Implement HLSL Diagnostics for LinAlg operations (microsoft#7430)

This PR implements checks to validate the linalg builtin functions:
__builtin_MatVecMul, __builtin_MatVecMulAdd,
__builtin_OuterProductAccumulate and __builtin_VectorAccumulate. This
includes:
- verify valid types for input and output vectors
- const checks for compile-time const parameters
- value checks for interpretation and layout (enum) parameters
- min/max checks for matrix dimensions
- verify input-output vector are the right dimensions for the given
matrix (dimM and dimK) : packed and unpacked cases
- verify matrix layout, transpose and stride rules
- incorrect shader model warning

Adds tests for all the above error checks.

Implements
microsoft#7336

Author: anupamachandra

include/dxc/dxcapi.internal.h

@@ -133,11 +133,15 @@ enum LEGAL_INTRINSIC_COMPTYPES {
   LICOMPTYPE_HIT_OBJECT = 51,
   LICOMPTYPE_RAY_QUERY = 52,
 
+  LICOMPTYPE_LINALG = 53, // f32, partial-precision-f32, f16,
+                          // i32, i16, u32, u16,
+                          // int8_4packed, uint8_4packed
+
 #ifdef ENABLE_SPIRV_CODEGEN
-  LICOMPTYPE_VK_BUFFER_POINTER = 53,
-  LICOMPTYPE_COUNT = 54
+  LICOMPTYPE_VK_BUFFER_POINTER = 54,
+  LICOMPTYPE_COUNT = 55
 #else
-  LICOMPTYPE_COUNT = 53
+  LICOMPTYPE_COUNT = 54
 #endif
 };
 

tools/clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -8018,6 +8018,43 @@ def err_hlsl_reorder_unsupported_stage : Error<
    "dx::MaybeReorderThread is unavailable in shader stage '%0' (requires 'raygeneration')">;
 def err_hlsl_hitobject_unsupported_stage : Error<
    "dx::HitObject is unavailable in shader stage '%0' (requires 'raygeneration', 'closesthit' or 'miss')">;
+
+// Linear Algebra Operations
+def err_hlsl_linalg_isunsigned_incorrect_for_given_type : Error<
+  "%0 must be %select{false|true}1 for vector of "
+  "%select{floating point|signed integer|unsigned integer}2 type">;
+def err_hlsl_linalg_interpretation_value_incorrect : Error<
+  "%0 is an invalid %select{memory|register}1 interpretation value">;
+def err_hlsl_linalg_matrix_layout_is_not_transposable : Error<
+  "RowMajor and ColumnMajor matrices are not transposable">;
+def err_hlsl_linalg_optimal_matrix_layout_matrix_stride_must_be_zero : Error<
+  "for optimal matrix layout, matrix stride must be 0">;
+def err_hlsl_linalg_matrix_dim_must_be_greater_than_zero: Error<
+  "matrix dimension must be greater than 0">;
+def err_hlsl_linalg_matrix_layout_invalid : Error<
+  "matrix layout %0 is not valid, must be in the range [%1, %2]">;
+
+def err_hlsl_linalg_mul_muladd_output_vector_size_not_equal_to_matrix_M : Error<
+  "output vector length must be equal to Matrix M dimension in a linalg Mul/MulAdd operation">;
+def err_hlsl_linalg_mul_muladd_unpacked_input_vector_size_not_equal_to_matrix_K : Error<
+  "unpacked input vector length must be equal to Matrix K dimension in a linalg Mul/MulAdd operation">;
+def err_hlsl_linalg_mul_muladd_packed_input_vector_size_incorrect : Error<
+  "packed input vector length must be the smallest number that can hold matrix dim K values of the "
+  "packed(smaller) type in linalg mul/muladd operations">;
+def err_hlsl_linalg_mul_muladd_isUnsigned_for_packed_input_must_be_true : Error<
+  "IsInputUnsigned must be true for packed input interpretations in linalg mul/muladd operations">;
+def err_hlsl_linalg_mul_muladd_packed_input_vector_must_be_uint : Error<
+  "packed input vector type must be a 32-bit unsigned int type in linalg mul/muladd operations">;
+def err_hlsl_linalg_mul_muladd_invalid_dim: Error<
+  "matrix dimension %select{M|K when using unpacked input vectors|K "
+  "when using packed input vectors}0 must be less than %1, in a linalg "
+  "Mul/MulAdd operation">;
+
+def err_hlsl_linalg_outer_prod_acc_vector_type_mismatch : Error<
+  "input vectors of outerproductaccumulate must have the same element type">;
+def err_hlsl_linalg_outer_prod_acc_matrix_layout_must_be_outer_prod_acc_optimal : Error<
+  "matrix layout for outerproductaccumulate must be %0">;
+
 // HLSL Change Ends
 
