From d3c4249dc3f932021a93785d8b29af9c1bbe2164 Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista@microsoft.com>
Date: Thu, 10 Apr 2025 16:21:29 -0700
Subject: [PATCH 01/18] a start

---
 tools/clang/lib/Headers/hlsl/linalg.h         | 176 ++++++++++++++++++
 .../CodeGenDXIL/hlsl/linalg/.clang-format     |   2 +
 .../hlsl/linalg/make-interp-vec-errors.hlsl   |  38 ++++
 .../hlsl/linalg/mat-vec-mul-errors.hlsl       |  18 ++
 .../CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl  |  31 +++
 .../hlsl/linalg/mat-vec-muladd-errors.hlsl    |  18 ++
 .../hlsl/linalg/mat-vec-muladd.hlsl           |  96 ++++++++++
 .../linalg/outerproductaccumulate-errors.hlsl |  20 ++
 .../hlsl/linalg/outerproductaccumulate.hlsl   |  17 ++
 .../hlsl/linalg/vector-matrix-proposal.hlsl   | 143 ++++++++++++++
 .../hlsl/linalg/vectoraccumulate.hlsl         |  16 ++
 11 files changed, 575 insertions(+)
 create mode 100644 tools/clang/lib/Headers/hlsl/linalg.h
 create mode 100644 tools/clang/test/CodeGenDXIL/hlsl/linalg/.clang-format
 create mode 100644 tools/clang/test/CodeGenDXIL/hlsl/linalg/make-interp-vec-errors.hlsl
 create mode 100644 tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-errors.hlsl
 create mode 100644 tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl
 create mode 100644 tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd-errors.hlsl
 create mode 100644 tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd.hlsl
 create mode 100644 tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl
 create mode 100644 tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate.hlsl
 create mode 100644 tools/clang/test/CodeGenDXIL/hlsl/linalg/vector-matrix-proposal.hlsl
 create mode 100644 tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate.hlsl

