP___builtin_LinAlg_MatrixVectorMultiplyAdd, EmptyLower,
DXIL::OpCode::MatrixVecMulAdd},
};
constexpr size_t NumLowerTableEntries =
sizeof(gLowerTable) / sizeof(gLowerTable[0]);
Expand Down
31 changes: 31 additions & 0 deletions utils/hct/gen_intrin_main.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -395,6 +395,37 @@ void [[min_sm=6.10]] __builtin_OuterProductAccumulate(in LinAlg<c> InputVector1,

void [[min_sm=6.10]] __builtin_VectorAccumulate(in LinAlg<c> InputVector, in RWByteAddressBuffer MatrixBuffer, in uint MatrixOffset);

// LinAlg intrinsics

// TODO: Update return type for CreateMatrix to MatrixRef
// TODO: Replace all int MatrixRef with MatrixRef type
// TODO: Replace all int GroupSharedMem with groupshared memory
void [[min_sm=6.10]] __builtin_LinAlg_CreateMatrix();
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);

} namespace


Expand Down
24 changes: 22 additions & 2 deletions utils/hct/hlsl_intrinsic_opcodes.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
{
"IntrinsicOpCodes": {
"Num_Intrinsics": 405,
"Num_Intrinsics": 425,
"IOP_AcceptHitAndEndSearch": 0,
"IOP_AddUint64": 1,
"IOP_AllMemoryBarrier": 2,
Expand Down Expand Up @@ -405,6 +405,26 @@
"IOP_TriangleObjectPosition": 401,
"MOP_CandidateTriangleObjectPosition": 402,
"MOP_CommittedTriangleObjectPosition": 403,
"MOP_DxHitObject_TriangleObjectPosition": 404
"MOP_DxHitObject_TriangleObjectPosition": 404,
"IOP___builtin_LinAlg_CopyConvertMatrix": 405,
"IOP___builtin_LinAlg_CreateMatrix": 406,
"IOP___builtin_LinAlg_FillMatrix": 407,
"IOP___builtin_LinAlg_MatrixGetCoordinate": 408,
"IOP___builtin_LinAlg_MatrixGetElement": 409,
"IOP___builtin_LinAlg_MatrixLength": 410,
"IOP___builtin_LinAlg_MatrixLoadFromDescriptor": 411,
"IOP___builtin_LinAlg_MatrixLoadFromMemory": 412,
"IOP___builtin_LinAlg_MatrixSetElement": 413,
"IOP___builtin_LinAlg_MatrixStoreToDescriptor": 414,
"IOP___builtin_LinAlg_MatrixStoreToMemory": 415,
"IOP___builtin_LinAlg_MatrixAccumulate": 416,
"IOP___builtin_LinAlg_MatrixMatrixMultiply": 417,
"IOP___builtin_LinAlg_MatrixMatrixMultiplyAccumulate": 418,
"IOP___builtin_LinAlg_MatrixQueryAccumulatorLayout": 419,
"IOP___builtin_LinAlg_MatrixAccumulateToDescriptor": 420,
"IOP___builtin_LinAlg_MatrixAccumulateToMemory": 421,
"IOP___builtin_LinAlg_MatrixOuterProduct": 422,
"IOP___builtin_LinAlg_MatrixVectorMultiply": 423,
"IOP___builtin_LinAlg_MatrixVectorMultiplyAdd": 424
}
}