Skip to content

Fix silent data corruption in JIT eltwise kernel for i8/u8 bitwise ops with broadcast #34639

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

Open
goyaladitya05 wants to merge 2 commits into
base: master
Choose a base branch
from
+48 −1
Open

Fix silent data corruption in JIT eltwise kernel for i8/u8 bitwise ops with broadcast #34639

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e2226b4
correct byte broadcast in JIT eltwise kernel for i8/u8 bitwise ops
goyaladitya05 Mar 11, 2026
592163f
Merge branch 'master' into fix/jit_eltwise_bitwise_i8_broadcast
maxnick Mar 12, 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
23 changes: 22 additions & 1 deletion src/plugins/intel_cpu/src/nodes/kernels/x64/jit_uni_eltwise_generic.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -549,7 +549,28 @@ void jit_uni_eltwise_generic<isa>::load_vector(Vmm vmm_src,
if (src_prc == dst_prc) {
if (broadcast) {
load_scalar(xmm_src, op, src_prc, dst_prc);
uni_vbroadcastss(vmm_src, xmm_src);
switch (src_prc.size()) {
case 1:
if (isa == x64::sse41) {
punpcklbw(xmm_src, xmm_src);
punpcklbw(xmm_src, xmm_src);
pshufd(xmm_src, xmm_src, 0);
} else {
vpbroadcastb(vmm_src, xmm_src);
}
break;
case 2:
if (isa == x64::sse41) {
punpcklwd(xmm_src, xmm_src);
pshufd(xmm_src, xmm_src, 0);
} else {
vpbroadcastw(vmm_src, xmm_src);
}
break;
Comment on lines +562 to +569
Copy link

Copilot AI Mar 12, 2026

Choose a reason for hiding this comment

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

[MEDIUM] load_vector() adds a 2-byte broadcast path (case 2), but this code calls load_scalar() first, and load_scalar() currently throws for src_prc == dst_prc with src_prc.size() == 2 (it only supports sizes 1 and 4 in that branch). As a result, the new 2-byte broadcast logic is effectively unreachable and any future attempt to broadcast u16/i16 without type conversion will still fail at runtime. Either add 2-byte support to load_scalar() for the src_prc == dst_prc case (load 16 bits and clear upper bits) or remove the case 2 handling here to avoid implying support that isn't actually implemented.

Suggested change
case 2:
if (isa == x64::sse41) {
punpcklwd(xmm_src, xmm_src);
pshufd(xmm_src, xmm_src, 0);
} else {
vpbroadcastw(vmm_src, xmm_src);
}
break;

Copilot uses AI. Check for mistakes.
default:
uni_vbroadcastss(vmm_src, xmm_src);
break;
}
} else {
uni_vmovups(vmm_src, op);
}
Expand Down
26 changes: 26 additions & 0 deletions ...plugins/intel_cpu/tests/functional/custom/single_layer_tests/instances/common/eltwise.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -276,6 +276,11 @@ static const std::vector<std::vector<InputShape>> bitwise_in_shapes_4D = {
},
};

static const std::vector<std::vector<InputShape>> bitwise_in_shapes_2D_broadcast = {
{{{1, -1}, {{1, 64}, {1, 1}}}, {{1, 64}, {{1, 64}}}},
{{{-1, -1}, {{32, 256}, {1, 1}}}, {{32, 256}, {{32, 256}}}},
};

static const std::vector<CPUSpecificParams>& bitwiseCpuParams() {
static const std::vector<CPUSpecificParams> params = {CPUSpecificParams({nhwc, nhwc}, {nhwc}, {}, {}),
CPUSpecificParams({nchw, nchw}, {nchw}, {}, {})};
Expand Down Expand Up @@ -316,6 +321,27 @@ const auto params_4D_bitwise = ::testing::Combine(

INSTANTIATE_TEST_SUITE_P(smoke_CompareWithRefs_4D_Bitwise, EltwiseLayerCPUTest, params_4D_bitwise, EltwiseLayerCPUTest::getTestCaseName);

const auto params_2D_bitwise_broadcast =
::testing::Combine(::testing::Combine(::testing::ValuesIn(bitwise_in_shapes_2D_broadcast),
::testing::ValuesIn({ov::test::utils::EltwiseTypes::BITWISE_AND,
ov::test::utils::EltwiseTypes::BITWISE_OR,
ov::test::utils::EltwiseTypes::BITWISE_XOR}),
::testing::ValuesIn(secondaryInputTypes()),
::testing::ValuesIn({ov::test::utils::OpType::VECTOR}),
::testing::ValuesIn({ov::element::Type_t::i8, ov::element::Type_t::u8}),
::testing::Values(ov::element::Type_t::dynamic),
::testing::Values(ov::element::Type_t::dynamic),
::testing::Values(ov::test::utils::DEVICE_CPU),
::testing::Values(ov::AnyMap())),
::testing::Values(CPUSpecificParams({}, {}, {}, {})),
::testing::Values(emptyFusingSpec),
::testing::Values(false));

INSTANTIATE_TEST_SUITE_P(smoke_CompareWithRefs_2D_Bitwise_i8u8_Broadcast,
EltwiseLayerCPUTest,
params_2D_bitwise_broadcast,
EltwiseLayerCPUTest::getTestCaseName);

const auto params_4D_bitwise_i32 = ::testing::Combine(
::testing::Combine(
::testing::ValuesIn(bitwise_in_shapes_4D),
Expand Down
Loading