diff --git a/tools/clang/lib/Headers/hlsl/linalg.h b/tools/clang/lib/Headers/hlsl/linalg.h
new file mode 100644
index 0000000000..f2e6e7bfc6
--- /dev/null
+++ b/tools/clang/lib/Headers/hlsl/linalg.h
@@ -0,0 +1,176 @@
+
+// Header for linear algebra APIs.
+//
+// This needs to find a proper home!
+//
+
+namespace dx {
+namespace linalg {
+
+// NOTE: can't be an enum class because we get this error:
+//     error: non-type template argument of type 'dx::linalg::DataType' is not
+//     an integral constant expression
+//
+enum DataType {
+  DATA_TYPE_SINT16 = 2,           // ComponentType::I16
+  DATA_TYPE_UINT16 = 3,           // ComponentType::U16
+  DATA_TYPE_SINT32 = 4,           // ComponentType::I32
+  DATA_TYPE_UINT32 = 5,           // ComponentType::U32
+  DATA_TYPE_FLOAT16 = 8,          // ComponentType::F16
+  DATA_TYPE_FLOAT32 = 9,          // ComponentType::F32
+  DATA_TYPE_SINT8_T4_PACKED = 17, // ComponentType::PackedS8x32
+  DATA_TYPE_UINT8_T4_PACKED = 18, // ComponentType::PackedU8x32
+  DATA_TYPE_UINT8 = 19,           // ComponentType::U8
+  DATA_TYPE_SINT8 = 20,           // ComponentType::I8
+  DATA_TYPE_FLOAT8_E4M3 = 21,     // ComponentType::F8_E4M3
+                                  // (1 sign, 4 exp, 3 mantissa bits)
+  DATA_TYPE_FLOAT8_E5M2 = 22,     // ComponentType::F8_E5M2
+                                  // (1 sign, 5 exp, 2 mantissa bits)
+};
+
+enum MatrixLayout {
+  MATRIX_LAYOUT_ROW_MAJOR = 0,
+  MATRIX_LAYOUT_COLUMN_MAJOR = 1,
+  MATRIX_LAYOUT_MUL_OPTIMAL = 2,
+  MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL = 3
+};
+
+//
+// Helper for signedness
+//
+namespace details {
+template <typename T> bool IsUnsigned() { return false; }
+
+#ifdef __HLSL_ENABLE_16_BIT
+template <> bool IsUnsigned<uint16_t>() { return true; }
+#endif
+
+template <> bool IsUnsigned<uint32_t>() { return true; }
+template <> bool IsUnsigned<uint64_t>() { return true; }
+} // namespace details
+
+//
+// (RW)MatrixRef
+//
+
+template <typename BufferTy, DataType DT, uint M, uint K, MatrixLayout ML,
+          bool Transpose>
+struct MatrixRefImpl {
+  BufferTy Buffer;
+  uint StartOffset;
+  uint Stride;
+};
+
+template <DataType DT, uint M, uint K, MatrixLayout ML, bool Transpose = false>
+using MatrixRef = MatrixRefImpl<ByteAddressBuffer, DT, M, K, ML, Transpose>;
+
+template <DataType DT, uint M, uint K, MatrixLayout ML, bool Transpose = false>
+using RWMatrixRef = MatrixRefImpl<RWByteAddressBuffer, DT, M, K, ML, Transpose>;
+
+//
+// (RW)VectorRef
+//
+
+template <typename BufferTy, DataType DT> struct VectorRefImpl {
+  BufferTy Buffer;
+  uint StartOffset;
+};
+
+template <DataType DT> using VectorRef = VectorRefImpl<ByteAddressBuffer, DT>;
+
+template <DataType DT>
+using RWVectorRef = VectorRefImpl<RWByteAddressBuffer, DT>;
+
+//
+// Vector
+//
+
+template <typename T, int N, DataType DT> struct InterpretedVector {
+  vector<T, N> Data;
+};
+
+template <DataType DT, typename T, int N>
+InterpretedVector<T, N, DT> MakeInterpretedVector(vector<T, N> Vec) {
+  InterpretedVector<T, N, DT> IV = {Vec};
+  return IV;
+}
+
+//
+// Mul
+//
+
+template <typename OutputElTy, typename InputElTy, int InputElCount,
+          typename MatrixBufferTy, DataType InputDT, DataType MatrixDT,
+          uint MatrixM, uint MatrixK, MatrixLayout MatrixLayout,
+          bool MatrixTranspose>
+vector<OutputElTy, MatrixM>
+Mul(MatrixRefImpl<MatrixBufferTy, MatrixDT, MatrixM, MatrixK, MatrixLayout,
+                  MatrixTranspose>
+        Matrix,
+    InterpretedVector<InputElTy, InputElCount, InputDT> InputVector) {
+
+  vector<OutputElTy, MatrixM> OutputVector;
+
+  __builtin_MatVecMul(
+      /*out*/ OutputVector, details::IsUnsigned<OutputElTy>(), InputVector.Data,
+      details::IsUnsigned<InputElTy>(), InputDT, Matrix.Buffer,
+      Matrix.StartOffset, MatrixDT, MatrixM, MatrixK, MatrixLayout,
+      MatrixTranspose, Matrix.Stride);
+
+  return OutputVector;
+}
+
+//
+// MulAdd
+//
+
+template <typename OutputElTy, typename InputElTy, int InputElCount,
+          typename MatrixBufferTy, DataType InputDT, DataType MatrixDT,
+          uint MatrixM, uint MatrixK, MatrixLayout MatrixLayout,
+          bool MatrixTranspose, typename BiasVectorBufferTy,
+          DataType BiasVectorDT>
+vector<OutputElTy, MatrixM>
+MulAdd(MatrixRefImpl<MatrixBufferTy, MatrixDT, MatrixM, MatrixK, MatrixLayout,
+                     MatrixTranspose>
+           Matrix,
+       InterpretedVector<InputElTy, InputElCount, InputDT> InputVector,
+       VectorRefImpl<BiasVectorBufferTy, BiasVectorDT> BiasVector) {
+
+  vector<OutputElTy, MatrixM> OutputVector;
+
+  __builtin_MatVecMulAdd(
+      /*out*/ OutputVector, details::IsUnsigned<OutputElTy>(), InputVector.Data,
+      details::IsUnsigned<InputElTy>(), InputDT, Matrix.Buffer,
+      Matrix.StartOffset, MatrixDT, MatrixM, MatrixK, MatrixLayout,
+      MatrixTranspose, Matrix.Stride, BiasVector.Buffer, BiasVector.StartOffset,
+      BiasVectorDT);
+
+  return OutputVector;
+}
+
+//
+// OuterProductAccumulate
+//
+
+template <typename ElTy, int MatrixM, int MatrixN, DataType MatrixDT,
+          MatrixLayout MatrixLayout>
+void OuterProductAccumulate(
+    vector<ElTy, MatrixM> InputVector1, vector<ElTy, MatrixN> InputVector2,
+    RWMatrixRef<MatrixDT, MatrixM, MatrixN, MatrixLayout, false> Matrix) {
+  __builtin_OuterProductAccumulate(InputVector1, InputVector2, Matrix.Buffer,
+                                   Matrix.StartOffset, MatrixDT, MatrixLayout,
+                                   Matrix.Stride);
+}
+
+//
+// VectorAccumulate
+//
+
+template <typename ElTy, int ElCount>
+void VectorAccumulate(vector<ElTy, ElCount> InputVector,
+                      RWByteAddressBuffer Buffer, uint Offset) {
+  __builtin_VectorAccumulate(InputVector, Buffer, Offset);
+}
+
+} // namespace linalg
+} // namespace dx
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/.clang-format b/tools/clang/test/CodeGenDXIL/hlsl/linalg/.clang-format
new file mode 100644
index 0000000000..70797e559f
--- /dev/null
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/.clang-format
@@ -0,0 +1,2 @@
+BasedOnStyle: LLVM
+ColumnLimit: 80
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/make-interp-vec-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/make-interp-vec-errors.hlsl
new file mode 100644
index 0000000000..a7e40cb374
--- /dev/null
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/make-interp-vec-errors.hlsl
@@ -0,0 +1,38 @@
+// RUN: not %dxc -T lib_6_9 %s 2>&1 | FileCheck %s
+
+#include "linalg.h"
+
+ByteAddressBuffer Buf;
+
+export float4 Test1(vector<float, 4> Input) {
+  using namespace dx::linalg;
+
+  MatrixRef<DATA_TYPE_UINT16, 4, 4, MATRIX_LAYOUT_MUL_OPTIMAL, true> Matrix = {
+      Buf, 0, 0};
+
+  // clang-format off
+  // CHECK: error: no matching function for call to 'MakeInterpretedVector'
+  // CHECK: note: candidate template ignored: invalid explicitly-specified argument for template parameter 'DT'
+  return Mul<float>(    
+      Matrix, MakeInterpretedVector<2>(Input));
+  // clang-format on
+}
+
+enum DataType {
+  DATA_TYPE_InvalidType = 40
+};
+
+export float4 Test2(vector<float, 4> Input) {
+  using namespace dx::linalg;
+
+  MatrixRef<DATA_TYPE_UINT16, 4, 4, MATRIX_LAYOUT_MUL_OPTIMAL, true> Matrix = {
+      Buf, 0, 0};
+
+  // clang-format off
+  // CHECK: error: no matching function for call to 'MakeInterpretedVector'
+  // CHECK: note: candidate template ignored: invalid explicitly-specified argument for template parameter 'DT'
+  return Mul<float>(    
+      Matrix, MakeInterpretedVector<DATA_TYPE_InvalidType>(Input));
+  // clang-format on
+}
+
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-errors.hlsl
new file mode 100644
index 0000000000..f60eb70e0f
--- /dev/null
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-errors.hlsl
@@ -0,0 +1,18 @@
+// RUN: not %dxc -T lib_6_9 %s 2>&1 | FileCheck %s
+
+#include "linalg.h"
+
+ByteAddressBuffer Buf;
+
+vector<float, 128> MixUpVectorAndMatrixArguments(vector<float, 128> Input) {
+  using namespace dx::linalg;
+
+  MatrixRef<DATA_TYPE_FLOAT16, 128, 128, MATRIX_LAYOUT_MUL_OPTIMAL> Matrix = {
+      Buf, 0, 0};
+
+  // clang-format off
+  // CHECK: error: no matching function for call to 'Mul'
+  // CHECK: note: candidate template ignored: could not match 'MatrixRefImpl' against 'InterpretedVector'
+  // clang-format on
+  return Mul<float>(MakeInterpretedVector<DATA_TYPE_FLOAT16>(Input), Matrix);
+}
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl
new file mode 100644
index 0000000000..4dbee65eaf
--- /dev/null
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl
@@ -0,0 +1,31 @@
+// RUN: %dxc -T lib_6_9 -enable-16bit-types %s | FileCheck %s
+
+#include "linalg.h"
+
+ByteAddressBuffer Buf;
+
+export float4 Test1(vector<float, 4> Input) {
+  using namespace dx::linalg;
+
+  MatrixRef<DATA_TYPE_FLOAT16, 4, 4, MATRIX_LAYOUT_MUL_OPTIMAL, true> Matrix = {
+      Buf, 0, 0};
+
+  // clang-format off
+  // CHECK: %{{.+}} = call <4 x float> @dx.op.matVecMul.v4f32.v4f32(i32 305, <4 x float> %{{.+}}, i1 false, i32 8, %dx.types.Handle %{{.+}}, i32 0, i32 8, i32 4, i32 4, i32 2, i1 true, i32 0, i1 false)
+  return Mul<float>(    
+      Matrix, MakeInterpretedVector<DATA_TYPE_FLOAT16>(Input));
+  // clang-format on
+}
+
+export vector<float, 8> Test2(vector<uint8_t4_packed, 6> Input) {
+  using namespace dx::linalg;
+
+  MatrixRef<DATA_TYPE_UINT8, 8, 6 * 4, MATRIX_LAYOUT_MUL_OPTIMAL> Matrix = {
+      Buf, 0, 0};
+
+  // clang-format off
+  // CHECK: %{{.+}} = call <8 x float> @dx.op.matVecMul.v8f32.v6f32(i32 305, <6 x float> %{{.+}}, i1 false, i32 18, %dx.types.Handle %{{.+}}, i32 0, i32 19, i32 8, i32 24, i32 2, i1 false, i32 0, i1 false)
+  // clang-format on
+  return Mul<float>(Matrix,
+                    MakeInterpretedVector<DATA_TYPE_UINT8_T4_PACKED>(Input));
+}
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd-errors.hlsl
new file mode 100644
index 0000000000..89a895accb
--- /dev/null
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd-errors.hlsl
@@ -0,0 +1,18 @@
+// RUN: not %dxc -T lib_6_9 %s 2>&1 | FileCheck %s
+
+#include "linalg.h"
+
+ByteAddressBuffer Buf;
+
+vector<float, 128> MixUpVectorAndMatrixArguments(vector<float, 128> Input) {
+  using namespace dx::linalg;
+
+  MatrixRef<1, 128, 128, MATRIX_LAYOUT_MUL_OPTIMAL> Matrix = {
+      Buf, 0, 0};
+
+  // clang-format off
+  // CHECK: error: no matching function for call to 'MulAdd'
+  // CHECK: note: candidate template ignored: could not match 'MatrixRefImpl' against 'InterpretedVector'
+  // clang-format on
+  return MulAdd<float>(MakeInterpretedVector<1>(Input), Matrix, MakeInterpretedVector<1>(Input));
+}
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd.hlsl
new file mode 100644
index 0000000000..e1efa3f929
--- /dev/null
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd.hlsl
@@ -0,0 +1,96 @@
+// RUN: %dxc -T lib_6_9 %s | FileCheck %s
+
+#include "linalg.h"
+
+ByteAddressBuffer Buf;
+
+export float4 Test1(float4 input) {
+  using namespace dx::linalg;
+
+  MatrixRef<DATA_TYPE_FLOAT16, 4, 4, MATRIX_LAYOUT_MUL_OPTIMAL> matrix = {Buf,
+                                                                          0, 0};
+  VectorRef<DATA_TYPE_FLOAT16> biasVector = {Buf, 256};
+
+  InterpretedVector<float, 4, DATA_TYPE_FLOAT16> theVector = {input};
+
+  // clang-format off
+  // CHECK: %{{.+}} = call <4 x float> @dx.op.matVecMulAdd.v4f32.v4f32(i32 306, <4 x float> %{{.+}}, i1 false, i32 8, %dx.types.Handle [[RES:%.+]], i32 0, i32 8, i32 4, i32 4, i32 2, i1 false, i32 0, %dx.types.Handle [[RES]], i32 256, i32 8, i1 false)
+  return MulAdd<float>(
+      matrix, theVector,
+      biasVector);
+  // clang-format on
+}
+
+export float4 Test2(float4 input) {
+  using namespace dx::linalg;
+
+  MatrixRef<DATA_TYPE_FLOAT16, 4, 4, MATRIX_LAYOUT_MUL_OPTIMAL, true> matrix = {
+      Buf, 0, 0};
+  VectorRef<DATA_TYPE_FLOAT16> biasVector = {Buf, 256};
+
+  InterpretedVector<float, 4, DATA_TYPE_FLOAT16> theVector = {input};
+
+  // clang-format off
+  // CHECK: %{{.+}} = call <4 x float> @dx.op.matVecMulAdd.v4f32.v4f32(i32 306, <4 x float> %{{.+}}, i1 false, i32 8, %dx.types.Handle [[RES:%.+]], i32 0, i32 8, i32 4, i32 4, i32 2, i1 true, i32 0, %dx.types.Handle [[RES]], i32 256, i32 8, i1 false)
+  return MulAdd<float>(
+      matrix, theVector,
+      biasVector);
+  // clang-format on
+}
+
+export float4 Test3(float4 input) {
+  using namespace dx::linalg;
+
+  MatrixRef<DATA_TYPE_FLOAT16, 4, 4, MATRIX_LAYOUT_MUL_OPTIMAL, true> matrix = {
+      Buf, 0, 0};
+  VectorRef<DATA_TYPE_FLOAT16> biasVector = {Buf, 256};
+
+  // clang-format off
+  // CHECK: %{{.+}} = call <4 x float> @dx.op.matVecMulAdd.v4f32.v4f32(i32 306, <4 x float> %{{.+}}, i1 false, i32 8, %dx.types.Handle [[RES:%.+]], i32 0, i32 8, i32 4, i32 4, i32 2, i1 true, i32 0, %dx.types.Handle [[RES]], i32 256, i32 8, i1 false)
+  return MulAdd<float>(
+      matrix, MakeInterpretedVector<DATA_TYPE_FLOAT16>(input),
+      biasVector);
+  // clang-format on
+}
+
+namespace ProposalExample {
+
+ByteAddressBuffer model;
+
+vector<float, 3> ApplyNeuralMaterial(vector<half, 8> inputVector) {
+  using namespace dx::linalg;
+
+  MatrixRef<DATA_TYPE_FLOAT8_E4M3, 32, 8, MATRIX_LAYOUT_MUL_OPTIMAL> matrix0 = {
+      model, 0, 0};
+
+  VectorRef<DATA_TYPE_FLOAT16> biasVector0 = {model, 1024};
+
+  MatrixRef<DATA_TYPE_FLOAT8_E4M3, 32, 32, MATRIX_LAYOUT_MUL_OPTIMAL> matrix1 =
+      {model, 2048, 0};
+
+  VectorRef<DATA_TYPE_FLOAT16> biasVector1 = {model, 3072};
+
+  MatrixRef<DATA_TYPE_FLOAT8_E4M3, 3, 32, MATRIX_LAYOUT_MUL_OPTIMAL> matrix2 = {
+      model, 4096, 0};
+
+  VectorRef<DATA_TYPE_FLOAT16> biasVector2 = {model, 5120};
+
+  vector<half, 32> layer0 = MulAdd<half>(
+      matrix0, MakeInterpretedVector<DATA_TYPE_FLOAT8_E4M3>(inputVector),
+      biasVector0);
+  layer0 = max(layer0, 0);
+
+  vector<half, 32> layer1 = MulAdd<half>(
+      matrix1, MakeInterpretedVector<DATA_TYPE_FLOAT8_E4M3>(layer0),
+      biasVector1);
+  layer1 = max(layer1, 0);
+
+  vector<float, 3> output = MulAdd<float>(
+      matrix2, MakeInterpretedVector<DATA_TYPE_FLOAT8_E4M3>(layer1),
+      biasVector2);
+  output = exp(output);
+
+  return output;
+}
+
+} // namespace ProposalExample
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl
new file mode 100644
index 0000000000..2729538720
--- /dev/null
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl
@@ -0,0 +1,20 @@
+// RUN: not %dxc -T lib_6_9 -enable-16bit-types %s 2>&1 | FileCheck %s
+
+#include "linalg.h"
+
+RWByteAddressBuffer RWBuf;
+
+export void Test4(vector<half, 128> Input1, vector<half, 64> Input2) {
+  using namespace dx::linalg;
+
+  RWMatrixRef<DATA_TYPE_FLOAT16, 128, 64, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL, true>
+      matrix = {RWBuf, 0, 0};
+
+  // clang-format off
+  // CHECK: error: no matching function for call to 'OuterProductAccumulate'
+  // CHECK: note: candidate template ignored: could not match 0 against 1
+  // __builtin_OuterProductAccumulate(Input1, Input2, RWBuf, 0, DATA_TYPE_FLOAT16, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL, 0);
+  // clang-format on
+
+  OuterProductAccumulate(Input1, Input2, matrix);  
+}
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate.hlsl
new file mode 100644
index 0000000000..c9b0d7410e
--- /dev/null
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate.hlsl
@@ -0,0 +1,17 @@
+// RUN: %dxc -T lib_6_9 -enable-16bit-types %s | FileCheck %s
+
+RWByteAddressBuffer RWBuf;
+
+export void Test4(vector<half, 128> Input1, vector<half, 64> Input2) {
+  using namespace dx::linalg;
+
+  RWMatrixRef<DATA_TYPE_FLOAT16, 128, 64, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL>
+      matrix = {RWBuf, 0, 0};
+
+  // clang-format off
+  // CHECK: call void @dx.op.outerProductAccumulate.v128f16.v64f16(i32 307, <128 x half> %{{.+}}, <64 x half> %{{.+}}, %dx.types.Handle %{{.+}}, i32 0, i32 8, i32 3, i32 0)
+  // __builtin_OuterProductAccumulate(Input1, Input2, RWBuf, 0, DATA_TYPE_FLOAT16, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL, 0);
+  // clang-format on
+
+  OuterProductAccumulate(Input1, Input2, matrix);  
+}
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vector-matrix-proposal.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vector-matrix-proposal.hlsl
new file mode 100644
index 0000000000..21119cf90c
--- /dev/null
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vector-matrix-proposal.hlsl
@@ -0,0 +1,143 @@
+// RUN: %dxc -T lib_6_9 -enable-16bit-types %s
+
+#include "linalg.h"
+
+namespace code_example {
+ByteAddressBuffer Model;
+
+vector<float, 3> ApplyNeuralMaterial(vector<half, 8> InputVector) {
+  using namespace dx::linalg;
+
+  MatrixRef<DATA_TYPE_FLOAT8_E4M3, 32, 8, MATRIX_LAYOUT_MUL_OPTIMAL> Matrix0 = {
+      Model, 0, 0};
+
+  VectorRef<DATA_TYPE_FLOAT16> BiasVector0 = {Model, 1024};
+
+  MatrixRef<DATA_TYPE_FLOAT8_E4M3, 32, 32, MATRIX_LAYOUT_MUL_OPTIMAL> Matrix1 =
+      {Model, 2048, 0};
+
+  VectorRef<DATA_TYPE_FLOAT16> BiasVector1 = {Model, 3072};
+
+  MatrixRef<DATA_TYPE_FLOAT8_E4M3, 3, 32, MATRIX_LAYOUT_MUL_OPTIMAL> Matrix2 = {
+      Model, 4096, 0};
+
+  VectorRef<DATA_TYPE_FLOAT16> BiasVector2 = {Model, 5120};
+
+  vector<half, 32> Layer0 = MulAdd<half>(
+      Matrix0, MakeInterpretedVector<DATA_TYPE_FLOAT8_E4M3>(InputVector),
+      BiasVector0);
+  Layer0 = max(Layer0, 0);
+
+  vector<half, 32> Layer1 = MulAdd<half>(
+      Matrix1, MakeInterpretedVector<DATA_TYPE_FLOAT8_E4M3>(Layer0),
+      BiasVector1);
+  Layer1 = max(Layer1, 0);
+
+  vector<float, 3> Output = MulAdd<float>(
+      Matrix2, MakeInterpretedVector<DATA_TYPE_FLOAT8_E4M3>(Layer1),
+      BiasVector2);
+  Output = exp(Output);
+
+  return Output;
+}
+} // namespace code_example
+
+namespace matrixref_example {
+ByteAddressBuffer ROBuffer;
+RWByteAddressBuffer RWBuffer;
+
+void Example() {
+  using namespace dx::linalg;
+
+  MatrixRef<DATA_TYPE_FLOAT16, 4, 4, MATRIX_LAYOUT_MUL_OPTIMAL, true> MatrixA =
+      {ROBuffer, /*offset=*/128, /*stride=*/0};
+
+  MatrixRef<DATA_TYPE_FLOAT16, 4, 4, MATRIX_LAYOUT_ROW_MAJOR, true> MatrixB = {
+      ROBuffer, /*offset=*/128, /*stride=*/16};
+
+  RWMatrixRef<DATA_TYPE_FLOAT16, 128, 256, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL>
+      MatrixC = {RWBuffer, /*offset=*/64, /*stride=*/0};
+}
+} // namespace matrixref_example
+
+namespace vectorref_example {
+ByteAddressBuffer ROBuffer;
+RWByteAddressBuffer RWBuffer;
+
+void Example() {
+  using namespace dx::linalg;
+
+  VectorRef<DATA_TYPE_FLOAT16> VectorA = {ROBuffer, /*offset=*/128};
+  VectorRef<DATA_TYPE_FLOAT32> VectorB = {ROBuffer, /*offset=*/128};
+  RWVectorRef<DATA_TYPE_SINT16> VectorC = {RWBuffer, /*offset=*/64};
+}
+} // namespace vectorref_example
+
+namespace vector_example {
+ByteAddressBuffer Buffer;
+void Example() {
+  using namespace dx::linalg;
+
+  MatrixRef<DATA_TYPE_FLOAT16, 128, 128, MATRIX_LAYOUT_MUL_OPTIMAL, true>
+      Matrix = {Buffer, 0, 0};
+
+  vector<float, 128> V = 0;
+  vector<float, 128> Result =
+      Mul<float>(Matrix, MakeInterpretedVector<DATA_TYPE_FLOAT8_E4M3>(V));
+
+  // alternative:
+  InterpretedVector<float, 128, DATA_TYPE_FLOAT8_E4M3> IV = {V};
+  vector<float, 128> Result2 = Mul<float>(Matrix, IV);
+}
+} // namespace vector_example
+
+namespace mul_example {
+ByteAddressBuffer Buffer;
+float4 Example(float4 Input) {
+  using namespace dx::linalg;
+
+  MatrixRef<DATA_TYPE_FLOAT16, 4, 4, MATRIX_LAYOUT_MUL_OPTIMAL, true> Matrix = {
+      Buffer, 0, 0};
+
+  return Mul<float>(Matrix, MakeInterpretedVector<DATA_TYPE_FLOAT16>(Input));
+}
+} // namespace mul_example
+
+namespace muladd_example {
+ByteAddressBuffer Buffer;
+
+void Example() {
+  using namespace dx::linalg;
+
+  MatrixRef<DATA_TYPE_FLOAT8_E4M3, 32, 8, MATRIX_LAYOUT_MUL_OPTIMAL> Matrix = {
+      Buffer, 0, 0};
+
+  VectorRef<DATA_TYPE_FLOAT16> BiasVector = {Buffer, 1024};
+
+  vector<float, 8> V = 0;
+  vector<float, 32> Result = MulAdd<float>(
+      Matrix, MakeInterpretedVector<DATA_TYPE_FLOAT8_E4M3>(V), BiasVector);
+}
+} // namespace muladd_example
+
+namespace outerproductaccumulate_example {
+RWByteAddressBuffer RWBuf;
+
+void Example(vector<half, 128> Input1, vector<half, 256> Input2) {
+  using namespace dx::linalg;
+
+  RWMatrixRef<DATA_TYPE_FLOAT16, 128, 256, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL>
+      Matrix = {RWBuf, 0, 0};
+
+  OuterProductAccumulate(Input1, Input2, Matrix);
+}
+} // namespace outerproductaccumulate_example
+
+namespace vector_accumulate {
+RWByteAddressBuffer RWBuf;
+
+void Test(vector<half, 128> Input) {
+  using namespace dx::linalg;
+  VectorAccumulate(Input, RWBuf, 0);
+}
+} // namespace vector_accumulate
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate.hlsl
new file mode 100644
index 0000000000..e6fcca1433
--- /dev/null
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate.hlsl
@@ -0,0 +1,16 @@
+// RUN: %dxc -T lib_6_9 %s | FileCheck %s
+
+#include "linalg.h"
+
+RWByteAddressBuffer RWBuf;
+
+export void Test5(vector<half, 128> Input) {
+  using namespace dx::linalg;
+
+  RWBuf.Store<vector<half, 128> >(0, Input);
+
+  // clang-format off
+  // CHECK: call void @dx.op.vectorAccumulate.v128f32(i32 308, <128 x float> %{{.*}}, %dx.types.Handle %{{.*}}, i32 0)
+  VectorAccumulate(Input, RWBuf, 0);
+  // clang-format on
+}

From de0f2ceac99de41c3355d856bdc38bdf01b32c47 Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista@microsoft.com>
Date: Mon, 14 Apr 2025 11:37:37 -0700
Subject: [PATCH 02/18] add missing tests

---
 .../linalg/mat-vec-mul-transpose-errors.hlsl  | 31 ++++++++++++++++++
 .../CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl  | 15 +++++++++
 .../linalg/outerproductaccumulate-errors.hlsl | 32 +++++++++++++++++++
 .../hlsl/linalg/vectoraccumulate-errors.hlsl  | 16 ++++++++++
 4 files changed, 94 insertions(+)
 create mode 100644 tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl
 create mode 100644 tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl

diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl
new file mode 100644
index 0000000000..6cb2afa7e5
--- /dev/null
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl
@@ -0,0 +1,31 @@
+// RUN: %dxc -T lib_6_9 -enable-16bit-types %s | FileCheck %s
+
+#include "linalg.h"
+
+ByteAddressBuffer Buf;
+
+export float4 Test1(vector<float, 4> Input) {
+  using namespace dx::linalg;
+
+  MatrixRef<DATA_TYPE_FLOAT16, 4, 4, MATRIX_LAYOUT_ROW_MAJOR, true> Matrix = {
+      Buf, 0, 0};
+
+  // clang-format off
+  // CHECK: error: something about transposing not supported for rowmajor / colmajor layouts
+  return Mul<float>(    
+      Matrix, MakeInterpretedVector<DATA_TYPE_FLOAT16>(Input));
+  // clang-format on
+}
+
+export vector<float, 8> Test2(vector<uint8_t4_packed, 6> Input) {
+  using namespace dx::linalg;
+
+  MatrixRef<DATA_TYPE_UINT8, 8, 6 * 4, MATRIX_LAYOUT_COLUMN_MAJOR> Matrix = {
+      Buf, 0, 0};
+
+  // clang-format off
+  // CHECK: error: something about transposing not supported for rowmajor / colmajor layouts
+  // clang-format on
+  return Mul<float>(Matrix,
+                    MakeInterpretedVector<DATA_TYPE_UINT8_T4_PACKED>(Input));
+}
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl
index 4dbee65eaf..0475dd7377 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl
@@ -24,8 +24,23 @@ export vector<float, 8> Test2(vector<uint8_t4_packed, 6> Input) {
       Buf, 0, 0};
 
   // clang-format off
+  // note the stride argument is dropped.
   // CHECK: %{{.+}} = call <8 x float> @dx.op.matVecMul.v8f32.v6f32(i32 305, <6 x float> %{{.+}}, i1 false, i32 18, %dx.types.Handle %{{.+}}, i32 0, i32 19, i32 8, i32 24, i32 2, i1 false, i32 0, i1 false)
   // clang-format on
   return Mul<float>(Matrix,
                     MakeInterpretedVector<DATA_TYPE_UINT8_T4_PACKED>(Input));
 }
+
+// test that "stride" isn't ignored in non-optimal layouts
+export vector<float, 8> Test2(vector<uint8_t4_packed, 6> Input) {
+  using namespace dx::linalg;
+
+  MatrixRef<DATA_TYPE_UINT8, 8, 6 * 4, MATRIX_LAYOUT_ROW_MAJOR> Matrix = {
+      Buf, 0, 6 * 4 * 8};
+
+  // clang-format off
+  // CHECK: %{{.+}} = call <8 x float> @dx.op.matVecMul.v8f32.v6f32(i32 305, <6 x float> %{{.+}}, i1 false, i32 18, %dx.types.Handle %{{.+}}, i32 0, i32 19, i32 8, i32 24, i32 2, i1 false, i32 0, i1 false, i32 192)
+  // clang-format on
+  return Mul<float>(Matrix,
+                    MakeInterpretedVector<DATA_TYPE_UINT8_T4_PACKED>(Input));
+}
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl
index 2729538720..e593f86c02 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl
@@ -18,3 +18,35 @@ export void Test4(vector<half, 128> Input1, vector<half, 64> Input2) {
 
   OuterProductAccumulate(Input1, Input2, matrix);  
 }
+
+// now test for an error when element types differ
+export void Test4(vector<int, 128> Input1, vector<uint, 128> Input2) {
+  using namespace dx::linalg;
+
+  RWMatrixRef<DATA_TYPE_FLOAT16, 128, 128, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL, true>
+      matrix = {RWBuf, 0, 0};
+
+  // clang-format off
+  // CHECK: error: no matching function for call to 'OuterProductAccumulate'
+  // CHECK: note: candidate template ignored: could not match 0 against 1
+  // __builtin_OuterProductAccumulate(Input1, Input2, RWBuf, 0, DATA_TYPE_FLOAT16, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL, 0);
+  // clang-format on
+
+  OuterProductAccumulate(Input1, Input2, matrix);  
+}
+
+// now test for an error when matrix transpose parameter is false
+export void Test4(vector<int, 128> Input1, vector<uint, 128> Input2) {
+  using namespace dx::linalg;
+
+  RWMatrixRef<DATA_TYPE_FLOAT16, 128, 128, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL, false>
+      matrix = {RWBuf, 0, 0};
+
+  // clang-format off
+  // CHECK: error: no matching function for call to 'OuterProductAccumulate'
+  // CHECK: note: candidate template ignored: could not match 0 against 1
+  // __builtin_OuterProductAccumulate(Input1, Input2, RWBuf, 0, DATA_TYPE_FLOAT16, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL, 0);
+  // clang-format on
+
+  OuterProductAccumulate(Input1, Input2, matrix);  
+}
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl
new file mode 100644
index 0000000000..0ceb47df28
--- /dev/null
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl
@@ -0,0 +1,16 @@
+// RUN: %dxc -T lib_6_9 %s | FileCheck %s
+
+#include "linalg.h"
+
+RWByteAddressBuffer RWBuf;
+
+export void Test5(vector<float, 128> Input) {
+  using namespace dx::linalg;
+
+  RWBuf.Store<vector<half, 128> >(0, Input);
+
+  // clang-format off
+  // CHECK: Perhaps this is an error?
+  VectorAccumulate(Input, RWBuf, 0);
+  // clang-format on
+}

From 4858fc29b3a3d7bb723d086c1ba55e66a0b82c7a Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista@microsoft.com>
Date: Tue, 15 Apr 2025 10:25:24 -0700
Subject: [PATCH 03/18] update comment

---
 tools/clang/lib/Headers/hlsl/linalg.h | 4 ----
 1 file changed, 4 deletions(-)

diff --git a/tools/clang/lib/Headers/hlsl/linalg.h b/tools/clang/lib/Headers/hlsl/linalg.h
index f2e6e7bfc6..1e69763e4c 100644
--- a/tools/clang/lib/Headers/hlsl/linalg.h
+++ b/tools/clang/lib/Headers/hlsl/linalg.h
@@ -1,8 +1,4 @@
-
 // Header for linear algebra APIs.
-//
-// This needs to find a proper home!
-//
 
 namespace dx {
 namespace linalg {

From 4422d803ee03952d5ec591dcd8b5479c0e1d1ef4 Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista@microsoft.com>
Date: Tue, 15 Apr 2025 10:41:47 -0700
Subject: [PATCH 04/18] clarify comment

---
 .../test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl   | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl
index 0ceb47df28..77e144f5c5 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl
@@ -10,7 +10,7 @@ export void Test5(vector<float, 128> Input) {
   RWBuf.Store<vector<half, 128> >(0, Input);
 
   // clang-format off
-  // CHECK: Perhaps this is an error?
+  // CHECK: Something about an error due to illegal conversions
   VectorAccumulate(Input, RWBuf, 0);
   // clang-format on
 }

From 34302d69128d18774f29f196cf541ba18fb88624 Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista@microsoft.com>
Date: Tue, 15 Apr 2025 10:48:31 -0700
Subject: [PATCH 05/18] fix up this test

---
 .../hlsl/linalg/outerproductaccumulate-errors.hlsl       | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl
index e593f86c02..da0c0fe042 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl
@@ -4,6 +4,7 @@
 
 RWByteAddressBuffer RWBuf;
 
+// test for inputs of different size
 export void Test4(vector<half, 128> Input1, vector<half, 64> Input2) {
   using namespace dx::linalg;
 
@@ -20,7 +21,7 @@ export void Test4(vector<half, 128> Input1, vector<half, 64> Input2) {
 }
 
 // now test for an error when element types differ
-export void Test4(vector<int, 128> Input1, vector<uint, 128> Input2) {
+export void Test5(vector<int, 128> Input1, vector<uint, 128> Input2) {
   using namespace dx::linalg;
 
   RWMatrixRef<DATA_TYPE_FLOAT16, 128, 128, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL, true>
@@ -35,11 +36,11 @@ export void Test4(vector<int, 128> Input1, vector<uint, 128> Input2) {
   OuterProductAccumulate(Input1, Input2, matrix);  
 }
 
-// now test for an error when matrix transpose parameter is false
-export void Test4(vector<int, 128> Input1, vector<uint, 128> Input2) {
+// now test for an error when matrix transpose parameter is true
+export void Test4(vector<half, 64> Input1, vector<half, 64> Input2) {
   using namespace dx::linalg;
 
-  RWMatrixRef<DATA_TYPE_FLOAT16, 128, 128, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL, false>
+  RWMatrixRef<DATA_TYPE_FLOAT16, 64, 64, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL, true>
       matrix = {RWBuf, 0, 0};
 
   // clang-format off

From 6747741da3d647d23fec9540397fe372278c197a Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista@microsoft.com>
Date: Tue, 15 Apr 2025 10:50:58 -0700
Subject: [PATCH 06/18] move header to dx dir, remove all unneeded include
 directives

---
 tools/clang/lib/Headers/hlsl/{ => dx}/linalg.h                  | 0
 .../test/CodeGenDXIL/hlsl/linalg/make-interp-vec-errors.hlsl    | 2 --
 .../clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-errors.hlsl  | 2 --
 .../CodeGenDXIL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl   | 2 --
 tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl       | 2 --
 .../test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd-errors.hlsl     | 2 --
 tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd.hlsl    | 2 --
 .../CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl  | 2 --
 .../test/CodeGenDXIL/hlsl/linalg/vector-matrix-proposal.hlsl    | 2 --
 .../test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl   | 2 --
 tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate.hlsl  | 2 --
 11 files changed, 20 deletions(-)
 rename tools/clang/lib/Headers/hlsl/{ => dx}/linalg.h (100%)

diff --git a/tools/clang/lib/Headers/hlsl/linalg.h b/tools/clang/lib/Headers/hlsl/dx/linalg.h
similarity index 100%
rename from tools/clang/lib/Headers/hlsl/linalg.h
rename to tools/clang/lib/Headers/hlsl/dx/linalg.h
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/make-interp-vec-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/make-interp-vec-errors.hlsl
index a7e40cb374..ca85a9be18 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/make-interp-vec-errors.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/make-interp-vec-errors.hlsl
@@ -1,7 +1,5 @@
 // RUN: not %dxc -T lib_6_9 %s 2>&1 | FileCheck %s
 
-#include "linalg.h"
-
 ByteAddressBuffer Buf;
 
 export float4 Test1(vector<float, 4> Input) {
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-errors.hlsl
index f60eb70e0f..01eb2e66c1 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-errors.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-errors.hlsl
@@ -1,7 +1,5 @@
 // RUN: not %dxc -T lib_6_9 %s 2>&1 | FileCheck %s
 
-#include "linalg.h"
-
 ByteAddressBuffer Buf;
 
 vector<float, 128> MixUpVectorAndMatrixArguments(vector<float, 128> Input) {
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl
index 6cb2afa7e5..473dc312ae 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl
@@ -1,7 +1,5 @@
 // RUN: %dxc -T lib_6_9 -enable-16bit-types %s | FileCheck %s
 
-#include "linalg.h"
-
 ByteAddressBuffer Buf;
 
 export float4 Test1(vector<float, 4> Input) {
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl
index 0475dd7377..81060b29da 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl
@@ -1,7 +1,5 @@
 // RUN: %dxc -T lib_6_9 -enable-16bit-types %s | FileCheck %s
 
-#include "linalg.h"
-
 ByteAddressBuffer Buf;
 
 export float4 Test1(vector<float, 4> Input) {
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd-errors.hlsl
index 89a895accb..a4b4ad71a5 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd-errors.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd-errors.hlsl
@@ -1,7 +1,5 @@
 // RUN: not %dxc -T lib_6_9 %s 2>&1 | FileCheck %s
 
-#include "linalg.h"
-
 ByteAddressBuffer Buf;
 
 vector<float, 128> MixUpVectorAndMatrixArguments(vector<float, 128> Input) {
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd.hlsl
index e1efa3f929..b06848f77b 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd.hlsl
@@ -1,7 +1,5 @@
 // RUN: %dxc -T lib_6_9 %s | FileCheck %s
 
-#include "linalg.h"
-
 ByteAddressBuffer Buf;
 
 export float4 Test1(float4 input) {
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl
index da0c0fe042..5f4f58c387 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl
@@ -1,7 +1,5 @@
 // RUN: not %dxc -T lib_6_9 -enable-16bit-types %s 2>&1 | FileCheck %s
 
-#include "linalg.h"
-
 RWByteAddressBuffer RWBuf;
 
 // test for inputs of different size
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vector-matrix-proposal.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vector-matrix-proposal.hlsl
index 21119cf90c..feee88cc98 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vector-matrix-proposal.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vector-matrix-proposal.hlsl
@@ -1,7 +1,5 @@
 // RUN: %dxc -T lib_6_9 -enable-16bit-types %s
 
