Add a cheap hash to Expr nodes to speed up IREquality by abadams · Pull Request #9432 · halide/Halide

abadams · 2026-09-07T00:13:37Z

Summary

Adds a 32-bit hash to every IR node, packed into the previously-unused
upper 24 bits of IRNode::node_type's storage word (the node type
itself keeps the low 8 bits), so this doesn't increase the memory
footprint of an IR node.

Every Expr node's make() method fills in the hash from an
ultra-simple multiply-add combination of the hashes/values of its
arguments (combine_hash in Expr.h). Stmt nodes leave the extra
bits zero, since equal/less_than on Stmts are much less
frequently called than on Exprs.
Type::hash() (a 4-byte memcpy of the type's code/bits/lanes) lets
type-bearing nodes like Cast fold their target type into the hash,
so a Cast doesn't collide with its own child when they share a
node type.
(U)IntImm's hash multiplies the 64-bit value by a large odd
constant and keeps the high 32 bits (Knuth multiplicative hashing),
so small immediates (by far the most common case) still get a well
distributed hash instead of colliding with each other.
IREquality.h's equal/graph_equal/less_than/graph_less_than
now compare/order by this hash before falling back to a full
recursive comparison. A hash mismatch already implies the node
types differ, so the separate node-type check was dropped. For
less_than/graph_less_than the resulting order is arbitrary
anyway (only used for map keys), so it's fine to order directly by
hash when hashes differ.

Performance

Measured with HL_TIME_LOWERING_PASSES=1, target=host, comparing
this branch against main (a3690b3b6), 4 runs per app, means shown:

app	before (ms)	after (ms)	delta
bilateral_grid	34.79	32.60	-6.3%
camera_pipe	55.10	54.17	-1.7%
conv_layer	13.51	12.68	-6.1%
hist	14.73	13.77	-6.5%
iir_blur	12.77	11.73	-8.1%
interpolate	131.60	123.84	-5.9%
max_filter	17.11	15.31	-10.5%
nl_means	56.58	53.21	-6.0%
stencil_chain	315.65	300.65	-4.8%
unsharp	17.56	16.40	-6.6%
harris	18.12	17.12	-5.5%
bgu	578.76	472.17	-18.4%
blur	8.38	7.05	-15.9%
depthwise_separable_conv	75.48	69.01	-8.6%
wavelet	11.52	10.51	-8.8%
local_laplacian	150.82	141.02	-6.5%
lens_blur	223.75	206.83	-7.6%

Geomean speed-up across these 17 apps: 1.086x (8.6% faster lowering).

(fft and resize were excluded from this sweep: fft's generator
needs specific size params via its own driver rather than bare -g fft
defaults, and resize needs an explicit input.type GeneratorParam;
neither is related to this change.)

Testing

correctness_ir_equality, correctness_simplify, correctness_bounds,
correctness_cse, correctness_lots_of_dimensions, and
correctness_constant_expr all pass.

🤖 Generated with Claude Code

Each Expr node's make() method now fills in BaseExprNode::hash, an ultra-simple multiply-add combination of the node type and the hashes/values of its arguments. equal()/graph_equal() use it to short-circuit on a hash mismatch before doing a full recursive comparison, and less_than()/graph_less_than() use it directly to order nodes when hashes differ, since that ordering is arbitrary. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Move the hash from a separate BaseExprNode::hash field into a union with IRNode::node_type: the low 8 bits are the node type (as before) and the upper 24 bits are the hash, so this doesn't grow IRNode (the comment already noted these bits were free padding). Stmt nodes leave the upper bits zero. IRNode::set_hash keeps the low byte's node type intact while masking in a newly-computed 32-bit hash's upper 24 bits (its low bits are of poor quality due to the multiply-add construction, so they're discarded rather than shifted into the result). Add Type::hash() (a memcpy of its first 4 bytes) so type-bearing nodes like Cast can fold their type into the hash instead of just passing their child's hash through unchanged, which would otherwise collide with equal-typed child nodes of the same kind. IREquality.h now compares IRNode::hash directly instead of node_type followed by a separate BaseExprNode hash check, since a hash mismatch already implies a node_type mismatch. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

