[FRONTEND] Add descriptive messages to bare asserts (triton-lang#10405)

## Summary
Improve error diagnostics for user-facing `assert` statements in
`semantic.py` and `core.py` by adding descriptive messages following
the "expected X, got Y" pattern.

As requested by @lezcano in triton-lang#10341 — previously, users hitting these
asserts (e.g., mismatched accumulator dtype in `tl.dot`) would get an
empty `AssertionError` with no guidance on what went wrong.

## Changes
- `semantic.py`: 17 bare asserts → descriptive messages
- `core.py`: 5 bare asserts → descriptive messages
- Split compound `assert shape == X and dtype == Y` into two separate
  asserts for clearer diagnostics

Author: Chennesxu

python/triton/language/core.py

@@ -3250,7 +3250,7 @@ def selu(x, alpha):
         :return: one tensor or a tuple of tensors, depending on the mapped function.
     '''
     # Build the block for the nested region first to discover the return types
-    assert pack >= 1
+    assert pack >= 1, f"pack must be >= 1, got {pack}"
     in_scalar_tys = [t.type.scalar for t in args]
     builder = _semantic.builder
     block = builder.new_block()
@@ -3882,8 +3882,8 @@ def builtin_max(*args, propagate_nan=_NOTHING, _semantic=None):
     is_constexpr = all(not isinstance(x, base_value) for x in args)
     if is_constexpr:
         assert propagate_nan is _NOTHING, "propagate_nan is not supported on builtin max"
-        assert not any(math.isnan(x) for x in args)
-        assert not any(is_negative_zero(x) for x in args)
+        assert not any(math.isnan(x) for x in args), "constexpr max does not support NaN values"
+        assert not any(is_negative_zero(x) for x in args), "constexpr max does not support negative zero"
         return constexpr(builtins.max(_unwrap_if_constexpr(args)))
 
     if propagate_nan is _NOTHING:
@@ -3906,8 +3906,8 @@ def builtin_min(*args, propagate_nan=_NOTHING, _semantic=None):
     is_constexpr = all(not isinstance(x, base_value) for x in args)
     if is_constexpr:
         assert propagate_nan is _NOTHING, "propagate_nan is not supported on builtin min"
-        assert not any(math.isnan(x) for x in args)
-        assert not any(is_negative_zero(x) for x in args)
+        assert not any(math.isnan(x) for x in args), "constexpr min does not support NaN values"
+        assert not any(is_negative_zero(x) for x in args), "constexpr min does not support negative zero"
         return constexpr(builtins.min(_unwrap_if_constexpr(args)))
 
     if propagate_nan is _NOTHING:

python/triton/language/semantic.py

@@ -214,8 +214,8 @@ def binary_op_sanitize_overflow_impl(self, lhs: TensorTy, rhs: TensorTy, binary_
             return
         lhs_sca_ty = lhs.type.scalar
         rhs_sca_ty = rhs.type.scalar
-        assert lhs_sca_ty == rhs_sca_ty
-        assert lhs_sca_ty.is_int()
+        assert lhs_sca_ty == rhs_sca_ty, f"expected matching operand types, got {lhs_sca_ty} and {rhs_sca_ty}"
+        assert lhs_sca_ty.is_int(), f"expected integer type, got {lhs_sca_ty}"
         lhs = self.cast(lhs, tl.int64)
         rhs = self.cast(rhs, tl.int64)
         ret = binary_op(lhs, rhs, False)
@@ -657,7 +657,7 @@ def expand_dims(self, input: TensorTy, axis: int) -> TensorTy:
 
     def cat(self, lhs: TensorTy, rhs: TensorTy, can_reorder: bool) -> TensorTy:
         assert can_reorder, "current implementation of `cat` always may reorder elements"
-        assert len(lhs.shape) == 1
+        assert len(lhs.shape) == 1, f"expected 1D input for cat, got {len(lhs.shape)}D"
         ret_type = tl.block_type(lhs.type.scalar, [lhs.shape[0] + rhs.shape[0]])
         return self.tensor(self.builder.create_cat(lhs.handle, rhs.handle), ret_type)
 
