Add EQUAL_ELTWISE support for eltwise test and kernel #30877

madhurthareja · 2025-06-05T17:43:01Z

Details:

Implement CPU plugin just-in-time emitter for Equal operation

Tickets:
30257

Testing/Temporary/CTestCostData.txt

src/plugins/intel_cpu/src/nodes/kernels/riscv64/jit_uni_eltwise_generic.cpp

src/plugins/intel_cpu/tests/functional/shared_tests_instances/single_layer_tests/eltwise.cpp

src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp

test.log

madhurthareja

Completed.

test.log

madhurthareja · 2025-06-06T07:06:58Z

Hi @a-sidorova . I made the changes as you suggested. Could you please check? I was able to get all the test cases passed.

a-sidorova

@madhurthareja please apply the comments, build the tests and launch them. Please check that all of them are passed and send the screenshot with them 😊

a-sidorova · 2025-06-06T07:38:01Z

src/plugins/intel_cpu/tests/functional/shared_tests_instances/single_layer_tests/eltwise.cpp

@@ -80,7 +80,7 @@ std::vector<EltwiseTypes> eltwise_op_types = {
        EltwiseTypes::FLOOR_MOD,
        EltwiseTypes::SQUARED_DIFF,
        EltwiseTypes::POWER,
-        EltwiseTypes::MOD
+        EltwiseTypes::MOD,


Minor: please remove extra comma

a-sidorova · 2025-06-06T07:39:00Z

src/plugins/intel_cpu/src/nodes/kernels/riscv64/jit_uni_eltwise_generic.cpp

@@ -593,7 +594,8 @@ std::set<std::vector<element::Type>> eltwise_precision_helper::get_supported_pre
              OV_CASE(Algorithm::EltwiseRelu, jit_relu_emitter),
              OV_CASE(Algorithm::EltwiseSigmoid, jit_sigmoid_emitter),
              OV_CASE(Algorithm::EltwiseSqrt, jit_sqrt_emitter),
-              OV_CASE(Algorithm::EltwiseSubtract, jit_subtract_emitter));
+              OV_CASE(Algorithm::EltwiseSubtract, jit_subtract_emitter),
+              OV_CASE(Algorithm::EltwiseEqual, jit_equal_emitter));


You moved EltwiseEqual to right place only in first registration list, what's about this one? :) Could you please fix it here too?

a-sidorova · 2025-06-06T07:40:58Z

src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp

+size_t jit_equal_emitter::aux_fp_gprs_count() const {
+    return 1;
+}


I see definition of the method aux_fp_gprs_count() here but I don't see the method declaration in the header. Please add it to the header.

a-sidorova · 2025-06-06T07:42:00Z

src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp

+jit_equal_emitter::jit_equal_emitter(ov::intel_cpu::riscv64::jit_generator* host,
+                                     ov::intel_cpu::riscv64::cpu_isa_t host_isa,
+                                     const std::shared_ptr<ov::Node>& node)
+    : jit_emitter(host, host_isa, get_arithmetic_binary_exec_precision(node)) {}


Since you use constants from the table (register_table_entries()), you need to call prepare_table() in constructors. Please see example in jit_exp_emitter.

a-sidorova · 2025-06-06T07:42:33Z

src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp

+    h->vmv_v_x(dst, zero); // set dst to 0
+    h->vfeq_vv(mask_vreg(), src0, src1); // compare, result in mask
+    h->vfadd_vf(dst, dst, one, VM::masked); // set 1.0 where mask is true   
+    }


Please remove extra }

madhurthareja · 2025-06-06T07:53:16Z

@madhurthareja please apply the comments, build the tests and launch them. Please check that all of them are passed and send the screenshot with them 😊

Yes, I'll do it. Thanks for your guidance. I'm a little naive in OC contributions, but loving them now with the help you and fellow contributors are providing.

…ce=GPU

madhurthareja · 2025-06-06T09:06:48Z

Hi! I've made the changes as discussed. Here are a few screenshots for the tests done. Please let me know if there's something I'm missing.

a-sidorova · 2025-06-06T09:09:38Z

he 11th test is taking a lot of time to run, and I get some segmentation faults in tensorflow related tests. Could you let me know what should I look forward to?