-#include "linalg.h"
-
 namespace code_example {
 ByteAddressBuffer Model;
 
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl
index 77e144f5c5..99d7902d29 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl
@@ -1,7 +1,5 @@
 // RUN: %dxc -T lib_6_9 %s | FileCheck %s
 
-#include "linalg.h"
-
 RWByteAddressBuffer RWBuf;
 
 export void Test5(vector<float, 128> Input) {
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate.hlsl
index e6fcca1433..26ebe43dc5 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate.hlsl
@@ -1,7 +1,5 @@
 // RUN: %dxc -T lib_6_9 %s | FileCheck %s
 
-#include "linalg.h"
-
 RWByteAddressBuffer RWBuf;
 
 export void Test5(vector<half, 128> Input) {

From ab78c627321ddd614829ef7218cacc2ec0dc0050 Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista@microsoft.com>
Date: Thu, 17 Apr 2025 14:05:36 -0700
Subject: [PATCH 07/18] ifdef header, add back includes in test files, with -i
 option, make verifier tests, and mark preview todos

---
 tools/clang/lib/Headers/hlsl/dx/linalg.h         |  6 ++++++
 .../hlsl/linalg/make-interp-vec-errors.hlsl      | 11 ++++++-----
 .../hlsl/linalg/mat-vec-mul-errors.hlsl          |  8 +++++---
 .../linalg/mat-vec-mul-transpose-errors.hlsl     |  6 +++++-
 .../CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl     |  6 ++++--
 .../hlsl/linalg/mat-vec-muladd-errors.hlsl       | 12 +++++++-----
 .../CodeGenDXIL/hlsl/linalg/mat-vec-muladd.hlsl  |  4 +++-
 .../linalg/outerproductaccumulate-errors.hlsl    | 16 +++++++++-------
 .../hlsl/linalg/outerproductaccumulate.hlsl      |  4 +++-
 .../hlsl/linalg/vector-matrix-proposal.hlsl      |  4 +++-
 .../hlsl/linalg/vectoraccumulate-errors.hlsl     |  5 ++++-
 .../hlsl/linalg/vectoraccumulate.hlsl            |  4 +++-
 12 files changed, 58 insertions(+), 28 deletions(-)

diff --git a/tools/clang/lib/Headers/hlsl/dx/linalg.h b/tools/clang/lib/Headers/hlsl/dx/linalg.h
index 1e69763e4c..63a4a63963 100644
--- a/tools/clang/lib/Headers/hlsl/dx/linalg.h
+++ b/tools/clang/lib/Headers/hlsl/dx/linalg.h
@@ -1,5 +1,9 @@
 // Header for linear algebra APIs.
 
+#if ((__SHADER_TARGET_MAJOR > 6) || \
+     (__SHADER_TARGET_MAJOR == 6 && __SHADER_TARGET_MINOR >= 9)) && \
+    (__HLSL_VERSION >= 2021)
+
 namespace dx {
 namespace linalg {
 
@@ -170,3 +174,5 @@ void VectorAccumulate(vector<ElTy, ElCount> InputVector,
 
 } // namespace linalg
 } // namespace dx
+
+#endif
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/make-interp-vec-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/make-interp-vec-errors.hlsl
index ca85a9be18..47110c099a 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/make-interp-vec-errors.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/make-interp-vec-errors.hlsl
@@ -1,5 +1,6 @@
-// RUN: not %dxc -T lib_6_9 %s 2>&1 | FileCheck %s
+// RUN: %dxc -I %hlsl_headers -T lib_6_9 %s -verify
 
+#include <dx/linalg.h>
 ByteAddressBuffer Buf;
 
 export float4 Test1(vector<float, 4> Input) {
@@ -9,8 +10,8 @@ export float4 Test1(vector<float, 4> Input) {
       Buf, 0, 0};
 
   // clang-format off
-  // CHECK: error: no matching function for call to 'MakeInterpretedVector'
-  // CHECK: note: candidate template ignored: invalid explicitly-specified argument for template parameter 'DT'
+  // expected-error@+3{{no matching function for call to 'MakeInterpretedVector'}}
+  // expected-note@dx/linalg.h:93{{candidate template ignored: invalid explicitly-specified argument for template parameter 'DT'}}
   return Mul<float>(    
       Matrix, MakeInterpretedVector<2>(Input));
   // clang-format on
@@ -27,8 +28,8 @@ export float4 Test2(vector<float, 4> Input) {
       Buf, 0, 0};
 
   // clang-format off
-  // CHECK: error: no matching function for call to 'MakeInterpretedVector'
-  // CHECK: note: candidate template ignored: invalid explicitly-specified argument for template parameter 'DT'
+  // expected-error@+3{{no matching function for call to 'MakeInterpretedVector'}}
+  // expected-note@dx/linalg.h:93{{candidate template ignored: invalid explicitly-specified argument for template parameter 'DT'}}
   return Mul<float>(    
       Matrix, MakeInterpretedVector<DATA_TYPE_InvalidType>(Input));
   // clang-format on
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-errors.hlsl
index 01eb2e66c1..a4ad213968 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-errors.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-errors.hlsl
@@ -1,4 +1,6 @@
-// RUN: not %dxc -T lib_6_9 %s 2>&1 | FileCheck %s
+// RUN: %dxc -I %hlsl_headers -T lib_6_9 %s -verify
+
+#include <dx/linalg.h>
 
 ByteAddressBuffer Buf;
 
@@ -9,8 +11,8 @@ vector<float, 128> MixUpVectorAndMatrixArguments(vector<float, 128> Input) {
       Buf, 0, 0};
 
   // clang-format off
-  // CHECK: error: no matching function for call to 'Mul'
-  // CHECK: note: candidate template ignored: could not match 'MatrixRefImpl' against 'InterpretedVector'
+  // expected-error@+3{{no matching function for call to 'Mul'}}
+  // expected-note@dx/linalg.h:107{{candidate template ignored: could not match 'MatrixRefImpl' against 'InterpretedVector'}}
   // clang-format on
   return Mul<float>(MakeInterpretedVector<DATA_TYPE_FLOAT16>(Input), Matrix);
 }
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl
index 473dc312ae..2e17cb3aef 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl
@@ -1,4 +1,6 @@
-// RUN: %dxc -T lib_6_9 -enable-16bit-types %s | FileCheck %s
+// RUN: %dxc -I %hlsl_headers -T lib_6_9 -enable-16bit-types %s -verify
+
+#include <dx/linalg.h>
 
 ByteAddressBuffer Buf;
 
@@ -9,6 +11,7 @@ export float4 Test1(vector<float, 4> Input) {
       Buf, 0, 0};
 
   // clang-format off
+  // PREVIEW CHECK TODO:
   // CHECK: error: something about transposing not supported for rowmajor / colmajor layouts
   return Mul<float>(    
       Matrix, MakeInterpretedVector<DATA_TYPE_FLOAT16>(Input));
@@ -22,6 +25,7 @@ export vector<float, 8> Test2(vector<uint8_t4_packed, 6> Input) {
       Buf, 0, 0};
 
   // clang-format off
+  // PREVIEW CHECK TODO:
   // CHECK: error: something about transposing not supported for rowmajor / colmajor layouts
   // clang-format on
   return Mul<float>(Matrix,
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl
index 81060b29da..3c2cf9b0b6 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl
@@ -1,4 +1,6 @@
-// RUN: %dxc -T lib_6_9 -enable-16bit-types %s | FileCheck %s
+// RUN: %dxc -I %hlsl_headers -T lib_6_9 -enable-16bit-types %s | FileCheck %s
+
+#include <dx/linalg.h>
 
 ByteAddressBuffer Buf;
 
@@ -30,7 +32,7 @@ export vector<float, 8> Test2(vector<uint8_t4_packed, 6> Input) {
 }
 
 // test that "stride" isn't ignored in non-optimal layouts
-export vector<float, 8> Test2(vector<uint8_t4_packed, 6> Input) {
+export vector<float, 8> Test3(vector<uint8_t4_packed, 6> Input) {
   using namespace dx::linalg;
 
   MatrixRef<DATA_TYPE_UINT8, 8, 6 * 4, MATRIX_LAYOUT_ROW_MAJOR> Matrix = {
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd-errors.hlsl
index a4b4ad71a5..cfe66f9c4b 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd-errors.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd-errors.hlsl
@@ -1,16 +1,18 @@
-// RUN: not %dxc -T lib_6_9 %s 2>&1 | FileCheck %s
+// RUN: %dxc -I %hlsl_headers -T lib_6_9 %s -verify
+
+#include <dx/linalg.h>
 
 ByteAddressBuffer Buf;
 
 vector<float, 128> MixUpVectorAndMatrixArguments(vector<float, 128> Input) {
   using namespace dx::linalg;
 
-  MatrixRef<1, 128, 128, MATRIX_LAYOUT_MUL_OPTIMAL> Matrix = {
+  MatrixRef<DATA_TYPE_SINT16, 128, 128, MATRIX_LAYOUT_MUL_OPTIMAL> Matrix = {
       Buf, 0, 0};
 
   // clang-format off
-  // CHECK: error: no matching function for call to 'MulAdd'
-  // CHECK: note: candidate template ignored: could not match 'MatrixRefImpl' against 'InterpretedVector'
+  // expected-error@+3{{no matching function for call to 'MulAdd'}}
+  // expected-note@dx/linalg.h:133{{candidate template ignored: could not match 'MatrixRefImpl' against 'InterpretedVector'}}
   // clang-format on
-  return MulAdd<float>(MakeInterpretedVector<1>(Input), Matrix, MakeInterpretedVector<1>(Input));
+  return MulAdd<float>(MakeInterpretedVector<DATA_TYPE_SINT16>(Input), Matrix, MakeInterpretedVector<DATA_TYPE_SINT16>(Input));
 }
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd.hlsl
index b06848f77b..32032fe71b 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd.hlsl
@@ -1,4 +1,6 @@
-// RUN: %dxc -T lib_6_9 %s | FileCheck %s
+// RUN: %dxc -I %hlsl_headers -T lib_6_9 %s | FileCheck %s
+
+#include <dx/linalg.h>
 
 ByteAddressBuffer Buf;
 
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl
index 5f4f58c387..b08ee0b6bc 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl
@@ -1,4 +1,6 @@
-// RUN: not %dxc -T lib_6_9 -enable-16bit-types %s 2>&1 | FileCheck %s
+// RUN: %dxc -I %hlsl_headers -T lib_6_9 -enable-16bit-types %s -verify
+
+#include <dx/linalg.h>
 
 RWByteAddressBuffer RWBuf;
 
@@ -10,8 +12,8 @@ export void Test4(vector<half, 128> Input1, vector<half, 64> Input2) {
       matrix = {RWBuf, 0, 0};
 
   // clang-format off
-  // CHECK: error: no matching function for call to 'OuterProductAccumulate'
-  // CHECK: note: candidate template ignored: could not match 0 against 1
+  // expected-error@+5{{no matching function for call to 'OuterProductAccumulate'}}
+  // expected-note@dx/linalg.h:157{{candidate template ignored: could not match 0 against 1}}
   // __builtin_OuterProductAccumulate(Input1, Input2, RWBuf, 0, DATA_TYPE_FLOAT16, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL, 0);
   // clang-format on
 
@@ -26,8 +28,8 @@ export void Test5(vector<int, 128> Input1, vector<uint, 128> Input2) {
       matrix = {RWBuf, 0, 0};
 
   // clang-format off
-  // CHECK: error: no matching function for call to 'OuterProductAccumulate'
-  // CHECK: note: candidate template ignored: could not match 0 against 1
+  // expected-error@+5{{no matching function for call to 'OuterProductAccumulate'}}
+  // expected-note@dx/linalg.h:157{{candidate template ignored: could not match 0 against 1}}
   // __builtin_OuterProductAccumulate(Input1, Input2, RWBuf, 0, DATA_TYPE_FLOAT16, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL, 0);
   // clang-format on
 
@@ -42,8 +44,8 @@ export void Test4(vector<half, 64> Input1, vector<half, 64> Input2) {
       matrix = {RWBuf, 0, 0};
 
   // clang-format off
-  // CHECK: error: no matching function for call to 'OuterProductAccumulate'
-  // CHECK: note: candidate template ignored: could not match 0 against 1
+  // expected-error@+5{{no matching function for call to 'OuterProductAccumulate'}}
+  // expected-note@dx/linalg.h:157{{candidate template ignored: deduced conflicting types for parameter 'ElTy' ('int' vs. 'unsigned int')}}
   // __builtin_OuterProductAccumulate(Input1, Input2, RWBuf, 0, DATA_TYPE_FLOAT16, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL, 0);
   // clang-format on
 
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate.hlsl
index c9b0d7410e..cc7f8778fb 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate.hlsl
@@ -1,4 +1,6 @@
-// RUN: %dxc -T lib_6_9 -enable-16bit-types %s | FileCheck %s
+// RUN: %dxc -I %hlsl_headers -T lib_6_9 -enable-16bit-types %s | FileCheck %s
+
+#include <dx/linalg.h>
 
 RWByteAddressBuffer RWBuf;
 
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vector-matrix-proposal.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vector-matrix-proposal.hlsl
index feee88cc98..90afad247f 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vector-matrix-proposal.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vector-matrix-proposal.hlsl
@@ -1,4 +1,6 @@
-// RUN: %dxc -T lib_6_9 -enable-16bit-types %s
+// RUN: %dxc -I %hlsl_headers -T lib_6_9 -enable-16bit-types %s
+
+#include "dx/linalg.h"
 
 namespace code_example {
 ByteAddressBuffer Model;
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl
index 99d7902d29..3de9615b4d 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl
@@ -1,4 +1,6 @@
-// RUN: %dxc -T lib_6_9 %s | FileCheck %s
+// RUN: %dxc -I %hlsl_headers -T lib_6_9 %s | FileCheck %s
+
+#include "dx/linalg.h"
 
 RWByteAddressBuffer RWBuf;
 
@@ -8,6 +10,7 @@ export void Test5(vector<float, 128> Input) {
   RWBuf.Store<vector<half, 128> >(0, Input);
 
   // clang-format off
+  // PREVIEW CHECK TODO:
   // CHECK: Something about an error due to illegal conversions
   VectorAccumulate(Input, RWBuf, 0);
   // clang-format on
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate.hlsl
index 26ebe43dc5..46523a9883 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate.hlsl
@@ -1,4 +1,6 @@
-// RUN: %dxc -T lib_6_9 %s | FileCheck %s
+// RUN: %dxc -I %hlsl_headers -T lib_6_9 %s | FileCheck %s
+
+#include <dx/linalg.h>
 
 RWByteAddressBuffer RWBuf;
 

From 9e79c492889c6b0868e154540b67216c7d14eb46 Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista@microsoft.com>
Date: Thu, 17 Apr 2025 14:08:13 -0700
Subject: [PATCH 08/18] add back xfails

---
 .../CodeGenDXIL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl    | 1 +
 .../test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl    | 1 +
 2 files changed, 2 insertions(+)

diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl
index 2e17cb3aef..8218c64431 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl
@@ -1,3 +1,4 @@
+// XFAIL: *
 // RUN: %dxc -I %hlsl_headers -T lib_6_9 -enable-16bit-types %s -verify
 
 #include <dx/linalg.h>
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl
index 3de9615b4d..337c3db237 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl
@@ -1,3 +1,4 @@
+// XFAIL: *
 // RUN: %dxc -I %hlsl_headers -T lib_6_9 %s | FileCheck %s
 
 #include "dx/linalg.h"

From a2b371d074cfb4fdc3ff4d03c7d3e3b0d0487c53 Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista@microsoft.com>
Date: Thu, 17 Apr 2025 14:11:17 -0700
Subject: [PATCH 09/18] use angle brackets

---
 .../test/CodeGenDXIL/hlsl/linalg/vector-matrix-proposal.hlsl    | 2 +-
 .../test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl   | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vector-matrix-proposal.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vector-matrix-proposal.hlsl
index 90afad247f..94c414c8ba 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vector-matrix-proposal.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vector-matrix-proposal.hlsl
@@ -1,6 +1,6 @@
 // RUN: %dxc -I %hlsl_headers -T lib_6_9 -enable-16bit-types %s
 
-#include "dx/linalg.h"
+#include <dx/linalg.h>
 
 namespace code_example {
 ByteAddressBuffer Model;
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl
index 337c3db237..6c12146773 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl
@@ -1,7 +1,7 @@
 // XFAIL: *
 // RUN: %dxc -I %hlsl_headers -T lib_6_9 %s | FileCheck %s
 
-#include "dx/linalg.h"
+#include <dx/linalg.h>
 
 RWByteAddressBuffer RWBuf;
 

From a791ede156023f98532aaa92127ad1149382a164 Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista@microsoft.com>
Date: Thu, 17 Apr 2025 16:52:30 -0700
Subject: [PATCH 10/18] move sema diag verifier tests over to semahlsl dir

---
 .../hlsl/linalg/make-interp-vec-errors.hlsl                       | 0
 .../{CodeGenDXIL => SemaHLSL}/hlsl/linalg/mat-vec-mul-errors.hlsl | 0
 .../hlsl/linalg/mat-vec-mul-transpose-errors.hlsl                 | 0
 .../hlsl/linalg/mat-vec-muladd-errors.hlsl                        | 0
 .../hlsl/linalg/outerproductaccumulate-errors.hlsl                | 0
 5 files changed, 0 insertions(+), 0 deletions(-)
 rename tools/clang/test/{CodeGenDXIL => SemaHLSL}/hlsl/linalg/make-interp-vec-errors.hlsl (100%)
 rename tools/clang/test/{CodeGenDXIL => SemaHLSL}/hlsl/linalg/mat-vec-mul-errors.hlsl (100%)
 rename tools/clang/test/{CodeGenDXIL => SemaHLSL}/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl (100%)
 rename tools/clang/test/{CodeGenDXIL => SemaHLSL}/hlsl/linalg/mat-vec-muladd-errors.hlsl (100%)
 rename tools/clang/test/{CodeGenDXIL => SemaHLSL}/hlsl/linalg/outerproductaccumulate-errors.hlsl (100%)

diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/make-interp-vec-errors.hlsl b/tools/clang/test/SemaHLSL/hlsl/linalg/make-interp-vec-errors.hlsl
similarity index 100%
rename from tools/clang/test/CodeGenDXIL/hlsl/linalg/make-interp-vec-errors.hlsl
rename to tools/clang/test/SemaHLSL/hlsl/linalg/make-interp-vec-errors.hlsl
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-errors.hlsl b/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-mul-errors.hlsl
similarity index 100%
rename from tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-errors.hlsl
rename to tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-mul-errors.hlsl
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl b/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl
similarity index 100%
rename from tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl
rename to tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd-errors.hlsl b/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-muladd-errors.hlsl
similarity index 100%
rename from tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd-errors.hlsl
rename to tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-muladd-errors.hlsl
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl b/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-errors.hlsl
similarity index 100%
rename from tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate-errors.hlsl
rename to tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-errors.hlsl

From 8f950ced8a146d95b5311dbc612c3cdcbc4d9118 Mon Sep 17 00:00:00 2001
From: "github-actions[bot]" <github-actions[bot]@users.noreply.github.com>
Date: Thu, 17 Apr 2025 21:25:10 +0000
Subject: [PATCH 11/18] chore: autopublish 2025-04-17T21:25:10Z

---
 tools/clang/lib/Headers/hlsl/dx/linalg.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/tools/clang/lib/Headers/hlsl/dx/linalg.h b/tools/clang/lib/Headers/hlsl/dx/linalg.h
index 63a4a63963..8241bc2d72 100644
--- a/tools/clang/lib/Headers/hlsl/dx/linalg.h
+++ b/tools/clang/lib/Headers/hlsl/dx/linalg.h
@@ -1,7 +1,7 @@
 // Header for linear algebra APIs.
 
-#if ((__SHADER_TARGET_MAJOR > 6) || \
-     (__SHADER_TARGET_MAJOR == 6 && __SHADER_TARGET_MINOR >= 9)) && \
+#if ((__SHADER_TARGET_MAJOR > 6) ||                                            \
+     (__SHADER_TARGET_MAJOR == 6 && __SHADER_TARGET_MINOR >= 9)) &&            \
     (__HLSL_VERSION >= 2021)
 
 namespace dx {

From d987a590ac1454cf06461e00db9ceac9a4d6a4bb Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista@microsoft.com>
Date: Fri, 18 Apr 2025 15:25:28 -0700
Subject: [PATCH 12/18] update check line

---
 tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl
index 3c2cf9b0b6..49541d9306 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl
@@ -39,7 +39,7 @@ export vector<float, 8> Test3(vector<uint8_t4_packed, 6> Input) {
       Buf, 0, 6 * 4 * 8};
 
   // clang-format off
-  // CHECK: %{{.+}} = call <8 x float> @dx.op.matVecMul.v8f32.v6f32(i32 305, <6 x float> %{{.+}}, i1 false, i32 18, %dx.types.Handle %{{.+}}, i32 0, i32 19, i32 8, i32 24, i32 2, i1 false, i32 0, i1 false, i32 192)
+  // CHECK: %{{.+}} = call <8 x float> @dx.op.matVecMul.v8f32.v6f32(i32 305, <6 x float> %{{.+}}, i1 false, i32 18, %dx.types.Handle %{{.+}}, i32 0, i32 19, i32 8, i32 24, i32 0, i1 false, i32 192, i1 false)
   // clang-format on
   return Mul<float>(Matrix,
                     MakeInterpretedVector<DATA_TYPE_UINT8_T4_PACKED>(Input));

From 0cf39438e9c1474ed42ed6542f725d82647a1785 Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista@microsoft.com>
Date: Mon, 21 Apr 2025 14:01:18 -0700
Subject: [PATCH 13/18] address Damyan

---
 .../test/CodeGenDXIL/hlsl/linalg/.clang-format    |  2 --
 .../test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl |  6 ------
 .../CodeGenDXIL/hlsl/linalg/mat-vec-muladd.hlsl   |  6 ------
 .../hlsl/linalg/outerproductaccumulate.hlsl       |  3 ---
 .../hlsl/linalg/vectoraccumulate-errors.hlsl      |  2 --
 .../CodeGenDXIL/hlsl/linalg/vectoraccumulate.hlsl |  2 --
 .../hlsl/linalg/make-interp-vec-errors.hlsl       |  4 ----
 .../SemaHLSL/hlsl/linalg/mat-vec-mul-errors.hlsl  |  4 +---
 .../hlsl/linalg/mat-vec-mul-transpose-errors.hlsl |  8 ++------
 .../hlsl/linalg/mat-vec-muladd-errors.hlsl        |  4 +---
 .../linalg/outerproductaccumulate-errors.hlsl     | 15 +++------------
 11 files changed, 7 insertions(+), 49 deletions(-)
 delete mode 100644 tools/clang/test/CodeGenDXIL/hlsl/linalg/.clang-format

diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/.clang-format b/tools/clang/test/CodeGenDXIL/hlsl/linalg/.clang-format
deleted file mode 100644
index 70797e559f..0000000000
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/.clang-format
+++ /dev/null
@@ -1,2 +0,0 @@
-BasedOnStyle: LLVM
-ColumnLimit: 80
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl
index 49541d9306..141801c71c 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-mul.hlsl
@@ -10,11 +10,9 @@ export float4 Test1(vector<float, 4> Input) {
   MatrixRef<DATA_TYPE_FLOAT16, 4, 4, MATRIX_LAYOUT_MUL_OPTIMAL, true> Matrix = {
       Buf, 0, 0};
 
-  // clang-format off
   // CHECK: %{{.+}} = call <4 x float> @dx.op.matVecMul.v4f32.v4f32(i32 305, <4 x float> %{{.+}}, i1 false, i32 8, %dx.types.Handle %{{.+}}, i32 0, i32 8, i32 4, i32 4, i32 2, i1 true, i32 0, i1 false)
   return Mul<float>(    
       Matrix, MakeInterpretedVector<DATA_TYPE_FLOAT16>(Input));
-  // clang-format on
 }
 
 export vector<float, 8> Test2(vector<uint8_t4_packed, 6> Input) {
@@ -23,10 +21,8 @@ export vector<float, 8> Test2(vector<uint8_t4_packed, 6> Input) {
   MatrixRef<DATA_TYPE_UINT8, 8, 6 * 4, MATRIX_LAYOUT_MUL_OPTIMAL> Matrix = {
       Buf, 0, 0};
 
-  // clang-format off
   // note the stride argument is dropped.
   // CHECK: %{{.+}} = call <8 x float> @dx.op.matVecMul.v8f32.v6f32(i32 305, <6 x float> %{{.+}}, i1 false, i32 18, %dx.types.Handle %{{.+}}, i32 0, i32 19, i32 8, i32 24, i32 2, i1 false, i32 0, i1 false)
-  // clang-format on
   return Mul<float>(Matrix,
                     MakeInterpretedVector<DATA_TYPE_UINT8_T4_PACKED>(Input));
 }
@@ -38,9 +34,7 @@ export vector<float, 8> Test3(vector<uint8_t4_packed, 6> Input) {
   MatrixRef<DATA_TYPE_UINT8, 8, 6 * 4, MATRIX_LAYOUT_ROW_MAJOR> Matrix = {
       Buf, 0, 6 * 4 * 8};
 
-  // clang-format off
   // CHECK: %{{.+}} = call <8 x float> @dx.op.matVecMul.v8f32.v6f32(i32 305, <6 x float> %{{.+}}, i1 false, i32 18, %dx.types.Handle %{{.+}}, i32 0, i32 19, i32 8, i32 24, i32 0, i1 false, i32 192, i1 false)
-  // clang-format on
   return Mul<float>(Matrix,
                     MakeInterpretedVector<DATA_TYPE_UINT8_T4_PACKED>(Input));
 }
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd.hlsl
index 32032fe71b..c19e601904 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/mat-vec-muladd.hlsl
@@ -13,12 +13,10 @@ export float4 Test1(float4 input) {
 
   InterpretedVector<float, 4, DATA_TYPE_FLOAT16> theVector = {input};
 
-  // clang-format off
   // CHECK: %{{.+}} = call <4 x float> @dx.op.matVecMulAdd.v4f32.v4f32(i32 306, <4 x float> %{{.+}}, i1 false, i32 8, %dx.types.Handle [[RES:%.+]], i32 0, i32 8, i32 4, i32 4, i32 2, i1 false, i32 0, %dx.types.Handle [[RES]], i32 256, i32 8, i1 false)
   return MulAdd<float>(
       matrix, theVector,
       biasVector);
-  // clang-format on
 }
 
 export float4 Test2(float4 input) {
@@ -30,12 +28,10 @@ export float4 Test2(float4 input) {
 
   InterpretedVector<float, 4, DATA_TYPE_FLOAT16> theVector = {input};
 
-  // clang-format off
   // CHECK: %{{.+}} = call <4 x float> @dx.op.matVecMulAdd.v4f32.v4f32(i32 306, <4 x float> %{{.+}}, i1 false, i32 8, %dx.types.Handle [[RES:%.+]], i32 0, i32 8, i32 4, i32 4, i32 2, i1 true, i32 0, %dx.types.Handle [[RES]], i32 256, i32 8, i1 false)
   return MulAdd<float>(
       matrix, theVector,
       biasVector);
-  // clang-format on
 }
 
 export float4 Test3(float4 input) {
@@ -45,12 +41,10 @@ export float4 Test3(float4 input) {
       Buf, 0, 0};
   VectorRef<DATA_TYPE_FLOAT16> biasVector = {Buf, 256};
 
-  // clang-format off
   // CHECK: %{{.+}} = call <4 x float> @dx.op.matVecMulAdd.v4f32.v4f32(i32 306, <4 x float> %{{.+}}, i1 false, i32 8, %dx.types.Handle [[RES:%.+]], i32 0, i32 8, i32 4, i32 4, i32 2, i1 true, i32 0, %dx.types.Handle [[RES]], i32 256, i32 8, i1 false)
   return MulAdd<float>(
       matrix, MakeInterpretedVector<DATA_TYPE_FLOAT16>(input),
       biasVector);
-  // clang-format on
 }
 
 namespace ProposalExample {
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate.hlsl
index cc7f8778fb..eda15c66f6 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/outerproductaccumulate.hlsl
@@ -10,10 +10,7 @@ export void Test4(vector<half, 128> Input1, vector<half, 64> Input2) {
   RWMatrixRef<DATA_TYPE_FLOAT16, 128, 64, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL>
       matrix = {RWBuf, 0, 0};
 
-  // clang-format off
   // CHECK: call void @dx.op.outerProductAccumulate.v128f16.v64f16(i32 307, <128 x half> %{{.+}}, <64 x half> %{{.+}}, %dx.types.Handle %{{.+}}, i32 0, i32 8, i32 3, i32 0)
-  // __builtin_OuterProductAccumulate(Input1, Input2, RWBuf, 0, DATA_TYPE_FLOAT16, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL, 0);
-  // clang-format on
 