@@ -686,8 +686,9 @@ def join(self, a: TensorTy, b: TensorTy) -> TensorTy:
         return ret
 
     def split(self, a: TensorTy) -> Tuple[TensorTy, TensorTy]:
-        assert (len(a.shape) > 0)
-        assert (tl._unwrap_if_constexpr(a.shape[-1]) == 2)
+        assert (len(a.shape) > 0), "split requires a non-scalar tensor"
+        assert (tl._unwrap_if_constexpr(a.shape[-1]) == 2), \
+            f"expected last dimension to be 2 for split, got {tl._unwrap_if_constexpr(a.shape[-1])}"
 
         new_shape = a.shape[:-1]
         ret_type = tl.block_type(a.type.scalar, new_shape)
@@ -754,7 +755,8 @@ def broadcast_impl_value(self, lhs: TensorTy, rhs: TensorTy) -> TensorTy:
                                       tl.block_type(rhs_ty.scalar, [1] + rhs_shape.values))
                     rhs_ty = rhs.type
                     rhs_shape = rhs_ty.get_block_shapes()
-            assert len(rhs_shape) == len(lhs_shape)
+            assert len(rhs_shape) == len(lhs_shape), \
+                f"expected tensors of equal rank for broadcast, got {len(lhs_shape)} and {len(rhs_shape)}"
 
             ret_shape = []
             for i, left in enumerate(lhs_shape):
@@ -1062,7 +1064,8 @@ def load(self, ptr: TensorTy, mask: Optional[TensorTy], other: Optional[TensorTy
 
     def descriptor_load(self, desc: tl.tensor_descriptor_base, offsets, cache_modifier: str,
                         eviction_policy: str) -> TensorTy:
-        assert isinstance(desc, tl.tensor_descriptor_base)
+        assert isinstance(desc, tl.tensor_descriptor_base), \
+            f"expected a tensor descriptor, got {type(desc).__name__}"
         ndim = len(desc.block_shape)
         assert len(offsets) == ndim, f"expected {ndim} offsets, but got {len(offsets)}"
 
@@ -1072,10 +1075,12 @@ def descriptor_load(self, desc: tl.tensor_descriptor_base, offsets, cache_modifi
         return self.tensor(x, desc.block_type)
 
     def validate_store_like(self, desc: tl.tensor_descriptor_base, value: TensorTy, offsets) -> None:
-        assert isinstance(desc, tl.tensor_descriptor_base)
+        assert isinstance(desc, tl.tensor_descriptor_base), \
+            f"expected a tensor descriptor, got {type(desc).__name__}"
         ndim = len(desc.block_shape)
         assert len(offsets) == ndim, f"expected {ndim} offsets, but got {len(offsets)}"
-        assert value.shape == desc.block_shape
+        assert value.shape == desc.block_shape, \
+            f"expected value shape {desc.block_shape}, got {value.shape}"
 
     def descriptor_store(self, desc: tl.tensor_descriptor_base, value: TensorTy, offsets) -> TensorTy:
         self.validate_store_like(desc, value, offsets)
@@ -1136,7 +1141,8 @@ def descriptor_atomic_xor(self, desc: tl.tensor_descriptor_base, value: TensorTy
         return self.tensor(self.builder.create_descriptor_reduce(kind, desc.handle, value.handle, offsets), tl.void)
 
     def descriptor_gather(self, desc, x_offsets, y_offset, cache_modifier: str, eviction_policy: str) -> TensorTy:
-        assert isinstance(desc, tl.tensor_descriptor_base)
+        assert isinstance(desc, tl.tensor_descriptor_base), \
+            f"expected a tensor descriptor, got {desc.__class__.__name__}"
         assert cache_modifier == "", "cache modifier is not supported yet"
         assert eviction_policy == "", "eviction policy is not supported yet"
 
@@ -1162,7 +1168,8 @@ def descriptor_gather(self, desc, x_offsets, y_offset, cache_modifier: str, evic
         return self.tensor(x, type)
 
     def descriptor_scatter(self, desc, value: TensorTy, x_offsets, y_offset) -> TensorTy:
-        assert isinstance(desc, tl.tensor_descriptor_base)
+        assert isinstance(desc, tl.tensor_descriptor_base), \
+            f"expected a tensor descriptor, got {type(desc).__name__}"
 