In general you don't need to launch all existing tests in OV - only ov_cpu_func_tests are needed (as described in the task #30257):

# If you use xuantie-gnu-toolchain:
<xuantie_install_path>/bin/qemu-riscv64 -cpu rv64,x-v=true,vext_spec=v1.0 ./bin/[platform]/[build_type]/ov_cpu_func_tests --gtest_filter="*smoke*Comparison*Equal*"

# If you use riscv-gnu-toolchain:
<riscv_install_path>/bin/qemu-riscv64 -cpu rv64,v=true,vext_spec=v1.0 ./bin/[platform]/[build_type]/ov_cpu_func_tests --gtest_filter="*smoke*Comparison*Equal*"

I believe that some tests might be failed due to missed packages/models/installed frontends. This is OK. Please pay attention only on ov_cpu_func_tests with --gtest_filter="*smoke*Comparison*Equal*" 😊

a-sidorova

@madhurthareja your changes look better now, thanks! I left a few comments + I launched CI jobs. Please take a look at the failed code-style jobs and apply suggested changes 😃

src/tests/test_utils/common_test_utils/include/common_test_utils/test_enums.hpp

src/tests/test_utils/common_test_utils/src/node_builders/eltwise.cpp

madhurthareja · 2025-06-06T11:54:05Z

@madhurthareja your changes look better now, thanks! I left a few comments + I launched CI jobs. Please take a look at the failed code-style jobs and apply suggested changes 😃

I made the required changes, but the issues with CI jobs seems to be regarding some permissions. Would I have to set it up or would that be done from the admin side?

a-sidorova · 2025-06-06T12:01:12Z

@madhurthareja your changes look better now, thanks! I left a few comments + I launched CI jobs. Please take a look at the failed code-style jobs and apply suggested changes 😃

I made the required changes, but the issues with CI jobs seems to be regarding some permissions. Would I have to set it up or would that be done from the admin side?

Anybody can open reports of GHA jobs. Please take a look at them and apply suggestions from this job (I see that you haven't made it yet):

src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp:216:-        prepare_table();
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp:217:-    }
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp:216:+    prepare_table();
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp:217:+}
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp:223:-        prepare_table();
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp:224:-    }
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp:223:+    prepare_table();
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp:224:+}
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp:241:-void jit_equal_emitter::emit_impl(const std::vector<size_t>& in_vec_idxs, const std::vector<size_t>& out_vec_idxs) const {
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp:241:+void jit_equal_emitter::emit_impl(const std::vector<size_t>& in_vec_idxs,
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp:242:+                                  const std::vector<size_t>& out_vec_idxs) const {
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp:251:-void jit_equal_emitter::emit_isa(const std::vector<size_t>& in_vec_idxs, const std::vector<size_t>& out_vec_idxs) const {
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp:252:+void jit_equal_emitter::emit_isa(const std::vector<size_t>& in_vec_idxs,
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp:253:+                                 const std::vector<size_t>& out_vec_idxs) const {
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp:258:-    h->vmv_v_x(dst, zero); // set dst to 0
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp:259:-    h->vfeq_vv(mask_vreg(), src0, src1); // compare, result in mask
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp:260:-    h->vfadd_vf(dst, dst, one, VM::masked); // set 1.0 where mask is true   
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp:260:+    h->vmv_v_x(dst, zero);                   // set dst to 0
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp:261:+    h->vfeq_vv(mask_vreg(), src0, src1);     // compare, result in mask
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.cpp:262:+    h->vfadd_vf(dst, dst, one, VM::masked);  // set 1.0 where mask is true
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.hpp:103:-                     ov::intel_cpu::riscv64::cpu_isa_t host_isa,
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.hpp:104:-                     const ov::element::Type exec_prc = ov::element::f32);
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.hpp:103:+                      ov::intel_cpu::riscv64::cpu_isa_t host_isa,
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.hpp:104:+                      const ov::element::Type exec_prc = ov::element::f32);
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.hpp:106:-                     ov::intel_cpu::riscv64::cpu_isa_t host_isa,
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.hpp:107:-                     const std::shared_ptr<ov::Node>& node);
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.hpp:106:+                      ov::intel_cpu::riscv64::cpu_isa_t host_isa,
src/plugins/intel_cpu/src/emitters/plugin/riscv64/jit_eltwise_emitters.hpp:107:+                      const std::shared_ptr<ov::Node>& node);
src/tests/test_utils/common_test_utils/src/node_builders/eltwise.cpp:15:-#include "openvino/op/erf.hpp"
src/tests/test_utils/common_test_utils/src/node_builders/eltwise.cpp:16:+#include "openvino/op/erf.hpp"

As for launching CI job, only Intel-workers can do it (like me). So when you apply code-style suggestions, please let me know - I will relaunch CI 😊

madhurthareja · 2025-06-06T12:25:39Z

@a-sidorova I checked the reports and found that:

The GitHub Actions token used by the workflow does not have permission to post review comments on the PR.

Diff in eltwise.cpp: This is a no-op change (removing and re-adding the same line). It is likely a formatting artifact and does not affect functionality.

I'm not sure why is it failing. Is there due to formatting issues? Maybe some spaces or extra tabs?

Please let me know if I'm unable to understand the issue. I'll retry to understand them.

a-sidorova · 2025-06-09T05:12:57Z

@a-sidorova I checked the reports and found that:

The GitHub Actions token used by the workflow does not have permission to post review comments on the PR.

