Revision 3657cf5f363fd64aeaf06432e62e3960800927b0 authored by Andrew Adams on 26 January 2024, 17:26:12 UTC, committed by GitHub on 26 January 2024, 17:26:12 UTC
* Fix bounds_of_nested_lanes

bounds_of_nested_lanes assumed that one layer of nested vectorization
could be removed at a time. When faced with the expression:

min(ramp(x8(a), x8(b), 5), x40(27))

It panicked, because on the left hand side it reduced the bounds to
x8(a) ... x8(a) + x8(b) * 4, and on the right hand side it reduced the
bounds to 27. It then attempted to take a min of mismatched types.

In general we can't assume that binary operators on nested vectors have
the same nesting structure on both sides, so I just rewrote it to reduce
directly to a scalar.

Fixes #8038
1 parent 4590a09
Raw File
StorageFolding.h
#ifndef HALIDE_STORAGE_FOLDING_H
#define HALIDE_STORAGE_FOLDING_H

/** \file
 * Defines the lowering optimization pass that reduces large buffers
 * down to smaller circular buffers when possible
 */
#include <map>
#include <string>

#include "Expr.h"

namespace Halide {
namespace Internal {

class Function;

/** Fold storage of functions if possible. This means reducing one of
 * the dimensions module something for the purpose of storage, if we
 * can prove that this is safe to do. E.g consider:
 *
 \code
 f(x) = ...
 g(x) = f(x-1) + f(x)
 f.store_root().compute_at(g, x);
 \endcode
 *
 * We can store f as a circular buffer of size two, instead of
 * allocating space for all of it.
 */
Stmt storage_folding(const Stmt &s, const std::map<std::string, Function> &env);

}  // namespace Internal
}  // namespace Halide

#endif
back to top