         # Validate descriptor.
         assert len(desc.block_shape) == 2, f"descriptor must be 2D, but got {desc.block_shape}"
@@ -1421,7 +1428,7 @@ def _str_to_dot_input_precision(self, input_precision):
 
     def dot(self, lhs: TensorTy, rhs: TensorTy, acc: TensorTy, input_precision: Optional[str],
             max_num_imprecise_acc: int, out_dtype: tl.dtype | None) -> TensorTy:
-        assert lhs.type.is_block() and rhs.type.is_block()
+        assert lhs.type.is_block() and rhs.type.is_block(), "dot operands must be block tensors (not scalars)"
 
         if lhs.dtype.is_fp8() and rhs.dtype.is_fp8():
             # All combinations of supported fp8 x fp8 are permitted
@@ -1500,7 +1507,11 @@ def dot(self, lhs: TensorTy, rhs: TensorTy, acc: TensorTy, input_precision: Opti
             acc_handle = self.builder.create_splat(ret_ty.to_ir(self.builder), _0)
         else:
             acc_handle = acc.handle
-            assert acc.type.shape == ret_ty.shape and acc.type.element_ty == out_dtype
+            assert acc.type.shape == ret_ty.shape, \
+                f"expected accumulator shape {ret_ty.shape}, got {acc.type.shape}"
+            assert acc.type.element_ty == out_dtype, \
+                f"expected accumulator dtype {out_dtype}, got {acc.type.element_ty}; " \
+                f"pass out_dtype={acc.type.element_ty} to use this accumulator dtype"
 
         # max_num_imprecise_acc only applies to fp8 -> fp32 dot on sm_90
         if max_num_imprecise_acc is None:
@@ -1565,7 +1576,7 @@ def verify_scaled_shape(self, M, N, K, lhs_scale, rhs_scale, scale_factor):
     def dot_scaled(self, lhs: TensorTy, lhs_scale: TensorTy, lhs_format: str, rhs: TensorTy,
                    rhs_scale: Optional[TensorTy], rhs_format: str, acc: TensorTy | None, fast_math: bool,
                    lhs_k_pack: bool, rhs_k_pack: bool, out_dtype: tl.dtype) -> TensorTy:
-        assert lhs.type.is_block() and rhs.type.is_block()
+        assert lhs.type.is_block() and rhs.type.is_block(), "dot_scaled operands must be block tensors (not scalars)"
         #TODO: validate types.
         lhs_rank = len(lhs.shape)
         rhs_rank = len(rhs.shape)
@@ -1604,7 +1615,11 @@ def dot_scaled(self, lhs: TensorTy, lhs_scale: TensorTy, lhs_format: str, rhs: T
             acc_handle = self.builder.create_splat(ret_ty.to_ir(self.builder), _0)
         else:
             acc_handle = acc.handle
-            assert acc.type.shape == ret_ty.shape and acc.type.element_ty == out_dtype
+            assert acc.type.shape == ret_ty.shape, \
+                f"expected accumulator shape {ret_ty.shape}, got {acc.type.shape}"
+            assert acc.type.element_ty == out_dtype, \
+                f"expected accumulator dtype {out_dtype}, got {acc.type.element_ty}; " \
+                f"pass out_dtype={acc.type.element_ty} to use this accumulator dtype"
         rhs_scale_handle = None if rhs_scale_is_none else rhs_scale.handle
         lhs_scale_handle = None if lhs_scale_is_none else lhs_scale.handle
 
@@ -1842,7 +1857,7 @@ def make_tensor_descriptor(self, base: TensorTy, shape: List[TensorTy], strides:
             raise ValueError(f"Expected {ndim} strides but got {len(strides)}")
         if len(block_shape) != ndim:
             raise ValueError(f"Expected block_shape to have {ndim} dimensions but got {len(strides)}")
-        assert isinstance(base.dtype, tl.pointer_type)
+        assert isinstance(base.dtype, tl.pointer_type), f"base must be a pointer type, got {base.dtype}"
         elem_size = base.dtype.element_ty.primitive_bitwidth // 8
         contig_dim_size = tl._unwrap_if_constexpr(block_shape[-1])
         if contig_dim_size * elem_size < 16:
@@ -1860,7 +1875,7 @@ def make_tensor_descriptor(self, base: TensorTy, shape: List[TensorTy], strides:
         # Check whether `block_shape` is static
         block_shape = tl._unwrap_shape(block_shape)
 
-        assert isinstance(base.type, tl.pointer_type)
+        assert isinstance(base.type, tl.pointer_type), f"base must be a pointer type, got {base.type}"
         type = tl.block_type(base.type.element_ty, block_shape)
         base_handle = base.handle
         is_signed_int = base.type.element_ty.is_int_signed()