   OuterProductAccumulate(Input1, Input2, matrix);  
 }
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl
index 6c12146773..4c8ae6f049 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl
@@ -10,9 +10,7 @@ export void Test5(vector<float, 128> Input) {
 
   RWBuf.Store<vector<half, 128> >(0, Input);
 
-  // clang-format off
   // PREVIEW CHECK TODO:
   // CHECK: Something about an error due to illegal conversions
   VectorAccumulate(Input, RWBuf, 0);
-  // clang-format on
 }
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate.hlsl
index 46523a9883..9157156f10 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate.hlsl
@@ -9,8 +9,6 @@ export void Test5(vector<half, 128> Input) {
 
   RWBuf.Store<vector<half, 128> >(0, Input);
 
-  // clang-format off
   // CHECK: call void @dx.op.vectorAccumulate.v128f32(i32 308, <128 x float> %{{.*}}, %dx.types.Handle %{{.*}}, i32 0)
   VectorAccumulate(Input, RWBuf, 0);
-  // clang-format on
 }
diff --git a/tools/clang/test/SemaHLSL/hlsl/linalg/make-interp-vec-errors.hlsl b/tools/clang/test/SemaHLSL/hlsl/linalg/make-interp-vec-errors.hlsl
index 47110c099a..cbcd8147e8 100644
--- a/tools/clang/test/SemaHLSL/hlsl/linalg/make-interp-vec-errors.hlsl
+++ b/tools/clang/test/SemaHLSL/hlsl/linalg/make-interp-vec-errors.hlsl
@@ -9,12 +9,10 @@ export float4 Test1(vector<float, 4> Input) {
   MatrixRef<DATA_TYPE_UINT16, 4, 4, MATRIX_LAYOUT_MUL_OPTIMAL, true> Matrix = {
       Buf, 0, 0};
 
-  // clang-format off
   // expected-error@+3{{no matching function for call to 'MakeInterpretedVector'}}
   // expected-note@dx/linalg.h:93{{candidate template ignored: invalid explicitly-specified argument for template parameter 'DT'}}
   return Mul<float>(    
       Matrix, MakeInterpretedVector<2>(Input));
-  // clang-format on
 }
 
 enum DataType {
@@ -27,11 +25,9 @@ export float4 Test2(vector<float, 4> Input) {
   MatrixRef<DATA_TYPE_UINT16, 4, 4, MATRIX_LAYOUT_MUL_OPTIMAL, true> Matrix = {
       Buf, 0, 0};
 
-  // clang-format off
   // expected-error@+3{{no matching function for call to 'MakeInterpretedVector'}}
   // expected-note@dx/linalg.h:93{{candidate template ignored: invalid explicitly-specified argument for template parameter 'DT'}}
   return Mul<float>(    
       Matrix, MakeInterpretedVector<DATA_TYPE_InvalidType>(Input));
-  // clang-format on
 }
 
diff --git a/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-mul-errors.hlsl b/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-mul-errors.hlsl
index a4ad213968..cdc3ed020f 100644
--- a/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-mul-errors.hlsl
+++ b/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-mul-errors.hlsl
@@ -10,9 +10,7 @@ vector<float, 128> MixUpVectorAndMatrixArguments(vector<float, 128> Input) {
   MatrixRef<DATA_TYPE_FLOAT16, 128, 128, MATRIX_LAYOUT_MUL_OPTIMAL> Matrix = {
       Buf, 0, 0};
 
-  // clang-format off
-  // expected-error@+3{{no matching function for call to 'Mul'}}
+  // expected-error@+2{{no matching function for call to 'Mul'}}
   // expected-note@dx/linalg.h:107{{candidate template ignored: could not match 'MatrixRefImpl' against 'InterpretedVector'}}
-  // clang-format on
   return Mul<float>(MakeInterpretedVector<DATA_TYPE_FLOAT16>(Input), Matrix);
 }
diff --git a/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl b/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl
index 8218c64431..2018acafab 100644
--- a/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl
+++ b/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-mul-transpose-errors.hlsl
@@ -11,12 +11,10 @@ export float4 Test1(vector<float, 4> Input) {
   MatrixRef<DATA_TYPE_FLOAT16, 4, 4, MATRIX_LAYOUT_ROW_MAJOR, true> Matrix = {
       Buf, 0, 0};
 
-  // clang-format off
   // PREVIEW CHECK TODO:
-  // CHECK: error: something about transposing not supported for rowmajor / colmajor layouts
+  // expected-error@+1{{something about transposing not supported for rowmajor / colmajor layouts}}
   return Mul<float>(    
       Matrix, MakeInterpretedVector<DATA_TYPE_FLOAT16>(Input));
-  // clang-format on
 }
 
 export vector<float, 8> Test2(vector<uint8_t4_packed, 6> Input) {
@@ -25,10 +23,8 @@ export vector<float, 8> Test2(vector<uint8_t4_packed, 6> Input) {
   MatrixRef<DATA_TYPE_UINT8, 8, 6 * 4, MATRIX_LAYOUT_COLUMN_MAJOR> Matrix = {
       Buf, 0, 0};
 
-  // clang-format off
   // PREVIEW CHECK TODO:
-  // CHECK: error: something about transposing not supported for rowmajor / colmajor layouts
-  // clang-format on
+  // expected-error@+1{{something about transposing not supported for rowmajor / colmajor layouts}}
   return Mul<float>(Matrix,
                     MakeInterpretedVector<DATA_TYPE_UINT8_T4_PACKED>(Input));
 }
diff --git a/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-muladd-errors.hlsl b/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-muladd-errors.hlsl
index cfe66f9c4b..03cdacb8f6 100644
--- a/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-muladd-errors.hlsl
+++ b/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-muladd-errors.hlsl
@@ -10,9 +10,7 @@ vector<float, 128> MixUpVectorAndMatrixArguments(vector<float, 128> Input) {
   MatrixRef<DATA_TYPE_SINT16, 128, 128, MATRIX_LAYOUT_MUL_OPTIMAL> Matrix = {
       Buf, 0, 0};
 
-  // clang-format off
-  // expected-error@+3{{no matching function for call to 'MulAdd'}}
+  // expected-error@+2{{no matching function for call to 'MulAdd'}}
   // expected-note@dx/linalg.h:133{{candidate template ignored: could not match 'MatrixRefImpl' against 'InterpretedVector'}}
-  // clang-format on
   return MulAdd<float>(MakeInterpretedVector<DATA_TYPE_SINT16>(Input), Matrix, MakeInterpretedVector<DATA_TYPE_SINT16>(Input));
 }
diff --git a/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-errors.hlsl b/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-errors.hlsl
index b08ee0b6bc..9c707d5c3e 100644
--- a/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-errors.hlsl
+++ b/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-errors.hlsl
@@ -11,11 +11,8 @@ export void Test4(vector<half, 128> Input1, vector<half, 64> Input2) {
   RWMatrixRef<DATA_TYPE_FLOAT16, 128, 64, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL, true>
       matrix = {RWBuf, 0, 0};
 
-  // clang-format off
-  // expected-error@+5{{no matching function for call to 'OuterProductAccumulate'}}
+  // expected-error@+3{{no matching function for call to 'OuterProductAccumulate'}}
   // expected-note@dx/linalg.h:157{{candidate template ignored: could not match 0 against 1}}
-  // __builtin_OuterProductAccumulate(Input1, Input2, RWBuf, 0, DATA_TYPE_FLOAT16, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL, 0);
-  // clang-format on
 
   OuterProductAccumulate(Input1, Input2, matrix);  
 }
@@ -27,11 +24,8 @@ export void Test5(vector<int, 128> Input1, vector<uint, 128> Input2) {
   RWMatrixRef<DATA_TYPE_FLOAT16, 128, 128, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL, true>
       matrix = {RWBuf, 0, 0};
 
-  // clang-format off
-  // expected-error@+5{{no matching function for call to 'OuterProductAccumulate'}}
+  // expected-error@+3{{no matching function for call to 'OuterProductAccumulate'}}
   // expected-note@dx/linalg.h:157{{candidate template ignored: could not match 0 against 1}}
-  // __builtin_OuterProductAccumulate(Input1, Input2, RWBuf, 0, DATA_TYPE_FLOAT16, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL, 0);
-  // clang-format on
 
   OuterProductAccumulate(Input1, Input2, matrix);  
 }
@@ -43,11 +37,8 @@ export void Test4(vector<half, 64> Input1, vector<half, 64> Input2) {
   RWMatrixRef<DATA_TYPE_FLOAT16, 64, 64, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL, true>
       matrix = {RWBuf, 0, 0};
 
-  // clang-format off
-  // expected-error@+5{{no matching function for call to 'OuterProductAccumulate'}}
+  // expected-error@+3{{no matching function for call to 'OuterProductAccumulate'}}
   // expected-note@dx/linalg.h:157{{candidate template ignored: deduced conflicting types for parameter 'ElTy' ('int' vs. 'unsigned int')}}
-  // __builtin_OuterProductAccumulate(Input1, Input2, RWBuf, 0, DATA_TYPE_FLOAT16, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL, 0);
-  // clang-format on
 
   OuterProductAccumulate(Input1, Input2, matrix);  
 }

From acec5c8d3d2b8c97392f562bb5c1ffee9826c27b Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista@microsoft.com>
Date: Mon, 21 Apr 2025 15:47:21 -0700
Subject: [PATCH 14/18] move proposal file

---
 .../hlsl/linalg => examples/LinAlg}/vector-matrix-proposal.hlsl   | 0
 1 file changed, 0 insertions(+), 0 deletions(-)
 rename {tools/clang/test/CodeGenDXIL/hlsl/linalg => examples/LinAlg}/vector-matrix-proposal.hlsl (100%)

diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vector-matrix-proposal.hlsl b/examples/LinAlg/vector-matrix-proposal.hlsl
similarity index 100%
rename from tools/clang/test/CodeGenDXIL/hlsl/linalg/vector-matrix-proposal.hlsl
rename to examples/LinAlg/vector-matrix-proposal.hlsl

From 31629b111fdef1b66601d443acfaac2480e0c445 Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista@microsoft.com>
Date: Mon, 21 Apr 2025 16:56:44 -0700
Subject: [PATCH 15/18] remove proposal file

---
 examples/LinAlg/vector-matrix-proposal.hlsl | 143 --------------------
 1 file changed, 143 deletions(-)
 delete mode 100644 examples/LinAlg/vector-matrix-proposal.hlsl

diff --git a/examples/LinAlg/vector-matrix-proposal.hlsl b/examples/LinAlg/vector-matrix-proposal.hlsl
deleted file mode 100644
index 94c414c8ba..0000000000
--- a/examples/LinAlg/vector-matrix-proposal.hlsl
+++ /dev/null
@@ -1,143 +0,0 @@
-// RUN: %dxc -I %hlsl_headers -T lib_6_9 -enable-16bit-types %s
-
-#include <dx/linalg.h>
-
-namespace code_example {
-ByteAddressBuffer Model;
-
-vector<float, 3> ApplyNeuralMaterial(vector<half, 8> InputVector) {
-  using namespace dx::linalg;
-
-  MatrixRef<DATA_TYPE_FLOAT8_E4M3, 32, 8, MATRIX_LAYOUT_MUL_OPTIMAL> Matrix0 = {
-      Model, 0, 0};
-
-  VectorRef<DATA_TYPE_FLOAT16> BiasVector0 = {Model, 1024};
-
-  MatrixRef<DATA_TYPE_FLOAT8_E4M3, 32, 32, MATRIX_LAYOUT_MUL_OPTIMAL> Matrix1 =
-      {Model, 2048, 0};
-
-  VectorRef<DATA_TYPE_FLOAT16> BiasVector1 = {Model, 3072};
-
-  MatrixRef<DATA_TYPE_FLOAT8_E4M3, 3, 32, MATRIX_LAYOUT_MUL_OPTIMAL> Matrix2 = {
-      Model, 4096, 0};
-
-  VectorRef<DATA_TYPE_FLOAT16> BiasVector2 = {Model, 5120};
-
-  vector<half, 32> Layer0 = MulAdd<half>(
-      Matrix0, MakeInterpretedVector<DATA_TYPE_FLOAT8_E4M3>(InputVector),
-      BiasVector0);
-  Layer0 = max(Layer0, 0);
-
-  vector<half, 32> Layer1 = MulAdd<half>(
-      Matrix1, MakeInterpretedVector<DATA_TYPE_FLOAT8_E4M3>(Layer0),
-      BiasVector1);
-  Layer1 = max(Layer1, 0);
-
-  vector<float, 3> Output = MulAdd<float>(
-      Matrix2, MakeInterpretedVector<DATA_TYPE_FLOAT8_E4M3>(Layer1),
-      BiasVector2);
-  Output = exp(Output);
-
-  return Output;
-}
-} // namespace code_example
-
-namespace matrixref_example {
-ByteAddressBuffer ROBuffer;
-RWByteAddressBuffer RWBuffer;
-
-void Example() {
-  using namespace dx::linalg;
-
-  MatrixRef<DATA_TYPE_FLOAT16, 4, 4, MATRIX_LAYOUT_MUL_OPTIMAL, true> MatrixA =
-      {ROBuffer, /*offset=*/128, /*stride=*/0};
-
-  MatrixRef<DATA_TYPE_FLOAT16, 4, 4, MATRIX_LAYOUT_ROW_MAJOR, true> MatrixB = {
-      ROBuffer, /*offset=*/128, /*stride=*/16};
-
-  RWMatrixRef<DATA_TYPE_FLOAT16, 128, 256, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL>
-      MatrixC = {RWBuffer, /*offset=*/64, /*stride=*/0};
-}
-} // namespace matrixref_example
-
-namespace vectorref_example {
-ByteAddressBuffer ROBuffer;
-RWByteAddressBuffer RWBuffer;
-
-void Example() {
-  using namespace dx::linalg;
-
-  VectorRef<DATA_TYPE_FLOAT16> VectorA = {ROBuffer, /*offset=*/128};
-  VectorRef<DATA_TYPE_FLOAT32> VectorB = {ROBuffer, /*offset=*/128};
-  RWVectorRef<DATA_TYPE_SINT16> VectorC = {RWBuffer, /*offset=*/64};
-}
-} // namespace vectorref_example
-
-namespace vector_example {
-ByteAddressBuffer Buffer;
-void Example() {
-  using namespace dx::linalg;
-
-  MatrixRef<DATA_TYPE_FLOAT16, 128, 128, MATRIX_LAYOUT_MUL_OPTIMAL, true>
-      Matrix = {Buffer, 0, 0};
-
-  vector<float, 128> V = 0;
-  vector<float, 128> Result =
-      Mul<float>(Matrix, MakeInterpretedVector<DATA_TYPE_FLOAT8_E4M3>(V));
-
-  // alternative:
-  InterpretedVector<float, 128, DATA_TYPE_FLOAT8_E4M3> IV = {V};
-  vector<float, 128> Result2 = Mul<float>(Matrix, IV);
-}
-} // namespace vector_example
-
-namespace mul_example {
-ByteAddressBuffer Buffer;
-float4 Example(float4 Input) {
-  using namespace dx::linalg;
-
-  MatrixRef<DATA_TYPE_FLOAT16, 4, 4, MATRIX_LAYOUT_MUL_OPTIMAL, true> Matrix = {
-      Buffer, 0, 0};
-
-  return Mul<float>(Matrix, MakeInterpretedVector<DATA_TYPE_FLOAT16>(Input));
-}
-} // namespace mul_example
-
-namespace muladd_example {
-ByteAddressBuffer Buffer;
-
-void Example() {
-  using namespace dx::linalg;
-
-  MatrixRef<DATA_TYPE_FLOAT8_E4M3, 32, 8, MATRIX_LAYOUT_MUL_OPTIMAL> Matrix = {
-      Buffer, 0, 0};
-
-  VectorRef<DATA_TYPE_FLOAT16> BiasVector = {Buffer, 1024};
-
-  vector<float, 8> V = 0;
-  vector<float, 32> Result = MulAdd<float>(
-      Matrix, MakeInterpretedVector<DATA_TYPE_FLOAT8_E4M3>(V), BiasVector);
-}
-} // namespace muladd_example
-
-namespace outerproductaccumulate_example {
-RWByteAddressBuffer RWBuf;
-
-void Example(vector<half, 128> Input1, vector<half, 256> Input2) {
-  using namespace dx::linalg;
-
-  RWMatrixRef<DATA_TYPE_FLOAT16, 128, 256, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL>
-      Matrix = {RWBuf, 0, 0};
-
-  OuterProductAccumulate(Input1, Input2, Matrix);
-}
-} // namespace outerproductaccumulate_example
-
-namespace vector_accumulate {
-RWByteAddressBuffer RWBuf;
-
-void Test(vector<half, 128> Input) {
-  using namespace dx::linalg;
-  VectorAccumulate(Input, RWBuf, 0);
-}
-} // namespace vector_accumulate

