Skip to content

Inject LetStmt of variables that are used downstream modulo some constant. #9428

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
mcourteaux wants to merge 6 commits into from
Draft

Inject LetStmt of variables that are used downstream modulo some constant. #9428

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
79f14c6
Add a minimal repro for aligned-split phases not cancelling through a…
mcourteaux Sep 3, 2026
69063ba
Deinterleave ramps by strides up to seven, not four
mcourteaux Sep 3, 2026
33cd9cd
Cover more pattern periods in the aligned-split mux phase test
mcourteaux Sep 3, 2026
bf36224
Bind a variable's value modulo a constant where the variable is defined
mcourteaux Sep 4, 2026
b3e0958
Add InjectModuloVars to the Makefile's file lists
mcourteaux Sep 4, 2026
bf39e80
Fix C++20 extension not allowed.
mcourteaux Sep 5, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions Makefile
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -524,6 +524,7 @@ SOURCE_FILES = \
ImageParam.cpp \
InferArguments.cpp \
InjectHostDevBufferCopies.cpp \
InjectModuloVars.cpp \
Inline.cpp \
InlineReductions.cpp \
IntegerDivisionTable.cpp \
Expand Down Expand Up @@ -730,6 +731,7 @@ HEADER_FILES = \
ImageParam.h \
InferArguments.h \
InjectHostDevBufferCopies.h \
InjectModuloVars.h \
Inline.h \
InlineReductions.h \
IntegerDivisionTable.h \
Expand Down
2 changes: 2 additions & 0 deletions src/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -136,6 +136,7 @@ target_sources(
ImageParam.h
InferArguments.h
InjectHostDevBufferCopies.h
InjectModuloVars.h
Inline.h
InlineReductions.h
IntegerDivisionTable.h
Expand Down Expand Up @@ -319,6 +320,7 @@ target_sources(
ImageParam.cpp
InferArguments.cpp
InjectHostDevBufferCopies.cpp
InjectModuloVars.cpp
Inline.cpp
InlineReductions.cpp
IntegerDivisionTable.cpp
Expand Down
15 changes: 13 additions & 2 deletions src/Deinterleave.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -520,6 +520,17 @@ class Interleaver : public IRMutator {
return visit_let(op);
}

// The largest stride we'll deinterleave a ramp by. This is what decides
// whether a vectorized loop over a repeating pattern -- a sensor's colour
// filter array, say -- gets split into one slice per phase, so that each
// slice has a phase known at compile time. Bayer needs two and X-Trans
// needs six.
//
// It stops below eight because a permute with a period of eight is often
// better done as a single shuffle than as eight slices: on Hexagon, for
// instance, deinterleaving (x/8)*9 + x%8 displaces a single vdelta.
static constexpr int max_deinterleave_stride = 7;

Expr visit(const Ramp *op) override {
if (op->stride.type().is_vector() &&
is_const_one(op->stride) &&
Expand All @@ -536,7 +547,7 @@ class Interleaver : public IRMutator {

Expr visit(const Mod *op) override {
const Ramp *r = op->a.as<Ramp>();
for (int i = 2; i <= 4; ++i) {
for (int i = 2; i <= max_deinterleave_stride; ++i) {
if (r &&
is_const(op->b, i) &&
(r->type.lanes() % i) == 0) {
Expand All @@ -550,7 +561,7 @@ class Interleaver : public IRMutator {

Expr visit(const Div *op) override {
const Ramp *r = op->a.as<Ramp>();
for (int i = 2; i <= 4; ++i) {
for (int i = 2; i <= max_deinterleave_stride; ++i) {
if (r &&
is_const(op->b, i) &&
(r->type.lanes() % i) == 0 &&
Expand Down
Loading
Loading