[0029][0031]In linalg Mul/MulAdd functions, the memory backing the Matrix and Bias buffers must be read-only

proposals/0029-cooperative-vector.md

@@ -84,10 +84,10 @@ this should give a sense of how these new operations can be used.
 > NOTE: see proposal [0031] for full details on the HLSL API.
 
 ```c++
-ByteAddressBuffer inputMatrix0; 
-ByteAddressBuffer inputMatrix1; 
-ByteAddressBuffer biasVector0; 
-ByteAddressBuffer biasVector1;
+ByteAddressBuffer inputMatrix0; // note read-only buffer
+ByteAddressBuffer inputMatrix1; // note read-only buffer
+ByteAddressBuffer biasVector0;  // note read-only buffer
+ByteAddressBuffer biasVector1;  // note read-only buffer
 
 void ps_main(args) // args: texture, normal, position
 {   
@@ -229,9 +229,11 @@ an unsigned integer.
 
 ##### Matrix
 
-The matrix is loaded from a raw-buffer, **matrix resource**,  starting at
-**matrix offset**. The **matrix interpretation** argument specifies the element
-type of the matrix (see [Type Interpretations]), no conversion is performed.
+The matrix is loaded from a read-only raw-buffer, **matrix resource**,  starting
+at **matrix offset**. The **matrix interpretation** argument specifies the
+element type of the matrix (see [Type Interpretations]), no conversion is
+performed.
+
 The **matrix M dimension** and **matrix K dimension** arguments specify the
 dimensions of the matrix. The **matrix layout** argument specifies the layout
 of the matrix (see [Matrix Layouts]). If the **matrix transpose** is non-zero
@@ -255,10 +257,10 @@ the matrix load is out of bounds then the entire operation is undefined.
 
 ##### Bias Vector
 
-The bias vector is loaded from the raw-buffer, **bias vector resource**,
-starting at **bias vector offset**. The **bias vector interpretation** argument
-specifies the element type of the bias vector (see [Type Interpretations]), no
-conversion is performed.
+The bias vector is loaded from the read-only raw-buffer, **bias vector
+resource**, starting at **bias vector offset**. The **bias vector
+interpretation** argument specifies the element type of the bias vector (see
+[Type Interpretations]), no conversion is performed.
 
 Only non-packed interpretations are valid for bias vectors.
 

proposals/0031-hlsl-vector-matrix-operations.md

@@ -459,7 +459,10 @@ InterpretedVector<T, N, DT> MakeInterpretedVector(vector<T, N> Vec) {
 ### Function: Mul
 
 The `dx::linalg::Mul` function performs a matrix-vector multiplication. The
-matrix is stored in memory, while the vector comes from a variable.
+matrix is stored in memory, while the vector comes from a variable. The Mul
+function expects the memory backing the matrix object to be read-only
+(for efficient loads and hardware optimizations) and, therefore, the matrix
+must be of MatrixRef type (and not RWMatrixRef type).
 
 > TODO: add an example for packed types, and make sure they work correctly
 
@@ -530,7 +533,9 @@ Mul(MatrixRefImpl<MatrixBufferTy, MatrixDT, MatrixM, MatrixK, MatrixLayout,
 ## Function: MulAdd
 
 The `dx::linalg::MulAdd` function behaves as `dx::linalg::Mul`, but also adds a
-bias vector (loaded from memory) to the result.
+bias vector (loaded from memory) to the result. Similar to the the matrix
+operand, the memory backing the bias vector must be read-only, and therefore an
+object of `VectorRef` type (and not `RWVectorRef` type).
 
 Example: