[TIR] Allow symbolic bounds in IndexMap analysis

include/tvm/tir/index_map.h

@@ -102,8 +102,7 @@ class IndexMapNode : public Object {
    * \returns The indices in the output space.  Contains one value for
    * each expression in `final_indices`.
    */
-  Array<PrimExpr> MapIndices(const Array<PrimExpr>& indices,
-                             arith::Analyzer* analyzer = nullptr) const;
+  Array<PrimExpr> MapIndices(const Array<PrimExpr>& indices, arith::Analyzer* analyzer) const;
 
   /*! \brief Map a memory range to the output space
    *
@@ -121,7 +120,7 @@ class IndexMapNode : public Object {
    * \returns The ranges in the output space.  Contains one value for
    * each expression in `final_indices`.
    */
-  Array<Range> MapRanges(const Array<Range>& ranges, arith::Analyzer* analyzer = nullptr) const;
+  Array<Range> MapRanges(const Array<Range>& ranges, arith::Analyzer* analyzer) const;
 
   /*! \brief Map a buffer shape to the output space
    *
@@ -134,7 +133,7 @@ class IndexMapNode : public Object {
    * \returns The buffer shape in the output space.  Contains one
    * value for each expression in `final_indices`.
    */
-  Array<PrimExpr> MapShape(const Array<PrimExpr>& shape, arith::Analyzer* analyzer = nullptr) const;
+  Array<PrimExpr> MapShape(const Array<PrimExpr>& shape, arith::Analyzer* analyzer) const;
 
   /* \brief Map an NDArray according to this index map
    *
@@ -203,7 +202,7 @@ class IndexMap : public ObjectRef {
    * If the user has supplied an `inverse_index_map`, that map is
    * assumed to be correct and bijective, and is returned.
    */
-  IndexMap Inverse(Array<Range> initial_ranges) const;
+  IndexMap Inverse(Array<Range> initial_ranges, arith::Analyzer* analyzer) const;
 
   /*! \brief Rename the variables in the index map and ensure the names are unique.
    *
@@ -225,7 +224,8 @@ class IndexMap : public ObjectRef {
    * \return The inverted index map, along with the predicate for
    * which the inverse maps to a valid range.
    */
-  std::pair<IndexMap, PrimExpr> NonSurjectiveInverse(Array<Range> initial_ranges) const;
+  std::pair<IndexMap, PrimExpr> NonSurjectiveInverse(Array<Range> initial_ranges,
+                                                     arith::Analyzer* analyzer) const;
 
   TVM_DEFINE_OBJECT_REF_METHODS(IndexMap, ObjectRef, IndexMapNode);
 };

include/tvm/topi/transform.h

@@ -24,6 +24,7 @@
 #ifndef TVM_TOPI_TRANSFORM_H_
 #define TVM_TOPI_TRANSFORM_H_
 
+#include <tvm/arith/analyzer.h>
 #include <tvm/te/operation.h>
 #include <tvm/tir/data_layout.h>
 #include <tvm/tir/index_map.h>
@@ -1738,16 +1739,18 @@ inline Tensor auto_scheduler_layout_transform(const Tensor& src, const String& s
 inline Tensor meta_schedule_layout_transform(const Tensor& src, const tir::IndexMap& index_map,
                                              const String name = "T_meta_schedule_layout_trans",
                                              const String tag = kInjective) {
+  arith::Analyzer analyzer;
   Array<Range> iter_domain;
   iter_domain.reserve(src->shape.size());
   for (const PrimExpr& e : src->shape) {
     iter_domain.push_back(Range::FromMinExtent(make_zero(e->dtype), e));
   }
-  Array<PrimExpr> post_transform_shape = index_map->MapShape(src->shape);
+  Array<PrimExpr> post_transform_shape = index_map->MapShape(src->shape, &analyzer);
   return compute(
       post_transform_shape,
-      [src, inv = index_map.Inverse(iter_domain)](const Array<Var>& indices) -> PrimExpr {
-        return src(inv->MapIndices(Array<PrimExpr>{indices.begin(), indices.end()}));
+      [src, inv = index_map.Inverse(iter_domain, &analyzer),
+       &analyzer](const Array<Var>& indices) -> PrimExpr {
+        return src(inv->MapIndices(Array<PrimExpr>{indices.begin(), indices.end()}, &analyzer));
       },
       name, tag);
 }

pyproject.toml

@@ -14,6 +14,9 @@
 # KIND, either express or implied.  See the License for the
 # specific language governing permissions and limitations
 # under the License.
+[tool.isort]
+profile = "black"
+src_paths = ["python", "tests/python"]
 
 [tool.black]
 line-length = 100

python/tvm/te/schedule.py

@@ -22,13 +22,12 @@
 
 import tvm._ffi
 from tvm._ffi.base import string_types
-
-from tvm.runtime import Object, convert
 from tvm.ir import container as _container
-from tvm.tir import IterVar, Buffer, Var, IndexMap
+from tvm.runtime import Object, convert
+from tvm.tir import Buffer, IndexMap, IterVar, Var
 
-from . import tensor as _tensor
 from . import _ffi_api
+from . import tensor as _tensor
 
 
 @tvm._ffi.register_object
@@ -600,7 +599,9 @@ def transform_layout(self, mapping_function: Callable[..., List[tvm.tir.PrimExpr
         """
 
         ndim = len(self.op.output(0).shape)
-        index_map, axis_separators = IndexMap.from_func_with_separators(mapping_function, ndim=ndim)
+        index_map, axis_separators = IndexMap.from_func_with_separators(
+            mapping_function, ndim=ndim, index_dtype="int32"
+        )
 
         new_iter_vars = _ffi_api.StageTransformLayout(
             self, index_map.initial_indices, index_map.final_indices

python/tvm/tir/function.py

@@ -67,7 +67,6 @@ def __init__(
         attrs=None,
         span=None,
     ):
-
         param_list = []
         buffer_map = {} if buffer_map is None else buffer_map
         for x in params:
@@ -266,6 +265,8 @@ def from_func(
         mapping_function: Callable,
         ndim: Optional[int] = None,
         inverse_index_map: Union[Callable, Optional["IndexMap"]] = None,
+        *,
+        index_dtype: str = "int64",
     ):
         """Create an index map from a function
 
@@ -302,7 +303,10 @@ def from_func(
 
         """
         index_map, axis_separators = IndexMap.from_func_with_separators(
-            mapping_function, ndim, inverse_index_map
+            mapping_function,
+            ndim,
+            inverse_index_map,
+            index_dtype=index_dtype,
         )
         assert not axis_separators, (
             "The mapping_function provided to IndexMap.from_func "
@@ -316,6 +320,8 @@ def from_func_with_separators(
         mapping_function: Callable,
         ndim: Optional[int] = None,
         inverse_index_map: Union[Callable, Optional["IndexMap"]] = None,
+        *,
+        index_dtype: str = "int64",
     ):
         """Create an index map from a function
 
@@ -346,6 +352,9 @@ def from_func_with_separators(
             It is the user's responsibility to ensure the correctness of the pre-defined inverse
             index map.
 
+        index_dtype : str
+            The default index dtype to use for input iters in the mapping function.
+
         Returns
         -------
         ret: Tuple[IndexMap, List[int]]
@@ -361,20 +370,19 @@ def from_func_with_separators(
         args = []
         var_arg_name = None
         kwargs = collections.OrderedDict()
-        default_index_dtype = "int32"
 
         for name, param in params.items():
             if param.kind in [
                 inspect.Parameter.POSITIONAL_ONLY,
                 inspect.Parameter.POSITIONAL_OR_KEYWORD,
             ]:
-                args.append(tvm.tir.Var(name, default_index_dtype))
+                args.append(tvm.tir.Var(name, index_dtype))
 
             elif param.kind == inspect.Parameter.VAR_POSITIONAL:
                 var_arg_name = name
 
             elif param.kind == inspect.Parameter.KEYWORD_ONLY:
-                kwargs[name] = tvm.tir.Var(name, default_index_dtype)
+                kwargs[name] = tvm.tir.Var(name, index_dtype)
 
             else:
                 raise ValueError("transform_layout mapping may not have *args")
@@ -386,7 +394,7 @@ def from_func_with_separators(
             assert ndim is not None, "ndim must be specified when *args is used"
             num_var_args = ndim - len(args) - len(kwargs)
             for i in range(num_var_args):
-                args.append(tvm.tir.Var(f"{var_arg_name}_{i}", default_index_dtype))
+                args.append(tvm.tir.Var(f"{var_arg_name}_{i}", index_dtype))
 
         mapping = mapping_function(*args, **kwargs)
 

python/tvm/tir/schedule/schedule.py

@@ -93,6 +93,15 @@ def _parse_seed(seed: Optional[int]) -> int:
     return seed
 
 
+def _get_block_default_dtype(block: Block) -> str:
+    for i in block.iter_vars:
+        return i.var.dtype
+    for buffer_region in list(block.reads) + list(block.writes):
+        for dom in buffer_region.region:
+            return dom.min.dtype
+    return "int64"
+
+
 @_register_object("tir.Schedule")
 class Schedule(Object):
     """The user-facing schedule class
@@ -1492,7 +1501,10 @@ def after_reindex_cache_read(a: T.handle, b: T.handle) -> None:
         block = self._normalize_block_arg(block)
 
         if callable(index_map):
-            index_map = IndexMap.from_func(index_map)
+            index_map = IndexMap.from_func(
+                index_map,
+                index_dtype=_get_block_default_dtype(self.get(block)),
+            )
         return _ffi_api.ScheduleReindexCacheRead(  # type: ignore # pylint: disable=no-member
             self, block, read_buffer_index, storage_scope, index_map
         )
@@ -1589,7 +1601,10 @@ def after_cache_write(a: T.handle, b: T.handle) -> None:
         block = self._normalize_block_arg(block)
 
         if callable(index_map):
-            index_map = IndexMap.from_func(index_map)
+            index_map = IndexMap.from_func(
+                index_map,
+                index_dtype=_get_block_default_dtype(self.get(block)),
+            )
         return _ffi_api.ScheduleReindexCacheWrite(  # type: ignore # pylint: disable=no-member
             self, block, write_buffer_index, storage_scope, index_map
         )
@@ -3246,14 +3261,22 @@ def two_elementwise_transformed_intermediate_buffer(a: T.handle, c: T.handle) ->
 
         ndim = len(buffer_obj.shape)
         if callable(index_map):
-            index_map, axis_separators = IndexMap.from_func_with_separators(index_map, ndim=ndim)
+            index_map, axis_separators = IndexMap.from_func_with_separators(
+                index_map,
+                ndim=ndim,
+                index_dtype=_get_block_default_dtype(self.get(block)),
+            )
         else:
             axis_separators = []
 
         if pad_value is None:
             pass
         elif callable(pad_value):
-            pad_value = IndexMap.from_func(pad_value, ndim=len(index_map.final_indices))
+            pad_value = IndexMap.from_func(
+                pad_value,
+                ndim=len(index_map.final_indices),
+                index_dtype=_get_block_default_dtype(self.get(block)),
+            )
         elif not isinstance(pad_value, IndexMap):
             # Explicitly convert python int/float arguments to the
             # buffer's type.  If the default `tvm.runtime.convert`
@@ -3264,7 +3287,9 @@ def two_elementwise_transformed_intermediate_buffer(a: T.handle, c: T.handle) ->
             elif "float" in buffer_obj.dtype and isinstance(pad_value, float):
                 pad_value = FloatImm(buffer_obj.dtype, pad_value)
             pad_value = IndexMap.from_func(
-                lambda *indices: pad_value, ndim=len(index_map.final_indices)
+                lambda *indices: pad_value,
+                ndim=len(index_map.final_indices),
+                index_dtype=_get_block_default_dtype(self.get(block)),
             )
 
         buffer_index_type_enum = 0 if buffer_index_type == "read" else 1
@@ -3337,7 +3362,10 @@ def after_transform_block_layout(
         """
         block = self._normalize_block_arg(block)
         if callable(index_map):
-            index_map = IndexMap.from_func(index_map)
+            index_map = IndexMap.from_func(
+                index_map,
+                index_dtype=_get_block_default_dtype(self.get(block)),
+            )
         _ffi_api.ScheduleTransformBlockLayout(  # type: ignore # pylint: disable=no-member
             self, block, index_map
         )

python/tvm/tir/schedule/testing.py

@@ -51,6 +51,8 @@ def verify_trace_roundtrip(
         The text format or formats whose round-trip behavior should be
         validated.  If a single string, validate round-trips through
     """
+    from tvm.script import tir as T  # pylint: disable=import-outside-toplevel
+
     if not isinstance(text_format, str):
         for opt in text_format:
             new_sch = verify_trace_roundtrip(sch, mod, debug_mask=debug_mask, text_format=opt)
@@ -66,7 +68,9 @@ def verify_trace_roundtrip(
         Trace.apply_json_to_schedule(json_obj=json_obj, sch=new_sch)
     elif text_format == "python":
         py_trace = "\n".join(trace.as_python())
-        exec(py_trace, tvm.tir.__dict__, {"sch": new_sch})  # pylint: disable=exec-used
+        vars_dict = {"T": T}
+        vars_dict.update(tvm.tir.__dict__)
+        exec(py_trace, vars_dict, {"sch": new_sch})  # pylint: disable=exec-used
     else:
         assert text_format in ("json", "python"), f"Unknown text format: {text_format}"
 

src/arith/ir_mutator_with_analyzer.cc

@@ -22,6 +22,7 @@
  */
 #include "ir_mutator_with_analyzer.h"
 
+#include <tvm/arith/iter_affine_map.h>
 #include <tvm/tir/analysis.h>
 #include <tvm/tir/op.h>
 
@@ -39,6 +40,25 @@ void IRMutatorWithAnalyzer::MarkBufferMapShapes(const tir::PrimFunc& func) {
   }
 }
 
+Array<PrimExpr> IRMutatorWithAnalyzer::IterMapSimplifyWithContext(const Array<PrimExpr>& indices,
+                                                                  bool non_trivial_only) {
+  PrimExpr pred = const_true();
+  for (PrimExpr val : iter_predicates_) {
+    pred = pred && val;
+  }
+  int n = indices.size();
+  Array<PrimExpr> simplified = arith::IterMapSimplify(
+      indices, this->iter_vars_, pred, arith::IterMapLevel::Surjective, this->analyzer_);
+  if (non_trivial_only) {
+    for (int i = 0; i < n; ++i) {
+      if (simplified[i]->IsInstance<IntImmNode>() && indices[i]->IsInstance<VarNode>()) {
+        simplified.Set(i, indices[i]);
+      }
+    }
+  }
+  return simplified;
+}
+
 Stmt IRMutatorWithAnalyzer::VisitStmt_(const ForNode* op) {
   // record the loop variable as iterators
   Range dom = Range::FromMinExtent(op->min, op->extent);

src/arith/ir_mutator_with_analyzer.h

@@ -70,6 +70,12 @@ class IRMutatorWithAnalyzer : public tir::StmtExprMutator {
    */
   void MarkBufferMapShapes(const tir::PrimFunc& func);
 
+  /*!
+   * \brief Use internal bound information to perform inter map simplification of indices.
+   * \note Only do this during layout remapping
+   */
+  Array<PrimExpr> IterMapSimplifyWithContext(const Array<PrimExpr>& indices, bool non_trivial_only);
+
   /*! \brief internal analyzer field. */
   Analyzer* analyzer_;
   // the following two fields are useful in case we want

src/arith/iter_affine_map.cc

@@ -2061,6 +2061,12 @@ class IterMapToExprNormalizer : public ExprMutator {
     if (analyzer_->CanProve(expr->extent == expr->source->extent) && is_one(expr->lower_factor)) {
       return source * expr->scale;
     } else if (analyzer_->CanProve(expr->source->extent == expr->lower_factor * expr->extent)) {
+      // Simplify if `expr` is always 0. The 2nd condition guarantess that we do not aggressively
+      // simplify trivial iters like `vi \in [0, 1)`, which can be useful for subsequent analysis
+      // like tensorization.
+      if (is_one(expr->extent) && !is_one(expr->source->extent)) {
+        return make_const(expr->extent->dtype, 0);
+      }
       return floordiv(source, expr->lower_factor) * expr->scale;
     } else {
       return floordiv(floormod(source, expr->lower_factor * expr->extent), expr->lower_factor) *

src/relay/backend/te_compiler_cache.cc

@@ -19,6 +19,7 @@
 
 #include "./te_compiler_cache.h"
 
+#include <tvm/arith/analyzer.h>
 #include <tvm/driver/driver_api.h>
 #include <tvm/ir/name_supply.h>
 #include <tvm/ir/type_functor.h>
@@ -594,7 +595,8 @@ class ScheduleBuilder : public ExprVisitor {
                       src_size_1d *= c->shape[i];
                       orig_shape.push_back(PrimExpr(static_cast<int>((c->shape[i]))));
                     }
-                    auto dst_shape = index_map->MapShape(orig_shape);
+                    arith::Analyzer analyzer;
+                    auto dst_shape = index_map->MapShape(orig_shape, &analyzer);
                     std::vector<int64_t> dst_shape_int;
                     size_t dst_size_1d = 1;
                     for (size_t i = 0; i < dst_shape.size(); ++i) {