Diff in eltwise.cpp: This is a no-op change (removing and re-adding the same line). It is likely a formatting artifact and does not affect functionality.

I'm not sure why is it failing. Is there due to formatting issues? Maybe some spaces or extra tabs?

Please let me know if I'm unable to understand the issue. I'll retry to understand them.

Hi!

The GitHub Actions token used by the workflow does not have permission to post review comments on the PR.

I think this is internal error and it come up only when there are problems with code-style. So no need to pay attention on it :)

Diff in eltwise.cpp: This is a no-op change (removing and re-adding the same line). It is likely a formatting artifact and does not affect functionality.

Code-style suggester suggests to swap lines with includes - please note that the difference between lines - number of lines. But this error is outdated - you reverted changes from the file src/tests/test_utils/common_test_utils/src/node_builders/eltwise.cpp. So you need to just fix all remaining errors from jit_eltwise_emitters.hpp and jit_eltwise_emitters.cpp.

Add EQUAL_ELTWISE support for eltwise test and kernel

87d4fea

madhurthareja requested review from a team as code owners June 5, 2025 17:43

madhurthareja requested review from mlukasze and removed request for a team June 5, 2025 17:43

madhurthareja mentioned this pull request Jun 5, 2025

implemented eltwise 'Equal' operation #30293

Closed

github-actions bot added category: IE Tests OpenVINO Test: plugins and common category: CPU OpenVINO CPU plugin no-match-files labels Jun 5, 2025

sys-openvino-ci added the ExternalPR External contributor label Jun 5, 2025

mlukasze linked an issue Jun 6, 2025 that may be closed by this pull request

[Good First Issue] [RISCV64]: Implement CPU plugin just-in-time emitter for Equal operation #30257

Open

a-sidorova reviewed Jun 6, 2025

View reviewed changes

madhurthareja added 3 commits June 6, 2025 11:54

Delete Testing/Temporary/LastTest.log

cdd457f

Delete pytest.log

48c42e9

Delete Testing/Temporary/CTestCostData.txt

f774481

t-jankowski reviewed Jun 6, 2025

View reviewed changes

test.log Outdated Show resolved Hide resolved

a-sidorova added the platform: risc-v OpenVINO on RISC-V label Jun 6, 2025

a-sidorova self-assigned this Jun 6, 2025

a-sidorova added this to the 2025.3 milestone Jun 6, 2025

Delete test.log

c541e17

github-actions bot removed the no-match-files label Jun 6, 2025

madhurthareja commented Jun 6, 2025

View reviewed changes

test.log Outdated Show resolved Hide resolved

Passed all the test and made the required changes

2c72fb5

Passed all the test and made the required changes

3d1d862

a-sidorova reviewed Jun 6, 2025

View reviewed changes

t-jankowski self-requested a review June 6, 2025 08:29

Successfully Tested for all the instances of Equal except target devi…

22bb33b

…ce=GPU

a-sidorova reviewed Jun 6, 2025

View reviewed changes

src/tests/test_utils/common_test_utils/include/common_test_utils/test_enums.hpp Outdated Show resolved Hide resolved

src/tests/test_utils/common_test_utils/src/node_builders/eltwise.cpp Outdated Show resolved Hide resolved

Minor fixes

8a11d28

github-actions bot removed the category: IE Tests OpenVINO Test: plugins and common label Jun 6, 2025

Add EQUAL_ELTWISE support for eltwise test and kernel #30877

Are you sure you want to change the base?

Add EQUAL_ELTWISE support for eltwise test and kernel #30877

Conversation

madhurthareja commented Jun 5, 2025

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

madhurthareja left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

madhurthareja commented Jun 6, 2025

Uh oh!

a-sidorova left a comment

Choose a reason for hiding this comment

Uh oh!

a-sidorova Jun 6, 2025

Choose a reason for hiding this comment

Uh oh!

a-sidorova Jun 6, 2025

Choose a reason for hiding this comment

Uh oh!

a-sidorova Jun 6, 2025

Choose a reason for hiding this comment

Uh oh!

a-sidorova Jun 6, 2025

Choose a reason for hiding this comment

Uh oh!

a-sidorova Jun 6, 2025

Choose a reason for hiding this comment

Uh oh!

madhurthareja commented Jun 6, 2025

Uh oh!

madhurthareja commented Jun 6, 2025

Uh oh!

a-sidorova commented Jun 6, 2025

Uh oh!

a-sidorova left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

madhurthareja commented Jun 6, 2025

Uh oh!

a-sidorova commented Jun 6, 2025

Uh oh!

madhurthareja commented Jun 6, 2025

Uh oh!

a-sidorova commented Jun 9, 2025

Uh oh!

Uh oh!

a-sidorova left a comment •

edited

Loading