 // SPIRV Change Starts

tools/clang/lib/Headers/hlsl/dx/linalg.h

@@ -43,14 +43,30 @@ enum MatrixLayout {
 // Helper for signedness
 //
 namespace details {
-template <typename T> bool IsUnsigned() { return false; }
+
+template <typename T> struct IsUnsigned {};
+
+#define _SPECIALIZE_ISUNSIGNED(type, value)                                    \
+  template <> struct IsUnsigned<type> {                                        \
+    static const bool Value = value;                                           \
+  }
+
+_SPECIALIZE_ISUNSIGNED(uint8_t4_packed, true);
+_SPECIALIZE_ISUNSIGNED(int8_t4_packed, true);
+_SPECIALIZE_ISUNSIGNED(uint32_t, true);
+_SPECIALIZE_ISUNSIGNED(int32_t, false);
+_SPECIALIZE_ISUNSIGNED(float32_t, false);
 
 #ifdef __HLSL_ENABLE_16_BIT
-template <> bool IsUnsigned<uint16_t>() { return true; }
-#endif
+_SPECIALIZE_ISUNSIGNED(uint16_t, true);
+_SPECIALIZE_ISUNSIGNED(int16_t, false);
+_SPECIALIZE_ISUNSIGNED(float16_t, false);
+#else  // //__HLSL_ENABLE_16_BIT
+_SPECIALIZE_ISUNSIGNED(half, false);
+#endif //__HLSL_ENABLE_16_BIT
+
+#undef _SPECIALIZE_ISUNSIGNED
 
-template <> bool IsUnsigned<uint32_t>() { return true; }
-template <> bool IsUnsigned<uint64_t>() { return true; }
 } // namespace details
 
 //
@@ -116,10 +132,10 @@ Mul(MatrixRefImpl<MatrixBufferTy, MatrixDT, MatrixM, MatrixK, MatrixLayout,
   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);
+      /*out*/ OutputVector, details::IsUnsigned<OutputElTy>::Value,
+      InputVector.Data, details::IsUnsigned<InputElTy>::Value, InputDT,
+      Matrix.Buffer, Matrix.StartOffset, MatrixDT, MatrixM, MatrixK,
+      MatrixLayout, MatrixTranspose, Matrix.Stride);
 
   return OutputVector;
 }
@@ -143,11 +159,11 @@ MulAdd(MatrixRefImpl<MatrixBufferTy, MatrixDT, MatrixM, MatrixK, MatrixLayout,
   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);
+      /*out*/ OutputVector, details::IsUnsigned<OutputElTy>::Value,
+      InputVector.Data, details::IsUnsigned<InputElTy>::Value, InputDT,
+      Matrix.Buffer, Matrix.StartOffset, MatrixDT, MatrixM, MatrixK,
+      MatrixLayout, MatrixTranspose, Matrix.Stride, BiasVector.Buffer,
+      BiasVector.StartOffset, BiasVectorDT);
 
   return OutputVector;
 }