From 6dd02480e59ab2104ebd4f4621073c342f46ba20 Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista@microsoft.com>
Date: Tue, 22 Apr 2025 11:00:15 -0700
Subject: [PATCH 16/18] move sema test

---
 .../hlsl/linalg/vectoraccumulate-errors.hlsl                      | 0
 1 file changed, 0 insertions(+), 0 deletions(-)
 rename tools/clang/test/{CodeGenDXIL => SemaHLSL}/hlsl/linalg/vectoraccumulate-errors.hlsl (100%)

diff --git a/tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl b/tools/clang/test/SemaHLSL/hlsl/linalg/vectoraccumulate-errors.hlsl
similarity index 100%
rename from tools/clang/test/CodeGenDXIL/hlsl/linalg/vectoraccumulate-errors.hlsl
rename to tools/clang/test/SemaHLSL/hlsl/linalg/vectoraccumulate-errors.hlsl

From 7fcd4ff3fd392f63c6e00fe9825215202884f6e1 Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista@microsoft.com>
Date: Tue, 22 Apr 2025 13:58:13 -0700
Subject: [PATCH 17/18] add spirv check

---
 tools/clang/lib/Headers/hlsl/dx/linalg.h      |   6 +-
 .../hlsl/linalg/make-interp-vec-errors.hlsl   |   4 +-
 .../hlsl/linalg/mat-vec-mul-errors.hlsl       |   2 +-
 .../hlsl/linalg/mat-vec-muladd-errors.hlsl    |   2 +-
 .../linalg/outerproductaccumulate-errors.hlsl |   6 +-
 .../outerproductaccumulate-spirv-errors.hlsl  |  19 ++
 .../outerproductaccumulate-spirv-errors.i     | 193 ++++++++++++++++++
 7 files changed, 224 insertions(+), 8 deletions(-)
 create mode 100644 tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-spirv-errors.hlsl
 create mode 100644 tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-spirv-errors.i

diff --git a/tools/clang/lib/Headers/hlsl/dx/linalg.h b/tools/clang/lib/Headers/hlsl/dx/linalg.h
index 8241bc2d72..ed3d0a0184 100644
--- a/tools/clang/lib/Headers/hlsl/dx/linalg.h
+++ b/tools/clang/lib/Headers/hlsl/dx/linalg.h
@@ -1,5 +1,9 @@
 // Header for linear algebra APIs.
 
+#if __spirv__
+#error "Cooperative matrix/vectors not (yet) supported for SPIRV"
+#endif
+
 #if ((__SHADER_TARGET_MAJOR > 6) ||                                            \
      (__SHADER_TARGET_MAJOR == 6 && __SHADER_TARGET_MINOR >= 9)) &&            \
     (__HLSL_VERSION >= 2021)
@@ -175,4 +179,4 @@ void VectorAccumulate(vector<ElTy, ElCount> InputVector,
 } // namespace linalg
 } // namespace dx
 
-#endif
+#endif // SM 6.9 check and HV version check
diff --git a/tools/clang/test/SemaHLSL/hlsl/linalg/make-interp-vec-errors.hlsl b/tools/clang/test/SemaHLSL/hlsl/linalg/make-interp-vec-errors.hlsl
index cbcd8147e8..9f2793d417 100644
--- a/tools/clang/test/SemaHLSL/hlsl/linalg/make-interp-vec-errors.hlsl
+++ b/tools/clang/test/SemaHLSL/hlsl/linalg/make-interp-vec-errors.hlsl
@@ -10,7 +10,7 @@ export float4 Test1(vector<float, 4> Input) {
       Buf, 0, 0};
 
   // expected-error@+3{{no matching function for call to 'MakeInterpretedVector'}}
-  // expected-note@dx/linalg.h:93{{candidate template ignored: invalid explicitly-specified argument for template parameter 'DT'}}
+  // expected-note@dx/linalg.h:97{{candidate template ignored: invalid explicitly-specified argument for template parameter 'DT'}}
   return Mul<float>(    
       Matrix, MakeInterpretedVector<2>(Input));
 }
@@ -26,7 +26,7 @@ export float4 Test2(vector<float, 4> Input) {
       Buf, 0, 0};
 
   // expected-error@+3{{no matching function for call to 'MakeInterpretedVector'}}
-  // expected-note@dx/linalg.h:93{{candidate template ignored: invalid explicitly-specified argument for template parameter 'DT'}}
+  // expected-note@dx/linalg.h:97{{candidate template ignored: invalid explicitly-specified argument for template parameter 'DT'}}
   return Mul<float>(    
       Matrix, MakeInterpretedVector<DATA_TYPE_InvalidType>(Input));
 }
diff --git a/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-mul-errors.hlsl b/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-mul-errors.hlsl
index cdc3ed020f..2d5a11e83e 100644
--- a/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-mul-errors.hlsl
+++ b/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-mul-errors.hlsl
@@ -11,6 +11,6 @@ vector<float, 128> MixUpVectorAndMatrixArguments(vector<float, 128> Input) {
       Buf, 0, 0};
 
   // expected-error@+2{{no matching function for call to 'Mul'}}
-  // expected-note@dx/linalg.h:107{{candidate template ignored: could not match 'MatrixRefImpl' against 'InterpretedVector'}}
+  // expected-note@dx/linalg.h:111{{candidate template ignored: could not match 'MatrixRefImpl' against 'InterpretedVector'}}
   return Mul<float>(MakeInterpretedVector<DATA_TYPE_FLOAT16>(Input), Matrix);
 }
diff --git a/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-muladd-errors.hlsl b/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-muladd-errors.hlsl
index 03cdacb8f6..f444f81c3a 100644
--- a/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-muladd-errors.hlsl
+++ b/tools/clang/test/SemaHLSL/hlsl/linalg/mat-vec-muladd-errors.hlsl
@@ -11,6 +11,6 @@ vector<float, 128> MixUpVectorAndMatrixArguments(vector<float, 128> Input) {
       Buf, 0, 0};
 
   // expected-error@+2{{no matching function for call to 'MulAdd'}}
-  // expected-note@dx/linalg.h:133{{candidate template ignored: could not match 'MatrixRefImpl' against 'InterpretedVector'}}
+  // expected-note@dx/linalg.h:137{{candidate template ignored: could not match 'MatrixRefImpl' against 'InterpretedVector'}}
   return MulAdd<float>(MakeInterpretedVector<DATA_TYPE_SINT16>(Input), Matrix, MakeInterpretedVector<DATA_TYPE_SINT16>(Input));
 }
diff --git a/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-errors.hlsl b/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-errors.hlsl
index 9c707d5c3e..6f503b367b 100644
--- a/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-errors.hlsl
+++ b/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-errors.hlsl
@@ -12,7 +12,7 @@ export void Test4(vector<half, 128> Input1, vector<half, 64> Input2) {
       matrix = {RWBuf, 0, 0};
 
   // expected-error@+3{{no matching function for call to 'OuterProductAccumulate'}}
-  // expected-note@dx/linalg.h:157{{candidate template ignored: could not match 0 against 1}}
+  // expected-note@dx/linalg.h:161{{candidate template ignored: could not match 0 against 1}}
 
   OuterProductAccumulate(Input1, Input2, matrix);  
 }
@@ -25,7 +25,7 @@ export void Test5(vector<int, 128> Input1, vector<uint, 128> Input2) {
       matrix = {RWBuf, 0, 0};
 
   // expected-error@+3{{no matching function for call to 'OuterProductAccumulate'}}
-  // expected-note@dx/linalg.h:157{{candidate template ignored: could not match 0 against 1}}
+  // expected-note@dx/linalg.h:161{{candidate template ignored: could not match 0 against 1}}
 
   OuterProductAccumulate(Input1, Input2, matrix);  
 }
@@ -38,7 +38,7 @@ export void Test4(vector<half, 64> Input1, vector<half, 64> Input2) {
       matrix = {RWBuf, 0, 0};
 
   // expected-error@+3{{no matching function for call to 'OuterProductAccumulate'}}
-  // expected-note@dx/linalg.h:157{{candidate template ignored: deduced conflicting types for parameter 'ElTy' ('int' vs. 'unsigned int')}}
+  // expected-note@dx/linalg.h:161{{candidate template ignored: deduced conflicting types for parameter 'ElTy' ('int' vs. 'unsigned int')}}
 
   OuterProductAccumulate(Input1, Input2, matrix);  
 }
diff --git a/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-spirv-errors.hlsl b/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-spirv-errors.hlsl
new file mode 100644
index 0000000000..3ec5998f75
--- /dev/null
+++ b/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-spirv-errors.hlsl
@@ -0,0 +1,19 @@
+// RUN: %dxc -I %hlsl_headers -T lib_6_9 -enable-16bit-types -spirv %s -verify
+
+// Tests that the header file cannot be included for spirv compilations
+// This is a copy of \tools\clang\test\CodeGenDXIL\hlsl\linalg\outerproductaccumulate.hlsl
+// except that spirv is targeted
+
+// expected-error@dx/linalg.h:4{{Cooperative matrix/vectors not (yet) supported for SPIRV}}
+#include <dx/linalg.h>
+
+RWByteAddressBuffer RWBuf;
+
+export void Test4(vector<half, 128> Input1, vector<half, 64> Input2) {
+  using namespace dx::linalg;
+
+  RWMatrixRef<DATA_TYPE_FLOAT16, 128, 64, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL>
+      matrix = {RWBuf, 0, 0};
+
+  OuterProductAccumulate(Input1, Input2, matrix);  
+}
diff --git a/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-spirv-errors.i b/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-spirv-errors.i
new file mode 100644
index 0000000000..44ac09f9e8
--- /dev/null
+++ b/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-spirv-errors.i
@@ -0,0 +1,193 @@
+#line 1 "D:\\DXC\\tools\\clang\\test\\SemaHLSL\\hlsl\\linalg\\outerproductaccumulate-spirv-errors.hlsl"
+
+
+
+
+
+
+#line 1 "D:/DXC/tools/clang/lib/Headers/hlsl\\dx/linalg.h"
+#line 11 "D:/DXC/tools/clang/lib/Headers/hlsl\\dx/linalg.h"
+namespace dx {
+namespace linalg {
+
+
+
+
+
+enum DataType {
+  DATA_TYPE_SINT16 = 2,
+  DATA_TYPE_UINT16 = 3,
+  DATA_TYPE_SINT32 = 4,
+  DATA_TYPE_UINT32 = 5,
+  DATA_TYPE_FLOAT16 = 8,
+  DATA_TYPE_FLOAT32 = 9,
+  DATA_TYPE_SINT8_T4_PACKED = 17,
+  DATA_TYPE_UINT8_T4_PACKED = 18,
+  DATA_TYPE_UINT8 = 19,
+  DATA_TYPE_SINT8 = 20,
+  DATA_TYPE_FLOAT8_E4M3 = 21,
+
+  DATA_TYPE_FLOAT8_E5M2 = 22,
+
+};
+
+enum MatrixLayout {
+  MATRIX_LAYOUT_ROW_MAJOR = 0,
+  MATRIX_LAYOUT_COLUMN_MAJOR = 1,
+  MATRIX_LAYOUT_MUL_OPTIMAL = 2,
+  MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL = 3
+};
+
+
+
+
+namespace details {
+template <typename T> bool IsUnsigned() { return false; }
+
+
+template <> bool IsUnsigned<uint16_t>() { return true; }
+
+
+template <> bool IsUnsigned<uint32_t>() { return true; }
+template <> bool IsUnsigned<uint64_t>() { return true; }
+}
+
+
+
+
+
+template <typename BufferTy, DataType DT, uint M, uint K, MatrixLayout ML,
+          bool Transpose>
+struct MatrixRefImpl {
+  BufferTy Buffer;
+  uint StartOffset;
+  uint Stride;
+};
+
+template <DataType DT, uint M, uint K, MatrixLayout ML, bool Transpose = false>
+using MatrixRef = MatrixRefImpl<ByteAddressBuffer, DT, M, K, ML, Transpose>;
+
+template <DataType DT, uint M, uint K, MatrixLayout ML, bool Transpose = false>
+using RWMatrixRef = MatrixRefImpl<RWByteAddressBuffer, DT, M, K, ML, Transpose>;
+
+
+
+
+
+template <typename BufferTy, DataType DT> struct VectorRefImpl {
+  BufferTy Buffer;
+  uint StartOffset;
+};
+
+template <DataType DT> using VectorRef = VectorRefImpl<ByteAddressBuffer, DT>;
+
+template <DataType DT>
+using RWVectorRef = VectorRefImpl<RWByteAddressBuffer, DT>;
+
+
+
+
+
+template <typename T, int N, DataType DT> struct InterpretedVector {
+  vector<T, N> Data;
+};
+
+template <DataType DT, typename T, int N>
+InterpretedVector<T, N, DT> MakeInterpretedVector(vector<T, N> Vec) {
+  InterpretedVector<T, N, DT> IV = {Vec};
+  return IV;
+}
+
+
+
+
+
+template <typename OutputElTy, typename InputElTy, int InputElCount,
+          typename MatrixBufferTy, DataType InputDT, DataType MatrixDT,
+          uint MatrixM, uint MatrixK, MatrixLayout MatrixLayout,
+          bool MatrixTranspose>
+vector<OutputElTy, MatrixM>
+Mul(MatrixRefImpl<MatrixBufferTy, MatrixDT, MatrixM, MatrixK, MatrixLayout,
+                  MatrixTranspose>
+        Matrix,
+    InterpretedVector<InputElTy, InputElCount, InputDT> InputVector) {
+
+  vector<OutputElTy, MatrixM> OutputVector;
+
+  __builtin_MatVecMul(
+              OutputVector, details::IsUnsigned<OutputElTy>(), InputVector.Data,
+      details::IsUnsigned<InputElTy>(), InputDT, Matrix.Buffer,
+      Matrix.StartOffset, MatrixDT, MatrixM, MatrixK, MatrixLayout,
+      MatrixTranspose, Matrix.Stride);
+
+  return OutputVector;
+}
+
+
+
+
+
+template <typename OutputElTy, typename InputElTy, int InputElCount,
+          typename MatrixBufferTy, DataType InputDT, DataType MatrixDT,
+          uint MatrixM, uint MatrixK, MatrixLayout MatrixLayout,
+          bool MatrixTranspose, typename BiasVectorBufferTy,
+          DataType BiasVectorDT>
+vector<OutputElTy, MatrixM>
+MulAdd(MatrixRefImpl<MatrixBufferTy, MatrixDT, MatrixM, MatrixK, MatrixLayout,
+                     MatrixTranspose>
+           Matrix,
+       InterpretedVector<InputElTy, InputElCount, InputDT> InputVector,
+       VectorRefImpl<BiasVectorBufferTy, BiasVectorDT> BiasVector) {
+
+  vector<OutputElTy, MatrixM> OutputVector;
+
+  __builtin_MatVecMulAdd(
+              OutputVector, details::IsUnsigned<OutputElTy>(), InputVector.Data,
+      details::IsUnsigned<InputElTy>(), InputDT, Matrix.Buffer,
+      Matrix.StartOffset, MatrixDT, MatrixM, MatrixK, MatrixLayout,
+      MatrixTranspose, Matrix.Stride, BiasVector.Buffer, BiasVector.StartOffset,
+      BiasVectorDT);
+
+  return OutputVector;
+}
+
+
+
+
+
+template <typename ElTy, int MatrixM, int MatrixN, DataType MatrixDT,
+          MatrixLayout MatrixLayout>
+void OuterProductAccumulate(
+    vector<ElTy, MatrixM> InputVector1, vector<ElTy, MatrixN> InputVector2,
+    RWMatrixRef<MatrixDT, MatrixM, MatrixN, MatrixLayout, false> Matrix) {
+  __builtin_OuterProductAccumulate(InputVector1, InputVector2, Matrix.Buffer,
+                                   Matrix.StartOffset, MatrixDT, MatrixLayout,
+                                   Matrix.Stride);
+}
+
+
+
+
+
+template <typename ElTy, int ElCount>
+void VectorAccumulate(vector<ElTy, ElCount> InputVector,
+                      RWByteAddressBuffer Buffer, uint Offset) {
+  __builtin_VectorAccumulate(InputVector, Buffer, Offset);
+}
+
+}
+}
+#line 7 "D:\\DXC\\tools\\clang\\test\\SemaHLSL\\hlsl\\linalg\\outerproductaccumulate-spirv-errors.hlsl"
+
+
+RWByteAddressBuffer RWBuf;
+
+export void Test4(vector<half, 128> Input1, vector<half, 64> Input2) {
+  using namespace dx::linalg;
+
+  RWMatrixRef<DATA_TYPE_FLOAT16, 128, 64, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL>
+      matrix = {RWBuf, 0, 0};
+
+
+  OuterProductAccumulate(Input1, Input2, matrix);
+}

From d81fd22622e5f34f991964f3e77852ab7e310c37 Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista@microsoft.com>
Date: Tue, 22 Apr 2025 16:47:56 -0700
Subject: [PATCH 18/18] adjust error msg

---
 tools/clang/lib/Headers/hlsl/dx/linalg.h      |   2 +-
 .../outerproductaccumulate-spirv-errors.hlsl  |   2 +-
 .../outerproductaccumulate-spirv-errors.i     | 193 ------------------
 3 files changed, 2 insertions(+), 195 deletions(-)
 delete mode 100644 tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-spirv-errors.i

diff --git a/tools/clang/lib/Headers/hlsl/dx/linalg.h b/tools/clang/lib/Headers/hlsl/dx/linalg.h
index ed3d0a0184..51e662bbc9 100644
--- a/tools/clang/lib/Headers/hlsl/dx/linalg.h
+++ b/tools/clang/lib/Headers/hlsl/dx/linalg.h
@@ -1,7 +1,7 @@
 // Header for linear algebra APIs.
 
 #if __spirv__
-#error "Cooperative matrix/vectors not (yet) supported for SPIRV"
+#error "Cooperative vectors not (yet) supported for SPIRV"
 #endif
 
 #if ((__SHADER_TARGET_MAJOR > 6) ||                                            \
diff --git a/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-spirv-errors.hlsl b/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-spirv-errors.hlsl
index 3ec5998f75..0213103926 100644
--- a/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-spirv-errors.hlsl
+++ b/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-spirv-errors.hlsl
@@ -4,7 +4,7 @@
 // This is a copy of \tools\clang\test\CodeGenDXIL\hlsl\linalg\outerproductaccumulate.hlsl
 // except that spirv is targeted
 
-// expected-error@dx/linalg.h:4{{Cooperative matrix/vectors not (yet) supported for SPIRV}}
+// expected-error@dx/linalg.h:4{{Cooperative vectors not (yet) supported for SPIRV}}
 #include <dx/linalg.h>
 
 RWByteAddressBuffer RWBuf;
diff --git a/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-spirv-errors.i b/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-spirv-errors.i
deleted file mode 100644
index 44ac09f9e8..0000000000
--- a/tools/clang/test/SemaHLSL/hlsl/linalg/outerproductaccumulate-spirv-errors.i
+++ /dev/null
@@ -1,193 +0,0 @@
-#line 1 "D:\\DXC\\tools\\clang\\test\\SemaHLSL\\hlsl\\linalg\\outerproductaccumulate-spirv-errors.hlsl"
-
-
-
-
-
-
-#line 1 "D:/DXC/tools/clang/lib/Headers/hlsl\\dx/linalg.h"
-#line 11 "D:/DXC/tools/clang/lib/Headers/hlsl\\dx/linalg.h"
-namespace dx {
-namespace linalg {
-
-
-
-
-
-enum DataType {
-  DATA_TYPE_SINT16 = 2,
-  DATA_TYPE_UINT16 = 3,
-  DATA_TYPE_SINT32 = 4,
-  DATA_TYPE_UINT32 = 5,
-  DATA_TYPE_FLOAT16 = 8,
-  DATA_TYPE_FLOAT32 = 9,
-  DATA_TYPE_SINT8_T4_PACKED = 17,
-  DATA_TYPE_UINT8_T4_PACKED = 18,
-  DATA_TYPE_UINT8 = 19,
-  DATA_TYPE_SINT8 = 20,
-  DATA_TYPE_FLOAT8_E4M3 = 21,
-
-  DATA_TYPE_FLOAT8_E5M2 = 22,
-
-};
-
-enum MatrixLayout {
-  MATRIX_LAYOUT_ROW_MAJOR = 0,
-  MATRIX_LAYOUT_COLUMN_MAJOR = 1,
-  MATRIX_LAYOUT_MUL_OPTIMAL = 2,
-  MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL = 3
-};
-
-
-
-
-namespace details {
-template <typename T> bool IsUnsigned() { return false; }
-
-
-template <> bool IsUnsigned<uint16_t>() { return true; }
-
-
-template <> bool IsUnsigned<uint32_t>() { return true; }
-template <> bool IsUnsigned<uint64_t>() { return true; }
-}
-
-
-
-
-
-template <typename BufferTy, DataType DT, uint M, uint K, MatrixLayout ML,
-          bool Transpose>
-struct MatrixRefImpl {
-  BufferTy Buffer;
-  uint StartOffset;
-  uint Stride;
-};
-
-template <DataType DT, uint M, uint K, MatrixLayout ML, bool Transpose = false>
-using MatrixRef = MatrixRefImpl<ByteAddressBuffer, DT, M, K, ML, Transpose>;
-
-template <DataType DT, uint M, uint K, MatrixLayout ML, bool Transpose = false>
-using RWMatrixRef = MatrixRefImpl<RWByteAddressBuffer, DT, M, K, ML, Transpose>;
-
-
-
-
-
-template <typename BufferTy, DataType DT> struct VectorRefImpl {
-  BufferTy Buffer;
-  uint StartOffset;
-};
-
-template <DataType DT> using VectorRef = VectorRefImpl<ByteAddressBuffer, DT>;
-
-template <DataType DT>
-using RWVectorRef = VectorRefImpl<RWByteAddressBuffer, DT>;
-
-
-
-
-
-template <typename T, int N, DataType DT> struct InterpretedVector {
-  vector<T, N> Data;
-};
-
-template <DataType DT, typename T, int N>
-InterpretedVector<T, N, DT> MakeInterpretedVector(vector<T, N> Vec) {
-  InterpretedVector<T, N, DT> IV = {Vec};
-  return IV;
-}
-
-
-
-
-
-template <typename OutputElTy, typename InputElTy, int InputElCount,
-          typename MatrixBufferTy, DataType InputDT, DataType MatrixDT,
-          uint MatrixM, uint MatrixK, MatrixLayout MatrixLayout,
-          bool MatrixTranspose>
-vector<OutputElTy, MatrixM>
-Mul(MatrixRefImpl<MatrixBufferTy, MatrixDT, MatrixM, MatrixK, MatrixLayout,
-                  MatrixTranspose>
-        Matrix,
-    InterpretedVector<InputElTy, InputElCount, InputDT> InputVector) {
-
-  vector<OutputElTy, MatrixM> OutputVector;
-
-  __builtin_MatVecMul(
-              OutputVector, details::IsUnsigned<OutputElTy>(), InputVector.Data,
-      details::IsUnsigned<InputElTy>(), InputDT, Matrix.Buffer,
-      Matrix.StartOffset, MatrixDT, MatrixM, MatrixK, MatrixLayout,
-      MatrixTranspose, Matrix.Stride);
-
-  return OutputVector;
-}
-
-
-
-
-
-template <typename OutputElTy, typename InputElTy, int InputElCount,
-          typename MatrixBufferTy, DataType InputDT, DataType MatrixDT,
-          uint MatrixM, uint MatrixK, MatrixLayout MatrixLayout,
-          bool MatrixTranspose, typename BiasVectorBufferTy,
-          DataType BiasVectorDT>
-vector<OutputElTy, MatrixM>
-MulAdd(MatrixRefImpl<MatrixBufferTy, MatrixDT, MatrixM, MatrixK, MatrixLayout,
-                     MatrixTranspose>
-           Matrix,
-       InterpretedVector<InputElTy, InputElCount, InputDT> InputVector,
-       VectorRefImpl<BiasVectorBufferTy, BiasVectorDT> BiasVector) {
-
-  vector<OutputElTy, MatrixM> OutputVector;
-
-  __builtin_MatVecMulAdd(
-              OutputVector, details::IsUnsigned<OutputElTy>(), InputVector.Data,
-      details::IsUnsigned<InputElTy>(), InputDT, Matrix.Buffer,
-      Matrix.StartOffset, MatrixDT, MatrixM, MatrixK, MatrixLayout,
-      MatrixTranspose, Matrix.Stride, BiasVector.Buffer, BiasVector.StartOffset,
-      BiasVectorDT);
-
-  return OutputVector;
-}
-
-
-
-
-
-template <typename ElTy, int MatrixM, int MatrixN, DataType MatrixDT,
-          MatrixLayout MatrixLayout>
-void OuterProductAccumulate(
-    vector<ElTy, MatrixM> InputVector1, vector<ElTy, MatrixN> InputVector2,
-    RWMatrixRef<MatrixDT, MatrixM, MatrixN, MatrixLayout, false> Matrix) {
-  __builtin_OuterProductAccumulate(InputVector1, InputVector2, Matrix.Buffer,
-                                   Matrix.StartOffset, MatrixDT, MatrixLayout,
-                                   Matrix.Stride);
-}
-
-
-
-
-
-template <typename ElTy, int ElCount>
-void VectorAccumulate(vector<ElTy, ElCount> InputVector,
-                      RWByteAddressBuffer Buffer, uint Offset) {
-  __builtin_VectorAccumulate(InputVector, Buffer, Offset);
-}
-
-}
-}
-#line 7 "D:\\DXC\\tools\\clang\\test\\SemaHLSL\\hlsl\\linalg\\outerproductaccumulate-spirv-errors.hlsl"
-
-
-RWByteAddressBuffer RWBuf;
-
-export void Test4(vector<half, 128> Input1, vector<half, 64> Input2) {
-  using namespace dx::linalg;
-
-  RWMatrixRef<DATA_TYPE_FLOAT16, 128, 64, MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL>
-      matrix = {RWBuf, 0, 0};
-
-
-  OuterProductAccumulate(Input1, Input2, matrix);
-}