The previous big-endian branch kept args_hash's low 24 bits (the low-quality end of a multiply-add hash) instead of discarding them. Shift right by 8 first to keep the high-quality high bits, matching what the little-endian branch already does by masking. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

set_hash keeps only the high 24 bits of its argument. The previous IntImm/UIntImm hash sliced the 64-bit value into two 32-bit halves and combined them, which put all the entropy of small values (the common case) in the low bits that set_hash then throws away, making every small IntImm/UIntImm of a given sign collide. Multiply the value by a large odd 64-bit constant and keep the high 32 bits of the product instead (Knuth multiplicative hashing), which mixes the low bits of the value into the high bits of the result even when the value itself is small. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

abadams · 2026-09-07T00:14:44Z

@mcourteaux please check if this hash serves your needs in #9400, instead of having to compute one on the fly

alexreinking · 2026-09-07T00:34:30Z

+ * 32 free bits in between the ref count and the Type field anyway. We use
+ * the first 8 to store the node type, and the next 24 as a hash of the
+ * children of the node, to make syntactic comparisons faster. */
+ union {


Why not use a bitfield here?

because I want "hash" to mean the entire 32 bits.

(otherwise I'd have to manually incorporate the node_type bits into the hash in every make method)

abadams · 2026-09-07T00:44:09Z

If I use this to swap our expr-keyed maps and sets to be unordered_maps and unordered_sets, lowering time gets worse, and if I use ankerl::unordered_dense set and maps, lowering time doesn't change. So for this PR I'll leave this hash as something operator< uses for a regular map or set.

abadams · 2026-09-07T01:10:42Z

A custom map/set optimized for the sizes we see in practice was only 1.5% faster than this branch, so not worth the code complexity.

codecov · 2026-09-07T01:53:55Z

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 70.11%. Comparing base (a3690b3) to head (2b43673).

Additional details and impacted files

@@ Coverage Diff @@## main #9432 +/- ##
==========================================
- Coverage 70.12% 70.11% -0.01% 
==========================================
Files 261 261 Lines 79405 79462 +57 Branches 19362 19365 +3 ==========================================
+ Hits 55684 55717 +33 + Misses 17896 17891 -5 - Partials 5825 5854 +29

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

abadamsand others added 5 commits September 6, 2026 16:32

Tighten IRNode hash comment
2b43673
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

alexreinking reviewed Sep 7, 2026
View reviewed changes

Navigation Menu

Uh oh!

Add a cheap hash to Expr nodes to speed up IREquality - #9432

Add a cheap hash to Expr nodes to speed up IREquality#9432
abadams wants to merge 5 commits into
mainfrom
abadams/expr_hash

abadams commented Sep 7, 2026

Uh oh!

abadams commented Sep 7, 2026•
edited
Loading

Uh oh!

alexreinkingSep 7, 2026

Uh oh!

abadamsSep 7, 2026

Uh oh!

abadamsSep 7, 2026

Uh oh!

abadams commented Sep 7, 2026•
edited
Loading

Uh oh!

abadams commented Sep 7, 2026

Uh oh!

codecovBot commented Sep 7, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

abadams commented Sep 7, 2026

Summary

Performance

Testing

Uh oh!

abadams commented Sep 7, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

alexreinkingSep 7, 2026

Choose a reason for hiding this comment

Uh oh!

abadamsSep 7, 2026

Choose a reason for hiding this comment

Uh oh!

abadamsSep 7, 2026

Choose a reason for hiding this comment

Uh oh!

abadams commented Sep 7, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

abadams commented Sep 7, 2026

Uh oh!

codecovBot commented Sep 7, 2026

Codecov Report

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

abadams commented Sep 7, 2026•
edited
Loading

abadams commented Sep 7, 2026•
edited
Loading